We’re in the midst of an unprecedented investment boom. Trillions of dollars are flowing into artificial intelligence, funding everything from foundation models to enterprise automation. Valuations soar. Capabilities multiply. Deployment accelerates.

But while we count the capital going in, we’re not accounting for what we’re taking on. For every dollar invested in AI, we’re accumulating liabilities that don’t appear on any balance sheet—technical debt we can’t audit, ethical questions we’ve deferred, legal exposure we haven’t quantified, and social contracts we’re quietly rewriting. The financial investment is visible and celebrated. The debt we’re accruing is invisible and, for now, ignored.

This isn’t a hypothetical future problem. It’s happening now, compounding with every deployment, and the bill is coming due faster than we think.

The Debt Portfolio

Technical Debt: Building on Quicksand

We’re deploying systems we can’t fully explain. That’s not a provocative claim—it’s a technical fact. Neural networks operate as black boxes where understanding input-output relationships doesn’t mean understanding the decision-making process itself. We can test for outcomes, but we can’t audit the reasoning.

This matters because these systems aren’t isolated experiments. They’re being integrated into legacy infrastructure never designed to accommodate them, creating brittle, untestable architectures where failure modes multiply faster than we can map them. A recommendation engine connects to inventory management, which triggers supply chain automation, which adjusts pricing algorithms, which influences customer behavior predictions—and somewhere in that chain, something breaks in a way no single team understands.

The gap isn’t just between what AI can do and what we understand about how it works. It’s between the speed of capability advancement and the speed of our comprehension. Every deployment on this asymmetric foundation is technical debt—functionality that works until it doesn’t, in ways we can’t fully predict or prevent.

Risk Debt: The Illusion of Precision

AI systems generate outputs with impressive precision: percentages to decimal points, confidence scores, probability distributions. This precision creates a dangerous illusion—that we understand the underlying uncertainty we’re operating with.

We don’t. We’re making consequential decisions based on models trained on historical data that may or may not represent future conditions, using architectures that may or may not generalize beyond their training distribution, deployed in contexts where the stakes may be vastly higher than anything the system was tested for.

Consider the cascading failure points. An AI recruiting tool inherits biases from historical hiring patterns. Those biased recommendations influence who gets interviewed. Those hiring decisions create new training data. The bias compounds, and by the time anyone notices, you’ve hired three years’ worth of cohorts using a systematically flawed process. That’s not a technical glitch—it’s structural risk we baked into operations before we understood what we were building.

Liability Debt: When Personalization Becomes Peril

Hyper-personalization is pitched as AI’s killer feature—systems that know customers so well they can anticipate needs, customize experiences, and optimize engagement. But personalization creates specificity, and specificity creates liability.

Send a generic marketing email to a million people and one person has a bad reaction? That’s unfortunate. Send a million individually customized messages and one of them says exactly the wrong thing to exactly the wrong person at exactly the wrong moment? That’s a lawsuit with your company’s name on it—and you may not even know which message caused it, because the system generated it dynamically.

This raises the fundamental question we’re avoiding: who’s responsible when AI makes a consequential error? The company that deployed it? The vendor that built it? The engineer who trained the model? The manager who approved the deployment? The executive who set the strategy?

We’re rapidly expanding what’s technically possible while the legal framework for what’s defensible remains stuck in an earlier era. Product liability law was written for physical goods with knowable failure modes. We’re deploying autonomous systems whose failure modes we’re still discovering—often after deployment, at scale, with real-world consequences.

Ethical Debt: Decisions Deferred, Not Made

Move fast and break things was always questionable advice. Applied to AI systems that affect people’s lives, it’s not just reckless—it’s compounding ethical debt with every deployment.

Consider what we’re actually doing when we deploy AI systems. We’re encoding values, making tradeoffs, and prioritizing some outcomes over others—but we’re doing it implicitly, embedded in model architectures and training objectives and optimization functions, rather than explicitly as ethical decisions that get debated and decided.

A content recommendation algorithm that optimizes for engagement isn’t neutral. It’s making a values judgment that engagement matters more than accuracy, that keeping users on platform matters more than informing them, that viral spread matters more than truthfulness. Those are profound ethical choices, but they’re embedded in code rather than articulated as policy.

The cost of “fix it later” thinking isn’t evenly distributed. Some communities are already bearing the brunt of biased facial recognition, discriminatory credit algorithms, and automated decision systems that lack accountability. By the time we get around to fixing these issues—if we do—generations of people will have been affected by systems we deployed before we bothered to understand their impact.

Share Facing Disruption - Accelerating innovation and growth

Governance Debt: Policy Moving at Dial-Up Speed

Board meetings happen quarterly. Model capabilities advance weekly. This velocity mismatch creates a dangerous gap between what leadership approves and what actually gets deployed.

Boards sign off on “implementing AI in customer service” or “automating underwriting processes” or “deploying personalization at scale.” What they’re often not signing off on—because they’re not being asked to, or don’t know to ask—are the specific tradeoffs, failure modes, risk tolerances, and accountability structures those deployments require.

Meanwhile, regulatory frameworks built for a different technological era are trying to govern systems that didn’t exist when the laws were written. We’re underwriting risks we don’t fully understand using standards that assume we do. We’re creating dependencies on systems we don’t control, operated by vendors who may not even understand the liability they’re transferring to us.

The Accountability Gap

The Third-Party Illusion

Outsourcing AI development doesn’t eliminate risk—it just obscures it. When something goes wrong with a vendor’s model deployed at your company, under your brand, affecting your customers, “we bought it from someone else” isn’t a defense. It’s an admission that you deployed systems you didn’t understand, affecting people you were responsible for.

The vendor relationship creates a particularly insidious form of liability. You’re trusting “best practices” that haven’t been tested at scale, relying on security audits that may not have examined what you actually need examined, and depending on contractual language that might not hold up when your use case inevitably differs from what was anticipated.

The Frontline Trap

When AI systems fail, we tend to blame the people closest to the failure. The customer service rep who didn’t catch the AI’s error. The loan officer who trusted the automated underwriting. The content moderator who approved what the system flagged as safe.

This is the accountability equivalent of punishing the factory worker for the bridge collapse. We give frontline practitioners tools without adequate guardrails, training, or oversight, then hold them responsible when systems fail in ways they had no power to prevent. It’s not just unfair—it’s a fundamental misunderstanding of where responsibility lies.

You cannot have responsible use without responsible guidance. If your AI governance strategy is “we trust our people to use AI responsibly,” you’ve abdicated the actual leadership obligation: creating structures that make responsible use possible.

Leadership’s Reckoning

Direction-setting is the fundamental responsibility of leadership, and in AI deployment, that means understanding—not just at a buzzword level, but genuinely—what systems you’re putting into operation, what failure modes they have, what risks they create, and who bears those risks.

“We didn’t know” won’t be a viable defense when the liability comes due. Fiduciary duty includes the obligation to understand the systems you’re deploying and the risks you’re taking on behalf of others. If your board can’t explain how your AI systems work, what assumptions they make, where they’re vulnerable to failure, and who’s accountable when things go wrong, you’re not governing responsibly—you’re hoping nothing explodes before your term ends.

The decisions that create downstream chaos are made at the top: the strategy that prioritizes speed over safety, the budget that funds deployment but not governance, the incentive structure that rewards scale over scrutiny, the organizational design that separates those building systems from those who bear the consequences.

What We’re Really Asking

Strip away the technical complexity and we’re confronting fundamental questions we’ve been avoiding:

How much uncertainty can we tolerate in pursuit of efficiency? We’ve always made decisions under uncertainty, but AI systems operate with uncertainties we can’t even fully characterize. When does acceptable risk-taking become reckless gambling with other people’s stakes?

When does “good enough for now” become negligent? There’s always pressure to ship, to deploy, to capture market share. But deploying a physical product with known defects is different from deploying an AI system whose defects you haven’t discovered yet and might not be able to fix even if you do.

What do we owe to those affected by systems we don’t fully understand? The people on the receiving end of AI decisions—loan applicants, job candidates, content viewers, medical patients—didn’t consent to experimental deployment. They didn’t sign up to be test cases while we figure out what our systems actually do.

Can we move fast without breaking fundamental social contracts? The contract is simple: the organizations wielding power over people’s lives should understand what they’re doing and be accountable for the consequences. We’re on the verge of breaking that contract at scale.

The Governance Imperative

Voluntary frameworks aren’t enough. “Ethics guidelines” and “responsible AI principles” and “fairness commitments” sound good in press releases, but they’re not governance structures. They’re aspiration without mechanism, values without accountability.

Robust AI governance means having internal expertise—not just external consultants telling you what you want to hear. It means technical staff who can actually audit what systems are doing, legal staff who understand both the technology and the exposure, risk managers who can model scenarios beyond the ones in your vendor’s marketing materials.

It means accountability structures that exist before you need them: clear ownership of decisions, documentation of tradeoffs, escalation paths for concerns, stopping mechanisms when uncertainty exceeds tolerance, and consequences when protocols are violated.

It means knowing what questions to ask before deployment, not just how to respond after failure. Who approved this? Based on what understanding? What testing happened? What risks were identified? What failure modes were anticipated? Who’s monitoring performance? Who has authority to shut it down? What’s the plan if it goes wrong?

The Stakes

The cost of AI’s invisible debt won’t be evenly distributed. It never is.

It will hit consumers who didn’t consent to being subjects of experimental deployment, who find themselves on the wrong side of algorithmic decisions they can’t contest or even understand.

It will hit workers who become scapegoats for systemic failures, blamed for trusting tools they were given and told to use, held accountable for risks leadership should have managed.

It will hit communities that bear the brunt of biased systems—the neighborhoods where facial recognition fails more often, the demographics where credit algorithms discriminate, the populations where medical AI performs worst.

And it will hit future stakeholders who inherit the shortcuts we’re taking now: the organizations trying to untangle brittle systems built for speed not sustainability, the regulators trying to govern technologies they’re just beginning to understand, the society trying to maintain trust in institutions that deployed systems they couldn’t explain or control.

What Happens Next

This isn’t a call to stop building AI. It’s a call to stop pretending that velocity is the same as progress, that innovation justifies recklessness, that complexity excuses incomprehensibility.

For leadership: Your board needs specific governance structures, not vague principles. You need to be asking—and able to understand the answers to—questions like: What are our AI systems optimizing for and who decided that? Where are the failure modes and what happens when they activate? Who has authority to stop deployment if risks exceed tolerance? What liability are we taking on and do we understand it?

The difference between risk management theater and actual accountability is whether you’re asking these questions before deployment or after something goes wrong.

For practitioners: You need to know when to escalate and when to refuse. Document decisions that leadership should be making but isn’t. Build internal coalitions for responsible deployment. You’re not just implementers—you’re often the last line of defense between a risky deployment and real-world harm.

For the industry: The race to deploy is a race to accumulate liability. The companies that will win long-term aren’t the ones that moved fastest—they’re the ones that moved responsibly, that built understanding alongside capability, that created accountability structures before they needed them.

What mature AI governance looks like in practice is: slower deployment schedules, more testing before launch, clear ownership of risk, meaningful oversight of vendor relationships, and the ability to explain your systems not just to your engineers but to a jury, your board, and the people whose lives they affect.

Leave a comment

The Questions That Matter

Before your next AI deployment, ask yourself:

What debts is your organization accumulating right now? Not financial debts—the technical, ethical, legal, and governance debts that don’t show up on balance sheets but will come due just as surely.

Who will ultimately pay when they come due? Spoiler: probably not the people who accumulated them.

What governance structures exist between “exciting new capability” and “deployed at scale”? If the answer is “not much” or “we move pretty fast,” you’re not governing—you’re gambling.

Can you explain your AI systems to a jury? To your board? To the people they affect? If not, you might want to figure that out before you have to.

The invisible ledger is growing. The question is whether we’ll start accounting for it honestly—or whether we’ll pretend these debts don’t exist until they all come due at once.

Walking out of the CTA Tech Trends presentation on day one of CES 2026, I had a surreal moment. Every trend they highlighted - personalization, platform ecosystems, connected spaces, digital health, precision healthcare - these were the exact things we were prototyping at Accenture Liquid Studios a decade ago.

My first thought: “Wow, we were really ahead of our time.”

My second thought: “Wait... why are these STILL the trends?”

Then it hit me: The work we did back then was exploring 5-10 year horizons. Technology has finally caught up. These aren’t future trends anymore - they’re active deployments happening right now.

And that realization reframed everything I saw over the next four days.

The Compute Abundance Paradox

Walking the AMD and NVIDIA exhibits, I watched demos showcasing rapidly declining costs for token generation - both monetary and computational. Processing efficiency in data centers is advancing faster than our ability to consume it.

Here’s the paradox: We’re building infrastructure for today’s computational requirements, but those requirements are dropping faster than we’re building capacity.

If inference optimization continues (or we move to entirely new fluid architectures), we might find ourselves in 2-3 years with massively excess capacity. We’re scaling for scarcity that may not exist.

But here’s where it gets interesting:

The AMD keynote showcased video generation, animation, and world-building capabilities that would have required render farms just a few years ago. The creative workflow is transforming:

1. Vision in your head

2. Rapid prototyping in minutes

3. Iterative refinement

4. Bridge to reality

We’ve democratized creativity. And if compute abundance becomes reality, we’ll unlock exponentially more creative output from people who were previously constrained by technical barriers.

What really excites me? Both NVIDIA and AMD are enabling developers to develop locally - not just with LLMs, but with multimodal action models, voice and language models, models that process sensor fusion data and react in real-time. Full-stack AI prototyping capability is sitting on your desk now.

Human-Readable Code is Dead (And That’s Okay)

A conversation with a former Liquid Studios colleague fundamentally shifted how I think about AI-generated code.

His take: “Human-readable code is dead.”

My initial reaction was defensive. But then he walked me through the history of abstraction:

Assembly → C → JavaScript → TypeScript → AI-generated code

We’ve always been abstracting away from “readable” lower-level code. Nobody’s writing raw assembly saying “this is the ONLY way.” We accept that compilers have bugs, memory leaks, and inefficiencies. We constantly update them with better patterns.

JavaScript is already a huge abstraction. TypeScript adds another layer. AI is just the next abstraction layer.

The shift in how we evaluate code:

OLD CRITERIA: Is this human-optimized, readable, well-commented?

NEW CRITERIA: Does this achieve the desired result with predictable outcomes? Are edge cases articulated thoroughly? Does it meet requirements and vision?

Here’s what makes me addicted to “vibe coding”: the ability to visualize what I want to happen and turn vision into reality in minutes.

The pushback I hear: “But AI code isn’t efficient!”

Neither was early compiler output. We improved it. The same will happen here.

The real question isn’t whether AI can write perfect code today. It’s whether you’re ready to shift your evaluation criteria from “is this how a human would write it?” to “does this work reliably?”

The Dark Factory Principle: Why Humanoid Robots Miss the Point

At CES 2026, I saw dozens of humanoid robots. Arms. Hands. Walking. Picking up boxes.

And I think we’re sandbagging progress.

A conversation about “dark factories” reframed everything. A dark factory is so automated you don’t need lights - because no humans enter.

Here’s the problem with humanoid robots: We’re building robots with arms to pick up boxes and move them to other places.

We’re automating human processes instead of eliminating them entirely.

Why move boxes at all? Why have discrete pickup/dropoff points? Why design systems that require human-shaped movement patterns?

This is the same mistake we made early in digital transformation: replicating paper processes in software instead of reimagining the workflow.

The real breakthrough isn’t making robots work like humans. It’s designing systems where human-shaped work isn’t necessary.

The pattern: We’re at an inflection point. Stop asking “how do we automate what humans do?” Start asking “what would this look like if humans were never part of the equation?”

The companies that figure this out won’t just be more efficient. They’ll be operating in a completely different paradigm.

AI Companions: From Isolation to Community (Why I Was Wrong)

I’ve been publicly concerned about AI home companions for months. At CES, a conversation with the Aviden team completely changed my perspective.

My fear: AI companions would anchor isolated seniors at home, making the loneliness epidemic worse.

Here’s what I missed: AI companions as stepping stones, not destinations.

Think about the bookends of aging well:

• Isolated: At home, sedentary, disconnected

• Thriving: Active, community-engaged, social

I assumed AI companions locked people into the isolated end.

The Aviden model showed me a graduated re-integration approach:

Phase 1: Break the isolation habit - promote movement in small steps, build micro-habits of engagement, lower the activation energy to re-enter community.

Phase 2: Virtual bridges - connect with other users virtually first, build comfort with social interaction, create shared experience and identity.

Phase 3: In-person community - facilitate real-world meetups. The AI becomes a vehicle to get started. The community becomes “so much more.”

This reminded me of my parents. They lived in the suburbs for years. Isolated. Never talked about neighbors. Then they got a dog. Now they know everyone in the neighborhood.

The dog didn’t replace community - it created a reason to engage with it.

Can AI intentionally reinforce beneficial behaviors and get people back into their communities? If designed right, yes.

This aligns with everything I believe about AI: augment humans, strengthen human connection, don’t replace it. The best AI interventions create scaffolding for behavior change, then gradually remove themselves as the human capability strengthens.

The Biometric Data Paradox: Why More Data Doesn’t Mean Better Healthcare

At CES, I spoke with HaloScape about the future of longitudinal health data. We’ve seen this vision before. The challenges haven’t changed.

The pitch: Collect biometric data continuously. Share it with healthcare providers. Enable better outcomes.

The problem: More data ≠ better outcomes.

Healthcare providers have very limited time. Overabundance of information creates cognitive overload. The pushback is valid:

“I don’t trust this data.”
“I can’t see how to use this data.”
“I don’t have time to interpret this.”

Here’s what we’re missing: qualitative context.

We obsess over quantitative metrics while ignoring qualitative measures that provide critical context:

• Habits and behavior patterns

• How patients are feeling (subjective experience)

• Journal entries

• “Little pains” they forget to mention

• Social determinants of health

Biometric data without qualitative context is like having vitals without knowing the patient just ran up three flights of stairs, or seeing elevated heart rate without knowing about housing instability.

This came up in my webcast with Dr. Garrett Sessel months ago. The critical information gets lost when we focus purely on quantitative metrics.

The real opportunity for AI in healthcare isn’t autonomous diagnosis or data collection at scale. It’s AI that synthesizes biometric data with qualitative context and presents providers with actionable insights - not data dumps.

The functional capabilities are there. The last mile is figuring out how to make it operationally viable in real clinical workflows.

Sound familiar? That’s the pattern across every CES trend I saw.

Physical AI & The Last Mile Problem

CES 2026 was full of “AI-powered” everything. But the thing that genuinely excites me? Physical AI.

AI that interacts with real-time, real-world data. Understands its environment. Reasons through the next best action.

This is the next evolution: edge + air-gapped AI that operates autonomously with logical reasoning. No cloud dependency. Real-world consequences. And yes, legitimate concerns we need to explore.

But here’s what I keep seeing across immersive reality, robotics, and physical AI deployments:

✅ Functional capabilities: There
❌ Operational viability: Not quite

The last mile challenge:

• Battery life constraints

• Device management in the field

• Charging infrastructure

• Tracking distributed hardware at scale

The tech works in demos. Managing these devices in production at scale? That’s still the blocker.

This is the pattern I saw everywhere at CES:

• Code generation: Capability ✓ | Trust in production ✗

• Healthcare AI: Capability ✓ | Provider workflow integration ✗

• Biometric monitoring: Data collection ✓ | Meaningful synthesis ✗

• Immersive tech: Functional ✓ | Field management ✗

There’s a gap between “technically possible” and “operationally viable at scale.”

The Meta-Pattern: Removing Constraints That Are Now Obsolete

Looking back across every conversation, every keynote, every demo, a single theme emerges:

Every meaningful advancement at CES 2026 was about removing intermediaries and constraints that were historically necessary but are now obsolete.

• Between thought and code

• Between human process and automation

• Between creative vision and output

• Between developer and deployed AI model

• Between isolation and community

• Between data collection and clinical insight

The technologies we prototyped 10 years ago assumed these constraints were permanent. They’re not.

What This Means for Leaders

If you’re making 5-10 year bets today, remember: Innovation timelines are longer than we think, but the acceleration at the end is faster than we expect.

Success requires more than technology capability - it requires ecosystem readiness, trust-building, workflow integration, and operational viability at scale.

The vendors will show you what’s possible. Someone needs to tell you what’s actually ready for your environment, your workflows, your constraints.

Three questions to ask about any “AI innovation” you’re evaluating:

1. Is this automating a human process, or eliminating the need for that process entirely? The latter is where transformational value lives.

2. What’s the gap between technical capability and operational viability? Battery life, device management, provider trust, workflow integration - these “last mile” problems kill more innovations than technical limitations.

3. Does this create more human connection or less? The best AI interventions strengthen human capability and community, then gradually remove themselves as scaffolding.

The Future We’re Building Now

The future we imagined in 2016 arrived in 2026. What are you building now that won’t fully materialize until 2036?

Because I guarantee you: It’s going to take longer than you think. And when it finally arrives, it’s going to happen faster than you expect.

The companies that survive that transition will be the ones who understood the difference between what’s technically possible and what’s operationally viable - and had the patience to bridge that gap thoughtfully.

AJ Bubb is the founder of MXP Studio, a fractional AI strategy consultancy, and host of the Facing Disruption podcast. He helps mid-market companies navigate AI adoption strategically, drawing on 15+ years of innovation leadership from AWS, Accenture, and building emerging tech solutions that actually make it to production.

We’re living through one of the most consequential technology shifts in business history, yet the narrative around AI adoption has become dangerously detached from reality. Companies are pouring billions into generative AI initiatives with the expectation of immediate transformation, only to discover that the technology alone doesn’t deliver the promised results. The disconnect isn’t about the capabilities of AI models; it’s about how organizations fundamentally misunderstand what they’re actually implementing.

During our first Coffee Bytes conversation, at a Yemeni coffee shop in Sterling, Virginia, I sat down with Mo Hafaz from Beyond the Byte to explore why AI implementations keep falling short of expectations. We’d both been watching the same pattern repeat across industries: enthusiastic adoption followed by disappointing outcomes, mounting frustration, and eventually, disillusionment. Recent MIT research has confirmed what we’ve been observing in the field: about 95% of enterprise AI pilot programs fail to deliver measurable business impact, despite companies investing an average of $1.9 million on generative AI initiatives. The problem isn’t the technology. It’s us.

The Intern Problem: Expecting Expertise from Day One

The most useful mental model I’ve found for understanding AI implementation failures is thinking about generative AI as your newest employee, specifically, your greenest intern. This isn’t a metaphor I use lightly. When organizations deploy ChatGPT, Claude, or any other large language model, they’re essentially hiring someone with impressive general knowledge but zero understanding of their specific business context, processes, or constraints.

Think about how absurd our expectations have become. You wouldn’t hire a fresh college graduate on Monday and expect them to run your company by Friday. You wouldn’t hand them the keys to your most critical processes without any onboarding, training, or supervision. Yet that’s precisely what happens when organizations implement AI tools. The technology arrives with immense capabilities and immediate availability, which creates a dangerous illusion that it’s ready to perform at an expert level right out of the box.

The reality is messier. AI systems require what we call fine-tuning, the process of taking a generalist model and honing it for specific, well-defined tasks with appropriate guardrails. The MIT research identified this integration gap as the core failure point: generic tools like ChatGPT excel for individuals because of their flexibility, but they stall in enterprise use since they don’t learn from or adapt to organizational workflows. Without this customization, you’re asking a tool to navigate your business with no map, no training, and no understanding of where the landmines are buried.

As a mostly ridiculous detour to the conversaion, Mo and I wrestled with a particularly vexing technology setup, trying to mount an Insta360 camera to a new conferencing device. It should have been intuitive. The device was clearly designed with user experience in mind. But we couldn’t figure out the mounting mechanism without the manual. This small frustration perfectly encapsulated the AI implementation challenge: even well-designed technology requires understanding and proper setup to deliver value.

The Cognitive Debt We’re Accumulating

Beyond implementation failures, there’s a more insidious problem emerging from our relationship with AI tools: what researchers are now calling cognitive debt. An MIT Media Lab study tracking students over four months found that those who relied on ChatGPT for essay writing showed weaker neural connectivity and poorer memory recall compared to students who wrote without AI assistance or used traditional search engines. The implications extend far beyond academic settings.

Cognitive debt works like technical debt in software development. Shortcuts taken today create compounding problems tomorrow. When we offload thinking to AI systems, we’re not just saving time in the moment. We’re potentially weakening our capacity for the deep cognitive work that creates lasting knowledge and genuine expertise. The MIT researchers found that students who used AI assistance couldn’t accurately recall what they had written, struggled to quote their own work, and exhibited reduced brain activity patterns indicating cognitive under-engagement.

This matters profoundly in business contexts. The executives and knowledge workers who become overdependent on AI for thinking, analysis, and problem-solving may find themselves less capable of those tasks over time. The muscle memory of critical thinking, like physical muscles, atrophies without regular use. We’re at risk of creating a generation of professionals who can prompt AI effectively but struggle to think independently when the tools aren’t available.

An important point was raised during our discussion: AI is poised to benefit people with the most experience the most. The technology augments existing expertise rather than replacing it. Someone with deep domain knowledge can effectively guide, critique, and refine AI outputs. But what happens to the professionals just starting their careers? If junior roles increasingly get automated, where do people develop the foundational expertise that makes AI augmentation valuable in the first place?

The Trough of Disillusionment Has Arrived

Gartner’s 2025 Hype Cycle for Artificial Intelligence confirms that generative AI has officially entered the trough of disillusionment, with less than 30% of AI leaders reporting their CEOs are satisfied with AI investment returns. This is the inevitable correction that follows any overhyped technology. We saw it with blockchain, with IoT, with big data platforms. The pattern is always the same: initial excitement, inflated expectations, disappointing reality, and finally, for technologies that survive, a more sober understanding of actual value.

The trough of disillusionment isn’t failure. It’s the place where real work happens. The survivors of this phase will be the organizations that move beyond AI theater, implementing AI purely for marketing value or perception, and focus on genuine integration that solves specific business problems. The MIT research revealed a critical misalignment in resource allocation, with more than half of generative AI budgets devoted to sales and marketing tools, yet the biggest ROI came from back-office automation, like eliminating business process outsourcing and streamlining operations.

This disconnect between spending and value creation reflects a broader strategic failure. Companies are chasing the sexy, customer-facing applications of AI chatbots, content generation, and personalized recommendations while overlooking the unglamorous but highly valuable operational improvements. The organizations succeeding with AI aren’t necessarily the ones with the most advanced models or the biggest budgets. They’re the ones with clear strategies for where AI actually fits into their operations.

The Human-Machine Collaboration Sweet Spot

The conversation with Mo kept circling back to a central question: where’s the line between helpful augmentation and dangerous dependency? I use ChatGPT regularly as a thought partner. It’s remarkably good at taking half-formed ideas and helping me structure them into something coherent. When I’m researching synthetic data models, for instance, AI can explain complex concepts like variational autoencoders or generative adversarial networks in terms I can understand, then show me the code implementation.

But I’ve also learned to recognize when AI starts leading me in directions I don’t want to go. After a few prompts, ChatGPT often takes conversations into territory that feels off-track from my original intent. The tool is too eager sometimes. It wants to help so much that it makes assumptions about where you’re heading and sprints ahead without checking if that’s actually where you want to be. This is where the intern metaphor breaks down slightly: a good intern asks clarifying questions. AI systems often just run with their best guess.

The sweet spot for AI use isn’t about maximizing automation. It’s about finding the right balance between AI assistance and human judgment at every stage of work. Use AI to find corners that need investigating, then do the investigation yourself. Let AI help structure your thinking, but make sure the thinking itself remains yours. Have AI generate first drafts, but only if you’re capable of critically evaluating and substantially revising the output.

The MIT research found that companies purchasing AI solutions from specialized vendors achieved a 67% success rate, while internal builds succeeded only about 33% of the time. This gap exists because specialized vendors have already done the hard work of determining optimal human-machine collaboration patterns for specific use cases. They’ve learned through trial and error where AI adds value and where human expertise remains essential.

Why Communication Skills Matter More Than Ever

During our conversation, Mo made a point that deserves more attention than it typically gets: people are already terrible at communicating with each other. We have body language, facial expressions, years of social conditioning, and shared cultural context, and we still misunderstand each other constantly. Now we’re trying to communicate with AI systems that have none of those contextual cues, and we’re surprised when things go wrong.

The myopic nature of AI interaction amplifies every weakness in how we express ideas, ask questions, and provide direction. If you can’t clearly articulate what you want from another person, you definitely can’t articulate it to an AI system. The technology is remarkably good at inferring intent from ambiguous inputs, which creates a dangerous illusion of understanding. Just because AI produces a response doesn’t mean it understood what you actually meant.

This communication gap has profound implications for AI literacy in organizations. It’s not enough to teach people which buttons to click or which prompts to use. Effective AI implementation requires developing new communication skills, learning to be more precise, more explicit, and more aware of ambiguity in our own thinking. The executives who succeed with AI won’t necessarily be the most technical. They’ll be the ones who can clearly define problems, articulate constraints, and recognize when AI outputs miss the mark.

The Knowledge Skills Gap We’re Ignoring

While everyone debates the AI skills gap, whether workers can adapt to AI tools, Mo identified a more troubling concern: the knowledge skills gap. What happens to institutional knowledge development when AI systems handle increasing amounts of cognitive work? Organizations have always struggled with capturing and transferring expertise from experienced employees to newer ones. AI was supposed to help solve this problem by serving as a repository for institutional knowledge.

But there’s a catch. If junior employees never develop deep expertise because AI handles too much of their learning process, where does the next generation of institutional knowledge come from? Research has shown that repeated reliance on AI tools can lead to cognitive debt, where dependence leads to shallow processing and reduced ownership of ideas, potentially creating a generation that struggles with independent problem-solving.

Manufacturing has been grappling with this challenge through technologies like augmented reality. Companies tried using AR to capture institutional knowledge from retiring experts, training younger workers to perform at higher levels without decades of experience. The intent was good to preserve valuable knowledge, reduce dependence on an aging workforce. But the unintended consequence was eliminating the pathways through which people develop expertise in the first place.

We’re seeing the same pattern with AI, just accelerated. Knowledge workers have traditionally been somewhat protected from automation because their value came from expertise, judgment, and the ability to navigate complexity. AI doesn’t eliminate that value, but it does create pressure to demonstrate ROI on expensive human capital. When an AI model can be trained on decades of institutional knowledge for a fraction of the cost of employing experts, the economic calculus shifts dramatically.

The solution isn’t rejecting AI adoption. It’s being far more intentional about how we integrate these tools while preserving the development pathways that create expertise in the first place. That means identifying which tasks genuinely benefit from AI assistance versus which ones serve as crucial learning opportunities for developing professionals. It means recognizing that efficiency isn’t always the highest value; sometimes struggle and difficulty are features, not bugs, of the learning process.

From Hype to Reality: What Actually Works

The small percentage of AI implementations that succeed share common characteristics: they pick one specific pain point, execute well on solving it, and partner strategically with specialized vendors who understand both the technology and the business context. These successful implementations aren’t the flashiest or most ambitious. They’re focused, practical, and realistic about what AI can and cannot do.

Retrieval-augmented generation represents one of the most reliable implementation patterns. RAG systems combine the language capabilities of large language models with the specificity of your own knowledge base. Instead of asking a general-purpose chatbot to understand your business, you’re giving it direct access to your actual documentation, processes, and institutional knowledge. This approach significantly reduces hallucination risks while increasing the relevance of outputs.

But even RAG implementations fail when organizations don’t invest in the fundamentals. The quality of data matters immensely. The structure of your knowledge base matters. How you define the scope and guardrails of the system matters. Fifty-seven percent of organizations estimate their data is not AI-ready, meaning it hasn’t been prepared to prove fitness for specific AI use cases. Without AI-ready data, even the best implementation strategies will struggle.

The organizations finding success with AI share another common trait: they empower line managers, not just central AI labs, to drive adoption. When AI initiatives stay trapped in innovation departments, they remain divorced from the actual workflows and pain points they’re meant to address. Real value comes from the people closest to the work, identifying opportunities for AI augmentation and having the agency to implement solutions.

Strategy Over Sprinkles

Mo and I kept coming back to this idea of “sprinkling AI” on everything, the equivalent of earlier eras when companies would sprinkle IoT, blockchain, or quantum computing on problems without a clear strategy. Every emerging technology goes through this phase where it becomes a magic ingredient that executives think will automatically improve whatever it touches.

The problem with sprinkling is that it treats AI as a feature rather than a fundamental shift in how work gets done. Features can be added without deep integration. Features don’t require rethinking processes. Features allow organizations to claim they’re “doing AI” without actually transforming anything meaningful. This approach might work for innovation theater using AI primarily for marketing perception, but it doesn’t deliver real business value.

The MIT research found that the biggest problem wasn’t that AI models weren’t capable enough; executives tended to think that was the issue, but rather that companies were making poor choices in how they used the technology. Strategic clarity beats technological sophistication every time. A focused implementation of a less advanced AI system will outperform an unfocused deployment of cutting-edge models.

Begin with the end in mind. What specific problem are you trying to solve? How will you measure success? What processes need to change to accommodate AI integration? Who needs training, and what do they need to learn? These aren’t sexy questions. They don’t make for exciting board presentations. But they’re the questions that separate the 5% of successful implementations from the 95% that stall.

The American Innovation Paradox

Toward the end of our conversation, we discussed an interesting historical parallel from the semiconductor industry. American companies have always excelled at tip-of-the-spear innovation, being first to market with breakthrough technologies. But we’ve historically struggled with the follow-through: the careful implementation, the integration into existing systems, the unglamorous work of making innovation actually productive.

Silicon Valley created the chips that powered the computing revolution, then licensed them to Japan. American companies assumed their technological leadership was permanent. Japan took those chips and figured out how to manufacture them efficiently, how to integrate them into consumer products, and how to build entire industries around them. By the time American companies realized what was happening, they’d ceded massive portions of the market.

We’re seeing the same pattern with AI. American companies pioneered large language models, raced to deploy them publicly, and assumed global leadership was assured. Then China released DeepSeek, claiming to train comparable models for a fraction of the cost OpenAI and Anthropic require. Whether those cost claims are accurate or not, the message is clear: first-mover advantage doesn’t guarantee sustained leadership without excellence in implementation.

The AI race isn’t won by whoever builds the most powerful model. It’s won by whoever figures out how to reliably create value with AI systems at scale. That requires the patient, detail-oriented work that American companies often undervalue. It requires thinking beyond the technology itself to the organizational, cultural, and process changes needed to make the technology productive.

Practical Steps for Avoiding the 95%

For organizations serious about joining the successful 5% rather than the failing 95%, the path forward requires discipline and realism. Start by identifying a specific, well-defined problem where AI can add clear value. Not a vague aspiration like “improve customer service” but something concrete like “reduce time spent on contract review” or “automate routine data entry in financial reporting.”

Invest in making your data AI-ready. This isn’t optional infrastructure - it’s the foundation that determines whether any AI implementation can succeed. That means cleaning data, establishing governance, documenting processes, and creating the structured knowledge bases that AI systems need to be useful in your specific context.

Partner with specialized vendors rather than building everything in-house. The success rate difference is too significant to ignore. Specialized vendors bring lessons learned from multiple implementations. They’ve already made mistakes on someone else’s dime. They understand both the technology and the specific business domain in ways that internal teams often can’t match without significant investment.

Empower the people closest to the work to identify opportunities and drive adoption. Central AI labs have their place in establishing standards and governance, but real value comes from line managers who understand exactly where AI could save time, reduce errors, or improve outcomes in their specific domains.

Most importantly, maintain the human-machine collaboration balance. Use AI to augment human capabilities, not replace human judgment. Be intentional about which tasks you automate versus which ones serve as crucial learning opportunities. Recognize that efficiency isn’t always the highest value; sometimes the struggle is the point.

Looking Forward: The Window Is Closing

The window for gaining a competitive advantage with AI is narrowing. Not because the technology is going away, but because the learning curve exists whether you engage with it now or later. The organizations investing time in understanding AI’s actual capabilities, limitations, and optimal use cases are building institutional knowledge that becomes increasingly valuable as the technology matures.

We’re not heading toward a world where AI replaces human workers overnight. We’re heading toward a world where humans who know how to work effectively with AI replace humans who don’t. That’s a crucial distinction. The threat isn’t the technology itself; it’s the widening gap between workers and organizations that develop AI fluency versus those that don’t.

Gartner predicts that, despite current challenges, continued steady investment in and adoption of AI will lead organizations to shift from experimentation to scaling foundational innovations. The trough of disillusionment is temporary. The organizations that use this period to figure out what actually works, rather than abandoning AI altogether, will emerge stronger when the technology reaches the plateau of productivity.

The conversation Mo and I had at that coffee shop ultimately reinforced something I already believed: the hard problems in AI adoption aren’t technical. They’re human. They’re about communication, strategy, organizational change, and maintaining cognitive capabilities while adopting powerful new tools. The technology will keep improving whether we engage thoughtfully with these challenges or not. The question is whether we’re willing to do the unglamorous work of figuring out what AI is actually good for, rather than what we wish it could do.

Your AI intern isn’t going to save you. But with proper onboarding, clear direction, and realistic expectations, it might actually be able to help.

In our recent Facing Disruption webcast conversation, I spoke once again with Ed Martin, a digital twin consultant and former manufacturing industry veteran who spent over 25 years working across Autodesk and Unity before launching his consulting practice. Martin brings deep expertise in simulation modeling, control systems, and the convergence of physical and digital systems. Throughout our discussion, we explored synthetic data - artificially generated information that mimics real-world statistical properties without containing actual observations is fundamentally changing how organizations develop digital twins, train artificial intelligence systems, and navigate the complex landscape of data privacy. Our conversation reveals that synthetic data represents far more than a technical workaround; it’s enabling entirely new approaches to innovation that weren’t previously possible.

Key Takeaways

Before diving into our conversation with digital twin consultant Ed Martin, here are the essential insights about synthetic data and its transformative role in digital innovation:

What Makes Synthetic Data Different

Synthetic data isn’t dummy data or anonymized records - it’s artificially generated information that maintains real-world statistical properties and correlations without containing actual observations. Think of it as a lucid dream: structured, coherent, and realistic, rather than random noise.

Why It Matters Now

The market is exploding (projected 35%+ annual growth through 2034) because synthetic data solves a critical paradox: organizations need vast datasets to train AI and test systems, but real-world data is expensive, slow to collect, and increasingly restricted by privacy regulations.

Core Applications

• Digital twins: Enabling what-if scenario modeling for conditions that haven’t occurred yet

• Edge case generation: Creating rare but critical scenarios (99% of autonomous vehicle training needs come from <1% of driving conditions)

• Privacy protection: Generating realistic data that can’t be traced back to individuals

• Accelerated development: Compressing years of data collection into weeks of generation

Critical Risks to Manage

• Incomplete coverage of real-world variability creates dangerous blind spots

• Source data biases amplify through generation

• False confidence when synthetic data looks realistic but misses critical patterns

• Adversarial manipulation as systems increasingly depend on generated data

Success Framework

1. Start with problem definition, not technology selection

2. Assess existing data against the five V’s: volume, velocity, variety, value, veracity

3. Match methodology to use case (classical simulation, GANs, VAEs, or diffusion models)

4. Invest in verification as seriously as generation - test against real-world data relentlessly

5. Maintain human expertise for judgment, validation, and strategic oversight

The Bottom Line

Synthetic data shifts organizations from pure observation to principled imagination - enabling exploration of possibilities at unprecedented scale. But quality answers depend entirely on whether synthetic scenarios realistically represent their real-world counterparts. It’s augmented intelligence, not autonomous replacement, requiring rigorous verification and human judgment to separate lucid dreams from hallucinations.

The Evolution From Noise to Intelligence

For decades, engineers and developers have worked with two primary types of non-production data. The first is anonymized real-world data, where personally identifiable information has been stripped or obscured. The second is dummy data randomly generated values that fill database fields during testing. Both approaches have limitations that synthetic data addresses in fundamentally different ways.

Anonymized data, while derived from actual observations, remains vulnerable to re-identification attacks. Teams of engineers may carefully remove names, addresses, and obvious identifiers, yet sophisticated analysis can often reverse-engineer the original information through pattern matching and correlation. This risk has only grown as machine learning techniques have become more sophisticated. A single overlooked data field or an incomplete understanding of how information can be cross-referenced has led to numerous high-profile data breaches and privacy violations.

Dummy data presents the opposite problem. As Ed explains during our conversation, dummy data functions essentially as white noise statistically random information with no meaningful patterns or correlations. While it can verify that a system accepts certain data types or that a database schema functions correctly, it offers nothing for training algorithms or understanding system behavior. It’s the data equivalent of testing a car by pushing it downhill rather than starting the engine. The mechanics may appear to work, but you learn nothing about actual performance.

Synthetic data occupies an entirely different category. It’s artificially generated, making it impossible to trace back to actual individuals, yet it maintains the statistical properties, correlations, and patterns of real-world data. When properly constructed, synthetic datasets capture the essence of how variables relate to each other, how distributions cluster, and how edge cases manifest what Ed memorably described as

“a lucid dream as opposed to one of those crazy dreams where you wake up and wonder what it was.”

The market has responded accordingly, with the global synthetic data generation market valued at over $310 million in 2024 and projected to grow at a compound annual growth rate exceeding 35% through 2034. This explosive growth reflects not just technological advancement but a fundamental shift in how organizations approach data challenges.

Methods and Models: Choosing the Right Tool

Creating effective synthetic data requires understanding both the type of information you need to generate and the characteristics of your source data. Organizations today have multiple methodological approaches available, each with distinct strengths and limitations.

The foundation starts with classical simulation techniques approaches that predate the current AI boom by decades. Ed recalled his early career working with Simulink and state-space models 25 years ago, developing control systems through entirely deterministic simulations. These classical approaches excel when you deeply understand the underlying system you’re modeling. For a manufacturing process with well-characterized physics, a boiler with known thermodynamic properties, or a mechanical system with defined tolerances, simulation models can generate synthetic data that perfectly captures system behavior across valid operating ranges. The synthetic data from these models doesn’t just resemble reality, it mathematically represents it within specified parameters.

Procedural authoring tools offer another non-AI approach, particularly valuable for visual and spatial data. Using node-based systems, engineers can define parameters and rules that generate variations within realistic ranges. Software platforms like Houdini have made this approach standard for creating surface textures, environmental variations, and other content where controlled randomness within constraints produces useful diversity. These techniques shine when you need many variations of fundamentally similar data, different weathering patterns on the same surface, for example, or variations in lighting conditions.

The AI-driven approaches expand these capabilities dramatically, particularly for complex data where underlying relationships aren’t fully understood or easily modeled. Recent research has demonstrated how Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) can be integrated with digital twin architectures to generate new datasets reflecting possible future scenarios while ensuring data integrity.

Generative Adversarial Networks

GANs operate through competition between two neural networks. A generator creates synthetic data candidates, while a discriminator attempts to distinguish real from generated samples. This adversarial relationship, I characterize as “the art forger and the art inspector” forces continuous improvement. The generator must become increasingly sophisticated to fool the discriminator, while the discriminator must develop more nuanced detection capabilities. The result, after sufficient training, is a generator capable of producing highly realistic synthetic data within its specific domain.

Variational Autoencoder

VAEs take a different approach by encoding data into a lower-dimensional latent space essentially capturing the fundamental features or “essence” of the data then decoding back into the original distribution. Here’s a useful analogy: imagine summarizing a movie in as few words as possible, then having someone reconstruct the movie from that summary. The encoding process identifies core patterns and relationships, while the decoding process can generate new variations that maintain those fundamental characteristics. This approach works particularly well for time series data and scenarios where you need to maintain complex correlations between variables.

Diffusion models

Diffusion models, the technology behind popular image generation tools like Midjourney, add controlled noise to data then learn to reverse the process. By training neural networks to progressively denoise images or other data, these systems learn to generate novel content that maintains the statistical properties of training data. While computationally expensive, diffusion models have demonstrated remarkable capabilities for generating high-quality synthetic images and are increasingly being applied to other data types.

Each method presents trade-offs in sample quality, training stability, computational cost, and data requirements. Classical simulations offer perfect accuracy within their valid ranges but require deep domain expertise to construct. GANs can produce exceptional quality but may suffer from mode collapse or training instability. VAEs train more reliably but may produce blurrier outputs for image data. Diffusion models generate excellent results but demand significant computational resources. The art lies in matching methodology to use case a determination that requires both technical expertise and practical experience.

Digital Twins: Where Synthetic Data Becomes Essential

The connection between synthetic data and digital twins extends beyond mere convenience into fundamental necessity. Digital twins virtual representations of physical objects or systems synchronized with real-world counterparts face unique data challenges that make synthetic generation not just useful but often indispensable.

Digital twins serve two primary functions: real-time monitoring and what-if scenario modeling. For monitoring, digital twins track the current state of physical assets manufacturing equipment, supply chains, building systems, or even financial portfolios. They ingest sensor data, operational metrics, and environmental conditions to maintain an up-to-date virtual representation. This monitoring function relies primarily on actual data streams from the physical world.

The scenario modeling function, however, demands synthetic data. When organizations want to understand how a system might behave under different conditions: a supply chain disrupted by weather events, a manufacturing line running at higher speeds, a building’s HVAC system responding to unusual heat they need data representing conditions that haven’t yet occurred or can’t be safely tested in reality. Synthetic data generation becomes the engine for exploring these hypothetical futures.

The autonomous vehicle domain provides a particularly stark illustration. Vehicles driving normally down expressways or through cities generate massive amounts of data, but more than 95% of that data contains no new information. The car stays between the lanes, maintains safe following distance, and encounters no unusual situations. This data does nothing to train perception systems or improve decision-making algorithms. The valuable data the corner cases where something unexpected happens rarely occurs in real-world driving but represents the scenarios that determine system safety and reliability.

What happens when a bug splatters across a camera lens? When sun glare temporarily blinds a sensor? When road markings are worn away or construction equipment partially obstructs a lane? These edge cases might collectively represent a tiny fraction of actual driving conditions, but they’re precisely the scenarios where autonomous systems must perform flawlessly. Collecting sufficient real-world examples would require millions of miles of driving and years of time. Synthetic data generation allows these scenarios to be created, varied, and tested systematically.

Financial services applications present similar dynamics. Using digital twins to identify fraudulent behaviors requires training systems on examples of fraud data that legitimate transactions provide in limited quantities. Synthetic data can generate realistic transaction patterns representing various fraud scenarios, allowing detection algorithms to learn patterns that might otherwise require years of real-world fraud to accumulate. The synthetic approach also avoids privacy concerns inherent in sharing actual customer financial data, even in anonymized form.

Recent medical research has demonstrated how Latent Diffusion Models can edit digital twins to create what researchers call “digital siblings” variations that maintain core characteristics while introducing subtle anatomic differences. These siblings enable comparative simulations revealing how anatomic variations impact medical device deployment, augmenting virtual cohorts for improved device assessment without requiring impossible numbers of actual patients.

Navigating the Pitfalls: Where Synthetic Data Fails

Despite its potential, synthetic data carries risks that organizations must actively manage. In our conversation, we identified several critical failure modes that can undermine synthetic data initiatives or, worse, create false confidence in flawed systems.

The most fundamental risk is incomplete coverage of real-world variability. Synthetic data generation systems can only produce variations they’ve been designed or trained to create. If your source data or generation methodology doesn’t account for certain conditions, behaviors, or edge cases, your synthetic data will have blind spots and systems trained on that data will fail when they encounter the missing scenarios in reality. This isn’t a theoretical concern - failing to capture outliers and corner cases can leave systems vulnerable to exactly the situations they most need to handle correctly.

Data quality issues from source data propagate insidiously through synthetic generation. If your real-world data contains bias, whether demographic bias in image datasets, operational bias in manufacturing data, or any other systematic skew, synthetic data trained on that source will reproduce and potentially amplify those biases. The garbage-in, garbage-out principle applies with particular force to synthetic generation because the amplification happens invisibly within model training processes.

For example of a boiler system where temperature increases should correlate with pressure increases in a closed system. If synthetic data shows temperature rising without corresponding pressure changes except in scenarios explicitly designed to simulate sensor failures the data fails to represent physical reality. Any system trained on that data will have fundamental misunderstandings about how the world works.

The bias problem extends beyond technical accuracy into ethical and legal dimensions. Privacy regulations like GDPR create complex compliance challenges for synthetic data, as it’s only exempt from regulation when it avoids memorization, overfitting, and indirect re-identification. Poorly constructed synthetic data can inadvertently contain patterns that allow reconstruction of original data sources, negating the privacy benefits that motivated synthetic generation in the first place.

Copyright and intellectual property concerns add another layer of complexity. Depending on how synthetic data is generated and what it’s used for, organizations may face questions about the rights and restrictions associated with source data. If training data includes copyrighted material or proprietary information, the legal status of resulting synthetic data may be unclear or restricted.

Cost considerations, while less dramatic than technical failures, can derail synthetic data initiatives through accumulated expenses. Generating high-quality synthetic data at scale using sophisticated AI models demands significant computational resources. Cloud computing costs can escalate quickly, particularly for diffusion models or other computationally intensive approaches. Organizations must evaluate whether synthetic generation truly offers economic advantages over alternative approaches for their specific use case.

Perhaps most insidious is the risk of false confidence. When synthetic data looks realistic, systems perform well in testing, and stakeholders see impressive demonstrations, it’s easy to assume the synthetic data adequately represents reality. Without rigorous verification against actual real-world conditions and continuous testing for edge cases, organizations can deploy systems that fail catastrophically when they encounter conditions their synthetic training data didn’t capture.

Building Robustness: Verification and Red Teams

Ed’s background in manufacturing and control systems shaped his emphasis on verification as a non-negotiable element of working with synthetic data. He recalled a formative early career lesson: you might fall in love with your model, but you must constantly check it against real-world behavior. Models always have flaws; perhaps they’re fundamentally sound but handle certain operating regions poorly. The discovery of those limitations, rather than being disappointing, represents essential knowledge about where your systems work and where they don’t.

This verification imperative becomes more challenging as AI systems grow more sophisticated. When synthetic data generation, digital twin operation, and decision-making all involve AI components, verification can’t rely on simple comparison between predicted and actual outcomes. The systems operate in high-dimensional spaces with complex interactions that may not manifest obviously when something goes wrong.

And yet, here is another crucial question of adversarial threats: the cybersecurity dimension of synthetic data. As systems increasingly depend on synthetically generated information, the attack surface expands. A malicious actor who can inject biased or corrupted data into generation processes could subtly manipulate the synthetic data, which then influences model training, which ultimately affects real-world decision-making. The social engineering of language models, where users convince chatbots to provide incorrect information or make invalid commitments, demonstrates how vulnerable AI systems can be to adversarial manipulation.

The solution involves red team approaches deliberately attempting to break or manipulate systems to identify vulnerabilities before adversaries exploit them. Ed characterizes this as essential for robustness, though implementing effective red teaming for synthetic data systems requires specialized expertise. How do you verify that bias hasn’t been injected? How do you detect when synthetic data has drifted from realistic distributions? How do you ensure the chain of custody for data and models remains secure?

The concept of “chain of custody” for data tracking its origins, transformations, and usage throughout its lifecycle emerged as particularly important. Organizations need to know what base models they’re using, what data trained those models, what fine-tuning has been applied, and what synthetic data derives from what sources. This traceability enables both security verification and quality assurance.

Explainable AI techniques offer partial solutions by providing visibility into how models make decisions, but the field has grappled with explainability challenges for over a decade without reaching definitive solutions. Even sophisticated explainability tools may struggle to surface subtle biases or detect adversarial manipulation designed to avoid detection.

The conversation touched on using AI to verify AI employing higher-level models with broader context to validate outputs from lower-level specialized models. This layered approach adds guardrails but also adds complexity and computational cost.

“It’s going to be AI all the way down.”

Ultimately, verification requires combining multiple approaches: systematic testing against real-world data when available, domain expert review of synthetic data and model outputs, adversarial testing through red teams, automated monitoring for distributional drift, and maintaining careful documentation of data provenance and transformations.

The Human-AI Partnership: Augmentation Over Automation

Throughout the conversation, we both pushed back against framing AI and synthetic data as replacements for human expertise. Instead, we characterized these technologies as powerful augmentation tools that amplify human capabilities while remaining dependent on human judgment for effective deployment.

AI systems function like “an over-eager intern who has zero knowledge but is willing to work incredibly hard and is going to come back to you with an answer no matter what, even if it’s wrong.”

This characterization captures both AI’s power, its ability to process vast amounts of data and identify patterns no human could track and its fundamental limitation: the absence of meaning, context, and judgment.

AI excels at pattern recognition across massive datasets. It can identify correlations in sensor data from thousands of devices, spot anomalies in financial transactions, or recognize objects in millions of images. These capabilities surpass human cognitive capacity by orders of magnitude. Yet AI lacks the semantic understanding, prior knowledge, and conceptual frameworks that humans bring to complex problems.

Personally, I prefer framing AI as “augmented intelligence” rather than “artificial intelligence” , a power tool for people to use rather than an autonomous decision-maker. This perspective manifests practically in domains like medical diagnostics, where language models can identify rare diseases by scanning comprehensive medical literature and matching symptom patterns. No individual physician can remember every rare condition or maintain current knowledge across all medical specialties. An AI system that surfaces relevant possibilities augments the physician’s diagnostic process without replacing the clinical judgment required to evaluate those suggestions.

The conversation acknowledged that this augmentation creates challenges for people entering professional fields. Junior professionals must now compete with AI-augmented experts who can leverage vast knowledge repositories and analytical capabilities. The knowledge gap between experienced professionals and newcomers potentially widens as senior people master AI tools while junior people are still building fundamental expertise.

Yet this challenge doesn’t diminish the fundamental truth that synthetic data and AI remain tools requiring skilled operators. Someone must define what problem needs solving. Someone must evaluate whether synthetic data adequately captures relevant real-world variability. Someone must interpret AI outputs, catch errors, and apply contextual judgment. The expertise required shifts from routine execution toward higher-level strategy, validation, and interpretation but expertise remains essential.

AI can generate synthetic data far faster than humans could collect real data. It can train models on that data and produce impressive results. But only human experts can meaningfully verify that the synthetic data captures what matters, that the models perform appropriately, and that the entire system will work reliably in real-world deployment.

Getting Started: A Practical Framework

For organizations considering synthetic data initiatives, our conversation offers a pragmatic framework grounded in both technical understanding and business reality. Ed emphasizes starting not with technology selection but with problem definition.

• What specific challenge are you trying to solve?

• Why is it important?

• What would success look like?

Only after clearly defining the problem should organizations evaluate whether synthetic data represents an appropriate solution. Not every data challenge requires synthetic generation. Sometimes real-world data collection, despite its costs and constraints, remains the better approach. Sometimes simpler techniques like data augmentation or transfer learning suffice. Synthetic data becomes compelling when you need data that doesn’t exist yet, can’t be safely collected, or would require prohibitive time and expense to acquire through real-world observation.

With the problem and approach defined, the next step involves assessing what data already exists and what characteristics it has. I think it’s appropriate to fall back on the traditional “five V’s” of big data: volume, velocity, variety, value, and veracity. Organizations should understand not just whether they have data, but whether they have enough of it, whether it captures necessary diversity, whether it’s reliable, and whether it updates with sufficient frequency. These characteristics determine what synthetic data needs to provide.

The technical methodology selection follows from this assessment. Organizations need expertise in both the problem domain and the synthetic data generation techniques likely to apply. Classical simulation models require different skills than training GANs or diffusion models. The optimal approach depends on data types (images versus time series versus tabular data), available source data, quality requirements, and computational constraints.

Ed strongly recommends bringing in experienced practitioners who have “lived the pain of deployments that didn’t go well.” Theoretical knowledge of synthetic data techniques doesn’t substitute for practical experience with the myriad ways implementations can fail. Someone who has debugged biased training data, recovered from mode collapse in GANs, or discovered that their synthetic data missed critical edge cases brings invaluable pattern recognition to new initiatives.

Organizations should start small and iterate. Rather than immediately attempting to generate comprehensive synthetic datasets for critical applications, begin with limited scope pilots that allow learning and refinement. Establish verification processes early, comparing synthetic data outputs against real-world observations whenever possible. Build expertise gradually while managing risk.

The conversation emphasized that this isn’t purely a technical initiative. Synthetic data projects require coordinating domain experts who understand what the data should represent, data scientists who can implement generation techniques, engineers who will use the synthetic data to develop systems, and business stakeholders who define success criteria and acceptable risk levels. Getting these groups aligned around shared understanding and expectations often determines project success more than any technical decision.

The Road Ahead: Promise and Precaution

As the conversation concluded, we reflected on synthetic data’s trajectory. Ed noted that synthetic data references have exploded in recent months, though he was careful not to overhype the trend. While new attention brings new applications and innovations, synthetic data itself isn’t new. Engineers and researchers have used these techniques for years under different names.

What has changed is the scale, sophistication, and accessibility of synthetic data generation tools. The same AI advances that brought large language models to mainstream awareness have also dramatically improved synthetic data capabilities across domains. Organizations that previously couldn’t justify the specialized expertise or computational resources required for synthetic data can now access cloud-based platforms and pre-trained models that lower entry barriers.

Industry experts project that by 2024, approximately 60% of data used to develop AI and analytics projects will be synthetically generated. This shift reflects synthetic data’s practical advantages: faster time-to-development, reduced privacy risk, better control over data characteristics, and the ability to generate corner cases that would be prohibitively expensive to capture through real-world collection.

A word of caution though. The same factors that make synthetic data powerful: its ability to look realistic while being entirely generated create risks when that data doesn’t adequately represent reality. Organizations must balance the efficiency gains and privacy benefits of synthetic data against the verification burden and potential for blind spots.

Our conversation returned repeatedly to the theme of transparency and accountability. As synthetic data becomes more prevalent, organizations need clear frameworks for documenting what data is synthetic, how it was generated, what it’s used for, and how it’s been validated. This transparency serves multiple purposes: enabling technical review and quality assurance, supporting regulatory compliance, and maintaining trust with stakeholders who depend on systems trained on synthetic data.

Looking forward, the integration of synthetic data generation with digital twin architectures represents a particularly promising direction. As organizations build digital twins of everything from manufacturing facilities to urban infrastructure, the ability to generate synthetic operational data enables more sophisticated scenario modeling and predictive analytics. The digital twin becomes not just a passive mirror of reality but an active experimentation platform for exploring possibilities.

The cybersecurity dimension will likely grow in importance as synthetic data becomes more critical to system development and operation. Just as organizations now routinely consider data security and access controls, they’ll need to develop expertise in protecting synthetic data generation processes from adversarial manipulation. This includes securing training data, validating model integrity, and continuously monitoring for signs of corruption or bias injection.

Actionable Recommendations

For Technical Leaders: Start by auditing existing data assets against the five V’s framework, identifying specific gaps that synthetic data could address. Prioritize use cases where synthetic generation offers clear advantages over real-world collection training data for rare scenarios, privacy-sensitive applications, or rapid prototyping needs. Invest in verification capabilities before scaling synthetic data usage, establishing processes for validating generated data against real-world observations and domain expert review.

For Business Executives: Frame synthetic data as an enabler rather than a silver bullet. It accelerates development cycles, reduces privacy risk, and enables scenario modeling that would otherwise be impractical but it doesn’t eliminate the need for domain expertise, quality assurance, or real-world validation. Budget for both generation capabilities and verification processes. Consider synthetic data’s strategic implications for product development timelines, competitive positioning, and risk management.

For Data Scientists and Engineers: Develop expertise across multiple synthetic data generation techniques rather than specializing narrowly. Different methods suit different data types and use cases. Build robust data pipelines that maintain clear lineage from source data through synthetic generation to final usage. Implement automated monitoring for distributional drift and quality degradation. Document generation processes and validation results thoroughly to support both technical review and compliance requirements.

For Risk and Compliance Teams: Engage early in synthetic data initiatives to ensure privacy, security, and regulatory requirements are addressed in system design rather than retrofitted later. Evaluate synthetic data not just on whether it contains personally identifiable information but on whether it could enable indirect re-identification through pattern matching. Develop frameworks for assessing when synthetic data requires the same controls as real data. Consider synthetic data’s role in your organization’s overall data governance strategy.

Synthesis: Data as Imagination

Synthetic data represents a subtle but fundamental shift in how organizations think about information. For decades, data meant observations recording what happened, measuring what exists, documenting reality. Synthetic data inverts this relationship. It generates what could be, creating plausible futures and hypothetical scenarios grounded in realistic patterns but unconstrained by having to have actually occurred.

This shift from observation to imagination opens new possibilities for innovation while creating new responsibilities for verification and validation. Organizations can explore possibilities faster and more safely than real-world experimentation allows, but they must ensure their synthetic explorations remain tethered to reality. The technology enables asking “what if” at unprecedented scale: what if this supply chain disruption occurred, what if this sensor failed, what if this rare condition appeared but the quality of answers depends entirely on whether the synthetic scenarios realistically represent their real-world counterparts.

Synthetic data isn’t replacing human expertise or eliminating the need for real-world data. It’s augmenting human capabilities and filling gaps that real-world data collection cannot practically address. The organizations that will benefit most from synthetic data are those that understand both its power and its limitations, that invest in verification as seriously as generation, and that maintain the human judgment necessary to separate lucid dreams from mere hallucinations.

The rise of synthetic data marks not the end of the data collection era but the beginning of a more sophisticated approach where observation and imagination work together where digital twins reflect reality while exploring possibility, where algorithms learn from experiences that haven’t happened, and where organizations navigate complexity through carefully constructed simulations that inform rather than replace engagement with the messy, unpredictable real world.

In this episode of Facing Disruption, I spoke with Eli Becker, senior creative strategist at Teak, to explore this tension. Eli’s career trajectory from book publishing to agency creative strategy provides a unique lens for understanding how storytelling evolves when machines can mimic human creativity. Our conversation reveals that the challenge isn’t whether AI will replace human storytellers, but rather how the fundamental meaning of authenticity itself is being redefined.

Key Takeaways

The Market Shift: We’ve moved from an intelligence-based economy to a creativity-based one. Information is now commoditized, making creative thinking and strategic questioning the new competitive advantage.

Authenticity Redefined: AI is degrading the meaning of authenticity from “vulnerable and distinctive” to simply “made by humans.” Brands must resist this dilution by maintaining genuine emotional connections.

The Human Element Stays: Creative strategy requires identifying nuance, tension, and contradiction - the opposite of what AI does well with aggregate data. Human empathy and cultural understanding remain irreplaceable.

Entry-Level Crisis: Junior positions are disappearing as AI handles repetitive tasks, creating a skills gap where emerging talent can’t develop the experience needed for senior creative roles.

All Purchases Are Emotional: Even B2B buying decisions are tied to identity, fear, and aspiration. Successful brands acknowledge these emotional drivers rather than treating business as purely rational.

Speed vs. Quality: AI accelerates workflows, but organizations must resist filling saved time with more output. Instead, reinvest in creative exploration, play, and thoughtful iteration.

From Intelligence to Creativity: A Market Transformation

We are witnessing a profound shift in what creates competitive advantage. For decades, access to information determined market leaders. Companies that could analyze data faster, identify patterns more efficiently, and synthesize insights more comprehensively held the upper hand. That era is ending.

Eli articulates this transformation clearly: “We’ve shifted from an era of intelligence based to an era of creative, creativity based because information is at our fingertips. Really, our only limitation now is the creativity and being able to come up with the ideas or the questions.” This observation aligns with research showing how AI adoption has fundamentally altered creative industries. Over 87% of U.S. creative professionals now use AI tools in their work, with creative professionals reporting 20% time savings.

The implications extend far beyond workflow efficiency. When information synthesis becomes commoditized, the ability to ask the right questions, identify meaningful patterns in human behavior, and craft narratives that resonate emotionally becomes the scarce resource. Organizations are responding accordingly. Eli notes an uptick in creative strategy positions and recruiting activity, suggesting that companies recognize the growing value of professionals who can bridge analytical insight with creative execution.

This shift manifests in measurable ways. Research involving 1,000 business leaders found that those who received AI-generated questions to help with ideation produced 56% more ideas, with a 13% increase in the diversity of ideas and a 27% increase in the level of detail compared to a control group. Yet the same study revealed that 35% of C-suite leaders said AI changed how they talk about products and services without changing what they actually do, suggesting a fundamental disconnect between technology adoption and strategic transformation.

Action Step: Audit your organization’s investment balance between data analytics roles and creative strategy positions. If you’re still heavily weighted toward intelligence gathering rather than creative execution, you’re fighting yesterday’s battle. Consider shifting budget and headcount toward roles that bridge strategic insight with creative storytelling.

The Authenticity Redefinition

Perhaps the most provocative insight from our conversation concerns how AI is changing the very definition of brand authenticity. Eli challenges the conventional wisdom around this widely-used marketing term: “Authenticity has been watered down to simply mean, like made by humans instead of this deeper acknowledgement of a person’s ability to be vulnerable or courageous or distinctive.”

This observation cuts to the heart of a critical tension. As AI-generated content floods digital channels, authenticity risks becoming a binary designation - human versus machine - rather than a measure of genuine connection, vulnerability, or distinctive perspective. The bar lowers from “Who are you?” to “Is it human?” This degradation threatens to desensitize audiences to what Eli calls “authentic living,” potentially creating a world where merely being human-created becomes sufficient for claiming authenticity.

Research supports concerns about this deterioration. Trust levels are eroding as people grapple with an increasingly disruptive environment, with the UK experiencing some of the greatest drops in trust between 2023 and 2024. Meanwhile, there is a significant divide between trust in the technology sector itself and innovation around artificial intelligence, with little confidence in the legitimacy of AI.

For brands, this creates both challenge and opportunity. Organizations that can maintain genuine authenticity - defined by vulnerability, distinctive perspective, and meaningful connection rather than simply human authorship - will increasingly stand apart. Research examining brand authenticity, attachment, trust and loyalty found that all dimensions of brand authenticity exert notable positive impacts on brand attachment, brand trust and brand loyalty, confirming that authentic connections drive tangible business outcomes.

Action Step: Review your brand’s last five major marketing campaigns. Ask honestly: Are we demonstrating vulnerability, distinctive perspective, and genuine understanding of customer lives? Or are we simply checking the “human-created” box while playing it safe? Identify one way to take a more courageous stance that reflects your actual values and customer insights.

The Irreplaceable Human Element in Creative Strategy

Eli’s description of creative strategy work illuminates why certain aspects of brand building resist automation. A creative strategist occupies the space between product and audience, between logic and creativity, working to understand not just what problems exist but the emotional relationship audiences have with those problems. The role demands identifying insights - truths hidden in plain sight that, when articulated, prompt recognition rather than revelation.

This work remains distinctly human for specific reasons. AI excels at pattern recognition across aggregate data, but creative strategy requires the opposite: finding nuance, tension, and contradiction.

She uses AI tools like Notebook LM for organization and pressure-testing ideas, but the core work of conducting street interviews, reading between conversational lines, and identifying unstated motivations remains human-centered.

The distinction matters because emotional connection drives purchasing behavior in ways that exceed rational calculation. Research from Harvard Business School estimates that 95% of purchasing decisions are subconscious, driven by emotional rather than rational factors. Moreover, emotionally connected customers have a 306% higher lifetime value and stay with brands longer while being less price-sensitive.

Consider Eli’s example of BarkBox, the dog subscription service. When a customer’s dog passes away and a box is already en route, BarkBox compensates that box, sends a handwritten condolence note with the customer’s dog’s picture, and invites them to donate the box to another dog. This strategy extends beyond the transactional relationship to acknowledge the full emotional continuum of pet ownership. The approach generates organic advocacy, with customers sharing these experiences on YouTube and LinkedIn, creating awareness and positive sentiment that traditional marketing cannot replicate.

This level of strategic thinking - understanding the complete emotional journey, identifying meaningful touchpoints beyond transactions, and creating experiences that generate authentic stories - requires human empathy, cultural understanding, and the ability to read unspoken dynamics. Research on emotional marketing reveals that it significantly shapes purchasing decisions by enhancing consumer perceptions of brand authenticity, trustworthiness, and alignment with personal values.

Action Step: Map your customer’s complete emotional journey, not just their buying journey. Where do they experience vulnerability, fear, or joy in relation to the problem you solve? Identify three touchpoints beyond the transaction where you could demonstrate genuine understanding of their emotional experience. Implement at least one within the next quarter.

The Entry-Level Skills Crisis

While senior creative roles flourish, Eli acknowledges a concerning trend I’ve been tracking across multiple conversations: entry-level positions are disappearing or transforming dramatically. The traditional career ladder, where junior professionals develop skills through repetitive tasks before graduating to strategic work, is being disrupted. AI handles many tasks that previously served as training grounds for emerging talent.

The gap manifests in a troubling way. Organizations increasingly demand candidates who already know how to craft effective AI prompts, iterate with generative tools, and maintain creative quality in AI-assisted workflows. Yet developing these skills requires practical experience that entry-level positions traditionally provided. How do emerging professionals learn what constitutes a good story, compelling brand positioning, or effective creative strategy without opportunities to practice?

Industry research reveals this implementation gap: while 89% of respondents view AI as helpful for creative ideation and strategy, only 54% have fully integrated AI into their processes, and 47% of marketers lack a clear understanding of how to use or measure AI’s impact. This creates a paradox where tools are available but practical knowledge about their effective application remains scarce.

Eli’s advice for emerging creatives focuses on demonstrating both creativity and AI proficiency: “Letting your creativity shine and showing and proving that you do know how to leverage tools like AI to better produce stories is only going to help.” Yet this guidance assumes access to opportunities for developing such competencies - opportunities that are themselves becoming scarce.

The industry must grapple with this training challenge. Without pathways for emerging talent to develop creative judgment, strategic thinking, and brand intuition, organizations risk creating a future where senior roles cannot be filled because the pipeline for developing such expertise has dried up.

Action Step: If you’re in a leadership position, create intentional development pathways for junior talent. Design projects that blend AI tool usage with fundamental skill building. Establish mentorship programs where senior strategists guide emerging professionals through the nuanced work that AI can’t replicate. If you’re early in your career, seek out organizations committed to training, not just exploiting AI-enabled productivity.

Emotional Connection in a Data-Driven World

A persistent myth suggests that B2B purchasing decisions operate on pure logic while consumer decisions involve emotion. Eli dismantles this false dichotomy, and I’ve seen this play out repeatedly in my work: “Buying anything is an emotional process, and it’s very hard to separate emotion from our purchasing because what we buy at the end of the day, again, I would argue even in a professional setting, is tied to our identity.”

This insight matters because it reframes how organizations approach brand building across categories. Even enterprise software purchases involve emotional drivers - fear of job loss, desire for career advancement, need for security, relief from anxiety. Research confirms that on average, 86% of consumers’ purchasing decisions are influenced by 10 emotional factors, while emotional factors drive 56% of B2B purchasing decisions specifically.

The challenge becomes identifying and addressing these emotional undercurrents while navigating increased data availability. AI enables unprecedented personalization, allowing brands to customize messaging at the individual level based on behavioral data, contextual signals, and predictive analytics. This capability creates opportunities for micro-storytelling - crafting narratives tailored to specific moments in a customer journey.

Eli describes how brands can use this granularity: “With the level of technology that we have today, you can hyper customize the buyer journey, right? And you can look at each of those touch points throughout a user journey and find ways to tell micro stories.” The abandoned cart email exemplifies this approach - a moment where purchase history, browsing behavior, and timing converge to create an opportunity for personalized engagement.

Yet personalization divorced from genuine understanding risks feeling manipulative rather than helpful. The distinction lies in whether customization serves the customer’s interests or merely exploits their data. Authentic personalization demonstrates that a brand understands and values the customer; extractive personalization simply uses information to maximize conversions without regard for relationship quality.

Action Step: Choose one customer segment and conduct ten in-depth conversations about their emotional relationship with the problem you solve. Don’t ask about your product - ask about their fears, aspirations, and identity related to the challenge space. Use these insights to redesign one key touchpoint that addresses emotional needs rather than just functional ones.

When Speed Becomes the Enemy

Our conversation touched on a critical tension I’ve observed across industries: AI accelerates workflows, but acceleration can undermine quality.

This observation aligns with research about AI’s impact on creative industries. Studies emphasize that the future success of AI in creative fields depends on finding the right balance between technological innovation and preservation of human creativity, supported by clear ethical and legal guidelines. The question becomes not whether to use AI for acceleration, but rather how to deploy it without sacrificing the creative experimentation, reflection, and refinement that produces exceptional work.

Eli advocates for using the time AI saves to reinvest in creative exploration: “Maybe just take back a little bit of that time, you’re still consolidating your overall time to kind of sit with things, play a little bit more.” This approach treats AI as a tool for eliminating drudgery rather than simply compressing timelines. By automating boilerplate tasks, organizations create space for the unstructured thinking, serendipitous connections, and creative play that generate breakthrough ideas.

The practical challenge involves resisting organizational pressure to fill time saved with additional tasks. When AI reduces content creation time by 20%, the reflexive response involves producing 20% more content. The strategic response involves maintaining output while investing saved time in deeper research, more extensive iteration, or thoughtful reflection. Brands that resist the acceleration trap position themselves to produce work that stands out precisely because it reflects investment in quality over quantity.

Action Step: Conduct a team audit of time saved through AI adoption. Calculate the hours recovered weekly. Then make an explicit decision: Will you use that time to increase output volume, or will you protect it for creative exploration, deeper research, and thoughtful iteration? Build this decision into your workflow and hold yourself accountable to it.

The Renaissance Perspective

Despite acknowledging challenges, Eli maintains an optimistic view of AI’s impact on creativity that resonated deeply with me.

This perspective merits examination. Historical renaissances emerged when technological advances democratized previously restricted capabilities. The printing press enabled broad literacy; photography freed painting from representational obligations; digital tools made music production accessible beyond traditional studios. Each transformation initially provoked anxiety about devaluation of expertise, yet ultimately expanded creative possibility by enabling new forms of expression.

AI potentially follows this pattern. By handling technical execution, it lowers barriers for individuals with creative vision but limited production skills. A strategist with compelling brand insights but modest design ability can now prototype visual concepts. A writer with powerful storytelling instincts but limited video experience can create multimedia narratives. This democratization could indeed spark a creative flowering.

Yet renaissance requires more than tool access. It demands audiences capable of discerning quality, markets that reward distinctive work, and systems that support sustained creative development. The risk involves tool proliferation without accompanying growth in creative judgment, resulting in increased volume without commensurate quality improvement.

Research underscores this tension, noting that while AI democratizes access to creative tools and enables more innovative content creation, the importance of human insight to drive the creative process and oversight to mitigate AI-generated inaccuracies remains critical. The renaissance materializes only when technological capability combines with developed creative sensibility.

Action Step: Embrace experimentation while maintaining quality standards. Allocate 10-15% of your creative capacity to testing new AI-enabled formats, styles, or approaches. But establish clear criteria for what constitutes success beyond efficiency. Does it create genuine emotional impact? Does it reflect distinctive perspective? Does it strengthen customer relationships? Use these quality gates to determine what experiments scale.

Building Human-Centric Brands in the AI Era

Brand building in the AI era requires navigating a fundamental paradox. Technology enables unprecedented scale, personalization, and efficiency in creating and distributing marketing content. Yet these very capabilities threaten to commoditize communication, erode authenticity, and replace meaningful connection with algorithmic optimization.

The path forward involves neither wholesale AI rejection nor uncritical adoption. Instead, successful brands will use AI strategically to enhance human creativity rather than replace it. They will maintain clear distinctions between tasks where machines excel - pattern recognition, content generation, workflow automation - and domains where human judgment remains irreplaceable: emotional intelligence, cultural interpretation, ethical reasoning, and creative vision.

This shift demands that organizations cultivate creativity with the same rigor they previously applied to data analysis and operational efficiency.

Through this conversation, I’ve become convinced that the brands that thrive will be those that use technological leverage to deepen rather than replace human connection. They will invest in understanding the full emotional continuum of customer relationships. They will create authentic experiences that reflect genuine values rather than optimized messaging. They will tell stories that resonate because they emerge from real insight into human complexity rather than pattern matching against aggregate data.

Most importantly, they will remember that behind every metric, algorithm, and optimization sits a human being seeking connection, meaning, and recognition. Technology provides remarkable tools for reaching these individuals at scale. But only human creativity, empathy, and storytelling can truly engage their hearts.

Your Next Steps

As you consider how these insights apply to your organization, here are concrete actions to take this week:

For Brand Leaders:

• Schedule a strategy session to evaluate whether your team structure reflects the shift from intelligence-based to creativity-based competition

• Review your brand’s authenticity through the lens Eli describes: Are you demonstrating vulnerability and distinctive perspective, or just checking the “human-made” box?

• Identify one customer touchpoint where you can demonstrate deeper emotional understanding beyond the transaction

For Creative Professionals:

• Experiment with AI tools specifically for eliminating repetitive tasks, then protect the time saved for creative exploration rather than increased output

• Practice street-level research: Go talk to five real customers this month about their emotional relationship with the problems you solve

• Build a portfolio that demonstrates both AI proficiency and creative judgment - show you can use the tools while maintaining authentic voice

For Organizations:

• Create development pathways for junior talent that blend AI skill-building with fundamental creative training

• Establish quality criteria beyond efficiency: How do you measure emotional impact, distinctive perspective, and relationship strength?

• Commit to one high-risk creative experiment that demonstrates your brand values rather than playing it safe

The future belongs to brands that recognize creativity as the new competitive advantage while maintaining the human elements that create genuine connection. The question isn’t whether to adopt AI, but how to deploy it in service of deeper humanity rather than at its expense.

Leave a comment

On a recent episode of Facing Disruption, guest-host Chris Barlow had a chance to chat with Adam McLintock, Managing Director at Korn Ferry specializing in aerospace, defense, and government services executive search, at the Emerging Technologies for Defense Conference. With a background as a Navy pilot and over a decade placing executives in the defense sector, McLintock brings a unique perspective on how talent, technology, and acquisition reform are converging to create what he describes as very positive tailwinds for the industry. The conversation revealed not just the promise of emerging technologies, but the fundamental structural changes needed to bring innovation from concept to deployment.

The Acquisition Reform Imperative: Breaking Down Barriers to Innovation

The defense community has long acknowledged that its acquisition system needs modernization, but the urgency has never been greater. McLintock identifies acquisition and requirements reform as the central challenge, and opportunity, facing the sector today. The potential impact extends far beyond procurement efficiency. Meaningful reform would enable dual-use technology adoption, creating efficiencies for government operations while delivering better tools to warfighters through platforms that are actually easier to use and maintain.

The current system often forces promising technologies into a binary outcome: win a single contract or face extinction. This winner-takes-all approach stifles innovation and prevents the military from accessing multiple competing solutions that could serve different needs or contexts. A reformed system would fundamentally shift this dynamic, allowing prime contractors to work with multiple vendors and test various technologies simultaneously. Instead of defaulting to the lowest bidder, procurement decisions could prioritize the best technologies and crucially, there wouldn’t need to be just one winner.

Consider the example of drone technology development. In the current environment, a Series B startup developing advanced autonomous drone systems might spend two to three years navigating procurement processes, only to lose out to an established contractor with existing relationships but potentially inferior technology. By the time the contract is awarded, the startup’s technology may be obsolete, its funding exhausted, and its talent scattered to other opportunities. Under a reformed system, multiple drone technologies could be tested simultaneously in operational environments, with procurement decisions based on demonstrated performance rather than incumbent advantage.

Research from the Defense Innovation Board has consistently highlighted these challenges. Their 2019 report on software acquisition noted that traditional defense procurement timelines average 5-7 years from requirement to deployment, while commercial software cycles operate in months or weeks. This temporal mismatch creates a fundamental barrier to incorporating cutting-edge technology into defense systems.

Early-stage companies face what McLintock describes as a critical vulnerability period, the valley of death between prototype and production. Reform would provide these companies the breathing room to mature their technology and get it on the board to compete, rather than being eliminated before their solutions can demonstrate full capability. This maturation window is critical, it provides the time needed for rigorous testing and development while maintaining the company’s viability through the process. For Series A through D companies trying various entry points into the defense market, reform would create clear platforms to demonstrate technology effectiveness rather than forcing them to find backdoor approaches.

Subscribe now

The Public-Private Partnership Evolution: Beyond Transactional Relationships

The relationship between defense primes, emerging technology companies, and government buyers is undergoing a fundamental transformation. McLintock describes an ecosystem of co-op competition where companies simultaneously compete and collaborate. When a major contractor like L3 Harris takes the lead on a program involving technology from an earlier-stage company, streamlined processes and greater visibility across all parties becomes essential. True acquisition reform, in McLintock’s view, means making it easy for everyone involved while ultimately delivering the best product.

This collaboration model represents a significant departure from traditional prime contractor dynamics. Historically, large defense contractors maintained tight control over their supply chains and subcontractors, often absorbing or squeezing out smaller innovators. The emerging model recognizes that no single company, regardless of size, can maintain cutting-edge capabilities across all relevant technology domains.

Take shipbuilding as an example. McLintock notes that while shipbuilding presents unique challenges and isn’t as central to his work as some sectors, the potential for transformation across multiple related industries generates genuine excitement. Modern naval vessels integrate systems from dozens of specialized vendors, propulsion, radar, weapons systems, communications, and cyber defense. A reformed acquisition approach would allow the Navy to work directly with specialized technology providers while maintaining the systems integration role of traditional shipbuilders, accelerating innovation cycles without abandoning the industrial capacity needed for large-scale production.

The venture capital and private equity communities have recognized this shift, pouring unprecedented funding into defense technology startups. According to PitchBook data, defense tech startups raised over $33 billion between 2021 and 2023, compared to less than $10 billion in the preceding three-year period. This capital influx creates new dynamics in the talent market and changes the risk calculus for innovative companies entering the sector.

McLintock observes the practical implications daily in his executive search work. The ecosystem now includes both traditional primes and an entirely new class of well-funded startups backed by venture capital and private equity. These well-funded startups can now compete for top talent previously available only to established defense contractors or commercial tech giants, provided they can offer credible paths to actual defense contracts rather than perpetual prototype development.

Artificial Intelligence: Promise, Skepticism, and Practical Application

While artificial intelligence dominates conference agendas and investment theses, McLintock offers a refreshingly pragmatic perspective that many in the sector privately share but rarely voice publicly. He positions himself as an outlier on AI, acknowledging significant promise while questioning whether the expense justifies the ultimate value. This skepticism isn’t rooted in technological pessimism but rather in practical considerations about implementation, verification, and actual value creation.

Autonomy has existed in defense applications for years, McLintock points out, and machine learning has been deployed in various forms across the sector for even longer. What’s changed isn’t the underlying capability but rather the supercharging of these existing concepts and the investment flows following them. The challenge lies in moving beyond exciting conversations to understanding true applications, which remains in early stages. Current applications appear in drone technology and manufacturing, though manufacturing has leveraged these capabilities for some time already.

The trust-but-verify principle becomes paramount when lives depend on system performance. McLintock emphasizes that trusting AI systems still requires verification mechanisms at every stage. Beyond verification concerns, the fundamental production levels and performance of the physical items themselves, whether drones, ships, or other platforms, must meet requirements before AI can shoulder significant responsibility. The technology enhancing a system cannot compensate for inadequate baseline performance of that system.

This perspective aligns with recent findings from the RAND Corporation, which documented numerous instances where AI systems performed well in controlled environments but failed when deployed in complex, adversarial settings. Their 2024 study on AI in military applications concluded that human-machine teaming remains essential for the foreseeable future, with AI augmenting rather than replacing human decision-making in contested environments.

Consider autonomous logistics planning, an area where AI shows genuine promise. The military moves vast quantities of personnel, equipment, and supplies across global supply chains with countless variables; transport availability, weather, threat conditions, fuel costs, and diplomatic constraints. AI systems can process these variables far faster than human planners, optimizing routes and timing. However, the final decision authority must remain with experienced logisticians who understand the operational context, can recognize when AI recommendations don’t align with ground truth, and can adapt when conditions change unexpectedly.

McLintock sees value where AI enables scale and reduces costs, particularly in manufacturing improvements that allow companies to compete effectively and thrive. He acknowledges broader societal concerns about AI while recognizing it as a key term in the current market environment. Manufacturing quality control, predictive maintenance, and production optimization represent areas where AI applications are mature, measurable, and increasingly essential. A defense contractor using AI-powered inspection systems might detect defects that human inspectors miss while processing 10x more components per hour, directly improving both quality and throughput.

The Talent Dimension: Attracting and Deploying Leadership in a Transforming Sector

Perhaps the most overlooked aspect of defense technology transformation is the human element, specifically, the challenge of attracting, developing, and deploying leadership talent in an industry undergoing rapid change. McLintock’s executive search work positions him at the intersection of talent supply and organizational demand, revealing patterns that illuminate the sector’s evolution.

The perpetual need for strong leadership becomes especially acute at earlier-stage companies seeking to bring in talent from large organizations. Success requires more than simply recruiting experienced executives; it demands careful attention to culture fit and alignment while ensuring these leaders bring capabilities that help smaller companies compete effectively, potentially even against the larger players they came from. This talent arbitrage, moving experienced executives from mature defense contractors to growth-stage startups, requires careful consideration beyond simple compensation packages.

The challenge extends beyond individual placements to sector-wide talent attraction. For decades, the most ambitious engineers and technologists gravitated toward commercial tech giants or venture-backed startups in Silicon Valley, viewing defense work as bureaucratic, slow-moving, and ethically complex. McLintock sees the current transformation as an opportunity to shift that perception. The combination of defense sector innovation and emerging technologies could change how people think about defense work entirely. Someone who might have previously targeted Silicon Valley or other commercial tech hubs might reconsider when recognizing what’s happening in Washington DC and other defense ecosystems.

This isn’t merely about recruitment marketing, it requires substantive changes in how defense organizations operate. A software engineer considering offers from a commercial tech company and a defense startup will compare not just compensation but development cycles, technology stacks, decision-making speed, and impact visibility. If the defense role involves two-year procurement cycles, legacy programming languages, and limited autonomy, no amount of mission-focused recruiting will overcome the structural disadvantages.

McLintock’s current work spans multiple technology domains, reflecting market diversity. At present, the portfolio includes work in armaments, munitions, and energetics, a sector drawing considerable attention. This diversity creates opportunities for specialized talent to find niches aligned with their expertise and interests. An engineer passionate about advanced materials might contribute to next-generation armor systems, while a software architect focused on distributed systems could revolutionize command and control networks.

The talent challenge varies significantly by company stage and type. Established defense contractors need leaders who can drive cultural transformation while maintaining their core competencies in systems integration and large-scale production. Growth-stage companies need operators who can scale manufacturing, navigate regulatory requirements, and build sustainable business models. Early-stage startups need technical visionaries who can also understand military requirements and customer dynamics.

Research from McKinsey on defense workforce trends indicates that the sector faces a demographic cliff, with experienced engineers and program managers retiring faster than junior talent developing the necessary expertise. Their 2023 analysis projected a 20-25% shortfall in critical technical roles by 2030 without significant intervention. This talent gap makes McLintock’s work in repurposing talent for earlier-stage or growth companies increasingly strategic; moving experienced leaders from mature organizations to emerging ones doesn’t just help individual companies, it multiplies the impact of scarce expertise across the ecosystem.

Sector-Specific Opportunities: Where Innovation Meets Operational Need

McLintock observes that selecting a single exciting technology becomes difficult because of their interconnected nature. However, certain sectors present particularly compelling opportunities where technology innovation, operational need, and market dynamics converge.

Armaments, munitions, and energetics have gained renewed attention as conflicts in Ukraine and the Middle East have revealed concerning inventory levels and production constraints. The United States and its allies have struggled to maintain ammunition supplies at the rates required for sustained conflict, highlighting the need for both increased production capacity and more efficient, cost-effective manufacturing processes. Technologies that can reduce production costs, improve manufacturing throughput, or create more capable systems represent high-priority opportunities.

Consider the example of advanced manufacturing techniques applied to artillery shell production. Traditional manufacturing involves multiple specialized facilities, complex logistics, and lengthy production cycles. Additive manufacturing and advanced robotics could potentially consolidate production steps, reduce material waste, and accelerate output but only if acquisition processes allow new manufacturers to enter the market and scale rapidly.

Drone technology represents another area of convergence between technological capability and operational requirement. McLintock’s assessment that future conflict increasingly involves spectrum warfare and drone warfare aligns with observations from Ukraine, where inexpensive commercial drones have proven devastatingly effective for reconnaissance, targeting, and even direct attack roles. The ability to rapidly iterate drone designs, test them in operational environments, and scale production of effective variants represents a competitive advantage that traditional acquisition processes struggle to enable.

The spectrum warfare dimension encompasses electronic warfare, cyber capabilities, and the increasingly contested electromagnetic environment. Modern military operations depend on secure communications, GPS navigation, and sophisticated sensors all vulnerable to jamming, spoofing, and cyber attack. Technologies that provide resilient communications, accurate positioning without GPS, and effective electronic countermeasures will prove essential in near-peer conflicts.

Interoperability emerges as a critical but often overlooked challenge across these domains. McLintock expresses particular enthusiasm about service alignment, noting that each military service currently uses different platforms. Better alignment would enable improved interoperability and joint service operations, encompassing everything from manufacturing to drones to software. A battlefield network that seamlessly connects Army ground forces, Marine Corps expeditionary units, Air Force aircraft, and Navy ships sharing targeting data, coordinating fires, and maintaining situational awareness requires not just compatible hardware but common protocols, shared security architectures, and aligned operational concepts.

The Path Forward: From Optimism to Implementation

McLintock maintains strong optimism about the sector’s future, grounded in the belief that current conditions will drive acquisition reform. The logic follows a clear path: desiring new technology creates questions about accelerating acquisition and helping innovative companies survive long enough to deliver what they’ve built. This demand-pull approach where operational requirements drive process reform rather than reform existing in isolation offers the most promising path forward.

The market environment presents what McLintock characterizes as favorable tailwinds: a new administration potentially open to reform, recognized near-peer threats creating urgency, and a vibrant ecosystem of companies developing relevant technologies. However, translating these favorable conditions into sustained transformation requires deliberate action across multiple dimensions.

First, acquisition reform must move beyond pilot programs and special authorities to systemic change. Programs like the Defense Innovation Unit and Army Futures Command have demonstrated alternative acquisition approaches, but they remain exceptions rather than norms. Scaling these models requires changes to regulations, training for acquisition professionals, and executive commitment to accepting the risks inherent in faster, more flexible procurement.

Second, public-private partnerships need formalization through clearer pathways, transparent criteria, and predictable timelines. Emerging companies need to understand what the military actually needs, how to demonstrate capability, and what the path from prototype to program of record looks like. Defense organizations need mechanisms to engage with multiple vendors simultaneously, test competing solutions, and make evidence-based procurement decisions.

Third, the talent challenge requires coordinated action beyond individual hiring decisions. Industry associations, educational institutions, and government organizations must collaborate to create career pathways that attract young people into defense technology roles, develop their capabilities, and retain them long enough to generate return on investment. This includes addressing security clearance backlogs, offering competitive compensation, and creating work environments where talented people can do their best work.

Fourth, the sector needs honest conversations about technology maturity and realistic assessments of capability gaps. The AI discussion provides a useful example; rather than treating every problem as solvable through artificial intelligence, the community needs rigorous analysis of where AI adds value, where it introduces unacceptable risks, and where traditional approaches remain superior. This applies equally to other hyped technologies: not every defense application needs blockchain, quantum computing doesn’t solve all encryption challenges, and directed energy weapons have real physics constraints.

McLintock points toward first-principles engineering thinking as essential in determining what problem actually needs solving rather than creating products searching for applications. While he jokes about the term “defense tech,” he acknowledges it has always existed but now represents the right moment to leverage what enterprise technology has accomplished and bring those capabilities into defense applications. The opportunity includes potential simplification across services using disparate platforms, enabling better interoperability through alignment.

Actionable Recommendations for Defense Industry Leaders

For Defense Contractors and Primes:

Start identifying technology domains where your organization lacks cutting-edge capability and cannot realistically develop it internally. Build genuine partnership models with emerging companies that go beyond traditional subcontracting, offer technical mentorship, provide access to testing facilities, and create clear pathways to program inclusion. One major aerospace contractor recently established an innovation partner program that embeds startup engineers alongside company program managers, accelerating both technology transfer and mutual understanding of requirements and constraints.

For Defense Technology Startups:

Focus relentlessly on understanding actual operational requirements rather than building technology in search of a problem. Engage directly with military end-users whenever possible, even before formal procurement processes begin. Invest early in understanding security requirements, manufacturing scalability, and compliance obligations; these non-technical challenges kill more promising technologies than technical failures. Consider that the most successful defense tech companies often hire former military operators and acquisition professionals to bridge the cultural and procedural gaps between commercial technology development and defense procurement.

For Government Acquisition Professionals:

Champion reforms within your sphere of control rather than waiting for top-down mandates. Leverage existing authorities like Other Transaction Agreements (OTAs), middle-tier acquisition, and rapid prototyping programs. Document both successes and failures rigorously to build the evidence base for broader reform. Create opportunities for vendors to demonstrate capabilities before formal procurement begins, reducing risk and improving decision quality.

For Investors and Board Members:

Adjust expectations about timeline to revenue for defense technology companies. The path from technology development to sustained defense revenue typically measures in years, not months. However, companies with genuine operational traction, strong military relationships, and clear procurement pathways represent increasingly attractive opportunities as reform momentum builds. Pressure portfolio companies to focus on production readiness and manufacturing scalability alongside technology development; the valley of death often appears between successful prototypes and scaled production rather than between idea and prototype.

For Military Leadership:

Articulate clear, stable demand signals that emerging companies can design toward. Nothing kills innovation faster than granular requirements that change every six months. Simultaneously, create protected spaces for experimentation where failure provides learning rather than career consequences. The most successful military innovation efforts combine clear outcome objectives with flexibility about technical approaches. When technologies prove valuable in pilots or experiments, commit to creating acquisition pathways that allow vendors to scale rather than leaving promising capabilities in perpetual prototype status.

Conclusion: A Sector at the Threshold

The defense technology sector stands at a pivotal moment. The combination of geopolitical pressure, technological opportunity, favorable policy environment, and capital availability creates conditions for transformation. McLintock’s perspective from the talent side of the equation reveals that this transformation extends far beyond hardware and software to encompass organizational culture, career trajectories, and the fundamental relationship between innovation and procurement.

The promise is substantial; faster technology adoption, more effective military capabilities, revitalized defense industrial base, and a sector that attracts rather than repels top talent. The challenges are equally significant: entrenched processes, risk-averse culture, competing stakeholder interests, and the inherent difficulty of changing complex systems.

Success requires action at every level from individual hiring decisions to legislative reform, from startup go-to-market strategies to prime contractor partnership models. McLintock envisions a future where diverse ecosystems and innovative technologies become visible enough to excite the broader public and attract young people into the sector, contributing in various capacities to national security needs.

That excitement must translate into sustained commitment. The tailwinds are favorable, but wind alone doesn’t sail ships; it requires skilled crews, seaworthy vessels, and clear destinations. The defense technology community has all three elements emerging. The question is whether the sector can coordinate its efforts effectively enough to capitalize on this moment before the window closes and favorable conditions shift once again.

For executives navigating this transformation, the path forward demands both boldness and pragmatism; boldness to challenge established processes and pursue new partnerships, pragmatism to focus on genuine capability gaps rather than technological fashion. Those who can balance these competing demands will not only succeed commercially but contribute meaningfully to national security in an increasingly contested era.

If you made it this far, thank you! We hope we’ve earned a subscription from you, your support helps keep us going.

The root cause isn’t the sophistication of algorithms or the pace of innovation in large language models. It’s something far more fundamental: the quality and governance of the data that fuels these systems. Or as my guest Chris LaCour put it during our conversation:

Chris is the organizer of the upcoming Put Data First conference (more on that below), and he’s spent the past year talking directly with chief data officers, chief AI officers, and digital transformation leaders. The recurring theme he hears is striking in its simplicity: organizations don’t actually know where their data is, what condition it’s in, or how to harness it effectively. In other words, the promise of AI is colliding with the messy reality of enterprise data.

This episode of Facing Disruption gave us the chance to unpack that reality - why so many leaders are struggling, what’s at stake if they don’t get it right, and how a different approach to data could unlock real AI value.

AI in Its Toddler Stage

Artificial intelligence may dominate headlines, but its maturity as a business tool is far less advanced than many assume. Chris LaCour described today’s AI as a toddler - eager, fast-moving, and sometimes prone to falling over.

“We’ve been feeding it everything we could feed it, and now we’re dealing with the repercussions,” he explained.

The image is apt: enterprises have rushed to pour data into models, but like a child experimenting without boundaries, the outputs can be erratic and difficult to predict.

The deeper problem is not AI’s raw potential, but the pace at which organizations have tried to scale it. The market has been conditioned to believe that adoption is an arms race - if you don’t move quickly, you’ll be left behind. Boards and investors are pressing for aggressive timelines, often before internal teams have a clear understanding of their data ecosystem. The result is an uncomfortable pattern: pilots that look promising, but fail to translate into sustainable, enterprise-wide outcomes.

Unpredictability compounds the issue. Large language models, the engines of today’s generative AI wave, are probabilistic systems. They don’t “know” truth; they predict likely sequences of words based on training data. That means hallucinations, inconsistency, and “catastrophic forgetfulness” are not bugs to be ironed out, but inherent risks that organizations must plan for. Yet too many companies deploy these systems as if they were deterministic.

For executives, this toddler stage raises a leadership challenge. Do you push ahead, hoping to capture early-mover advantage, or do you slow down and strengthen your footing? The answer lies in balance. Ignoring AI entirely is not an option - competitors will find efficiencies and innovations that could leave laggards behind. But treating AI as a mature, plug-and-play solution is equally risky. The most resilient organizations are those willing to embrace AI’s potential while still putting guardrails in place, starting with a brutally honest assessment of their data readiness.

The Core Problem: Data Chaos and “Data Sewage”

Every executive understands the value of data in theory. In practice, most organizations are overwhelmed by it. Leaders repeatedly echo to Chris LaCour that they don’t actually know where their data resides, who controls it, or whether it can be trusted. The situation isn’t new - companies have been grappling with information sprawl since the dawn of email - but the stakes are higher now that AI depends on data as its raw material.

Structured data - the kind neatly organized in databases - is challenging enough to manage. But it represents only a fraction of what enterprises actually generate. Up to 80% of enterprise data is unstructured, buried in PDFs, videos, chat logs, and other formats that don’t fit easily into conventional systems. This is where the term “data sewage” has begun to surface: information that exists in volume, but is so disorganized, duplicated, or misclassified that it actively hinders rather than helps.

Consider what this looks like inside a large enterprise:

• Duplicate records that cause conflicting outputs when AI models attempt to generate customer insights.

• Poor classification, where sensitive data is mislabeled or overlooked entirely, creating compliance risks.

• Fragmented ownership, with marketing, finance, and operations each using their own tools and standards.

This data chaos is why so many AI efforts fail to progress beyond proof-of-concept. Models are only as good as the data they ingest. When that data is messy, incomplete, or contradictory, the results are inevitably unreliable. Worse, companies often don’t discover these flaws until after they’ve invested heavily in pilots or vendor contracts.

For leaders, the message is clear: before AI can become a source of competitive advantage, data must be treated as a strategic asset. That means not only cleaning and consolidating what exists, but also creating governance processes that ensure new data doesn’t simply add to the sewage problem.

Short Tenures, Big Expectations for CDOs

If the enterprise data landscape is chaotic, the role of the Chief Data Officer (CDO) is one of the most thankless jobs in business. On paper, CDOs are tasked with turning sprawling information ecosystems into coherent, trustworthy assets that power AI and analytics. In reality, many enter their roles only to find the situation far worse than advertised.

Chris LaCour shared what he’s repeatedly heard from data leaders: new CDOs are often given a two-year window to “get the company’s data together.” If they succeed, they are hailed as transformation leaders. If they don’t, they are quickly replaced.

The problem is that the expectations rarely match the reality on the ground. Executives may tell themselves their organization’s data quality is strong, but once a CDO begins peeling back the layers, the truth emerges: duplicates, inconsistent metadata, and incomplete governance frameworks are everywhere. What looked like a sprint to AI readiness becomes a long-distance marathon.

This revolving-door dynamic has consequences. Short CDO tenures mean institutional memory is lost just as progress begins to take shape. Teams grow cynical after multiple “data transformations” that never seem to stick. Meanwhile, the pressure to show quick wins drives some leaders toward cosmetic fixes - deploying flashy tools without solving underlying classification or governance problems.

For boards and CEOs, this should be a wake-up call. Treating the CDO role as a short-term experiment undermines the very foundations AI initiatives depend on. Building reliable, scalable data practices is not a 24-month project; it’s an organizational capability that requires sustained investment, patience, and cross-functional commitment. The companies that get this right won’t be the ones that churn through CDOs - they’ll be the ones that empower them with time, authority, and resources to address systemic issues.

Stakeholders and the Knowledge Gap

Artificial intelligence doesn’t sit neatly in one department. Its impact cuts across every function, which means no single executive can own the strategy outright. This makes AI a uniquely complex leadership challenge: the CFO cares about cost and ROI, the CISO worries about security vulnerabilities, the Chief Risk Officer frames it as a compliance problem, and the General Counsel sees legal exposure. Meanwhile, business line leaders want speed, efficiency, and new customer value.

This fragmentation is both natural and dangerous. Without deliberate coordination, organizations risk talking past one another - chasing vendor promises in one area while underestimating risks in another.

The knowledge gap widens the problem. Few leaders outside of data or technology roles fully understand how AI systems work, or what their limitations are. For example, large language models are “token guessers,” as Chris described them. They don’t represent truth; they generate predictions based on probabilities. To a compliance officer or lawyer, that distinction is critical - yet many companies deploy these systems without fully briefing stakeholders on how outputs should be interpreted.

The result is often a mismatch of expectations: business leaders assume AI is a reliable decision-making engine, while technical teams know it is a probabilistic system prone to hallucinations. Bridging this gap requires more than technical training; it requires creating shared literacy at the leadership level.

Forward-looking organizations are addressing this by forming cross-functional AI committees where CISOs, CFOs, legal officers, and data leaders sit together to align on priorities. These groups not only surface blind spots early, but also accelerate adoption by ensuring that AI is framed as a business transformation, not just a technology project.

For executives, the key is recognizing that AI is not a siloed initiative - it is a collective endeavor. Closing the knowledge gap is just as important as closing the data gap.

The AI Bubble and Sustainability Concerns

The AI boom has triggered a flood of investment. Trillions of dollars are projected for new data centers, high-performance chips, and cloud infrastructure. Every major software vendor now advertises some “AI-powered” capability. On the surface, it looks like unstoppable momentum. But underneath, questions about sustainability are growing louder.

Despite the hype, leading AI companies like OpenAI and Anthropic are not yet profitable. Their revenue models remain unproven, even as their infrastructure costs soar. Scaling large language models requires enormous compute power, and projections suggest that achieving artificial general intelligence (AGI) could require trillions in additional investment.

The environmental and energy implications add another layer of concern. Data centers already consume around 2% of global electricity, and AI workloads are set to accelerate that demand dramatically. Some in the industry are openly discussing nuclear reactors as a future enabler of AI infrastructure - a signal of just how energy-intensive the current approach is. For executives, that means AI adoption is not only a financial risk but also a reputational one, as sustainability metrics become core to investor and customer expectations.

There is also the market psychology to consider. Just as the dot-com bubble was defined by overinvestment in unproven business models, today’s AI surge is marked by inflated expectations. Every new funding round or press release seems to imply that AI will solve every business problem. When results inevitably fall short, disillusionment follows. Gartner’s hype cycle describes this pattern as the “trough of disillusionment” - and AI is heading there fast.

This doesn’t mean AI is a fad. As I mentioned during the conversation, “AI is an innovation engine, but it’s not invention.” It is extraordinarily powerful at recombining information, but it cannot create fundamentally new knowledge. For leaders, this means adjusting expectations: AI can drive efficiency and enable faster iteration, but it is not a silver bullet.

The bubble question isn’t whether AI has long-term value - it does. The real question is which companies will emerge from this phase with sustainable models, and which will collapse under the weight of overinvestment and unmet promises.

Practical Steps for Organizations

For all the uncertainty around AI, one truth is clear: success will not come from chasing the latest tool, but from building a resilient foundation. Chris LaCour emphasized this repeatedly, reminding us that “AI governance is becoming a top priority.” Organizations that rush to deploy without addressing fundamentals will find themselves stuck in endless pilots, or worse - facing regulatory and reputational fallout.

So what does a practical path forward look like?

1. Form cross-functional AI committees
AI is not an IT project; it is an enterprise transformation. Companies seeing traction are bringing together data, legal, risk, compliance, security, and business leaders in structured forums. These committees ensure AI is evaluated not only for technical feasibility but also for financial impact, ethical use, and regulatory exposure.

2. Start small with high-ROI use cases
Instead of aiming for sweeping transformation, focus on narrow projects where data quality is manageable and ROI is measurable. Customer support automation, invoice processing, and knowledge search are common entry points.

3. Tie every AI investment to clear ROI metrics
Boards and CFOs want to know: what will this save, replace, or enable? Map investments directly to CapEx and OpEx implications. This disciplines teams to pursue impact-driven initiatives rather than technology experiments.

4. Invest in governance and risk management
AI introduces new risk profiles, from hallucinated outputs to privacy breaches. Treat AI governance like cybersecurity: a continuous program of monitoring, controls, and education. Include policies for data classification, model usage, and accountability.

5. Build organizational literacy
Executives don’t need to become data scientists, but they do need to understand what AI is - and isn’t. Demystifying probabilistic models, explaining limitations like hallucinations, and clarifying where human oversight is required helps align expectations.

Taken together, these steps create an environment where AI can deliver real, compounding value instead of hype-driven frustration. They also prepare organizations for the next wave of regulatory scrutiny, which is certain to increase as governments grapple with the implications of AI in critical industries.

Building Community Around Data and AI

Note: While we aren’t sponsored by PutDataFirst, AJ Bubb will be facilitating roundtable discussions during the conference

If data quality and governance are the bottlenecks, and if success requires cross-functional alignment, then one truth becomes clear: leaders cannot solve these challenges in isolation. The problems are too broad, too complex, and too deeply interconnected across technology, legal, risk, and finance. What’s needed is not another vendor pitch or glossy panel - it’s structured dialogue among practitioners facing the same realities.

That was the motivation behind Chris LaCour’s Put Data First conference. Instead of centering on sponsor-driven keynotes, the event is designed around practitioner-led roundtables. Each participant has the chance to join multiple sessions, not as a passive listener, but as a contributor. The goal is to create a forum where executives can compare notes on the messy realities of AI - what’s working, what isn’t, and what lessons can be carried back to their organizations.

The October 27–29, 2025 gathering in Las Vegas reflects this ethos. Leaders from across industries - finance, healthcare, defense, energy - will step away from day-to-day pressures to focus on issues that can’t be solved in silos:

• How to build durable AI governance structures

• How legal, compliance, and risk functions intersect with innovation

• How to translate data strategy into tangible ROI

• How to prepare boards and investors for realistic outcomes

This intentional format addresses a gap in the current ecosystem: “There’s so much content out there already. What people want is intentional human connection and the chance to dive deeper into problems with peers.”

The lesson here isn’t only about one event. It’s about recognizing that AI maturity will come from communities of practice - not from isolated pilot projects or vendor promises. Executives who invest in building those connections will be better positioned to lead their organizations through the hype cycle and toward sustainable impact.

The Path Forward

The story of AI so far has been defined by bold ambition colliding with messy reality. Models have advanced rapidly, but most organizations remain bogged down by data chaos, governance gaps, and misaligned expectations. It’s little wonder that up to 95% of AI initiatives fail to meet their goals.

Yet the lesson isn’t to slow down or retreat. It’s to redirect focus. As Chris LaCour reminded us, “To be AI first, you must put data first.” That means confronting uncomfortable truths about the state of your data, investing in classification and governance, and giving your Chief Data Officer the mandate and support to succeed. It means equipping executives across the C-suite with enough shared literacy to close the knowledge gap. And it means resisting the temptation to treat AI as magic when it is, in fact, a tool - powerful, but only as reliable as the foundations it rests on.

Leaders who embrace this mindset will be better prepared for what comes next. They will avoid the trap of overinvestment in unsustainable initiatives, and instead build a portfolio of use cases that compound value over time. They will cultivate resilience by grounding AI in practical ROI, not hype. And they will strengthen their organizations by engaging in the kind of intentional, practitioner-driven dialogue that events like Put Data First are pioneering.

The path forward is not about chasing every new algorithm or racing to the biggest compute cluster. It’s about doing the unglamorous work of putting your data house in order, aligning your leadership team, and building communities of practice that can sustain progress. Those who take that path won’t just survive the trough of disillusionment - they’ll emerge from it stronger, with AI that truly serves their business and their customers.

I’d like to thank Chris LaCour for joining me on Facing Disruption and sharing his perspective on why putting data first is the foundation for meaningful AI adoption.

If you’re interested in continuing this conversation, the Put Data First Conference takes place October 27–29, 2025 at Planet Hollywood in Las Vegas. The event is built around practitioner-led roundtables that bring executives together to tackle the toughest questions around AI, data governance, and organizational readiness.

You can learn more and explore the agenda at www.putdatafirst.com.

This challenge was at the heart of a rich discussion on “The Future of” webcast, where I sat down with Dr. Gareth Sessel, Chief Growth & Product Officer at innovahealth.ai. Dr. Sessel, an Oxford-trained physician-engineer, brings hands-on experience at the intersection of digital biomarkers, clinical innovation, and health AI. Together, we explored why now is a turning point for precision healthcare, the practical tools emerging to bridge persistent care gaps, and what industry leaders must know to keep up with this accelerating future.

Why Healthcare Must Move Beyond “One-Size-Fits-All”

Historically, most of medicine has relied on broad protocols and population averages, delivering care that is “evidence-based” but insufficiently tailored to the individual. As Dr. Sessel pointed out, “Your doctor will treat you according to these guidelines or these protocols... but essentially they’re still designed around a population average rather than the actual individual sitting in front of you.” This generalized approach often leads to trial-and-error cycles that are both inefficient and, in some cases, risky.

Consider chronic disease management: In diabetes or depression, the standard practice involves progressing through a first-line, second-line, and third-line treatment sequence. My own multi-year journey with Long COVID mirrored the experiences of many, mired in subjective interviews and generic interventions - a frustrating process that clearly illustrates the drawbacks of healthcare’s rigid playbooks. This is supported by research in The New England Journal of Medicine, which noted that many drugs are effective for less than 60% of patients, underscoring the urgent need for more personalized approaches.

The Rise of Digital Biomarkers: Practical, Passive, and Scalable

Precision medicine’s initial promise centered on genetic and biochemical tests, but obstacles of cost, episodic measurement, and limited accessibility kept these tools from everyday practice. Improvements in genome sequencing—dropping from $3 billion and 13 years to under $200 in a single day—are remarkable, yet such approaches remain largely siloed and clinic-bound.

Digital biomarkers break down these barriers. Defined as “calculated scores” drawn from a tapestry of digital signals—including wearables, smartphones, smart home devices, and even behavioral data—digital biomarkers provide a living, breathing picture of individual health. Their defining features:

• Non-invasive and passive: No labs or clinician visits needed; everyday activities generate useful data.

• Continuous, real-world measurement: Health status is tracked in real time, reflecting real life, not the stress of a clinical setting.

• Affordable and accessible: From smart scales used in rural homes to wearables in developing countries, digital biomarkers reach where traditional medicine can’t.

A real-world narrative: Traditional monitoring of depression depends on patients’ self-reporting and infrequent check-ins, often missing subtle early warning signs. With digital biomarkers, changes in sleep quality, communication patterns, or social engagement—all easily tracked with consumer devices—can serve as sensitive indicators, prompting timely support before crises arise.

Why This Moment Is Different: Data, Compute, and Validation

The explosion of digital health data would be meaningless without the concurrent leaps in AI-driven analysis and robust data sharing platforms. As Dr. Sessel shared,

“It’s a combination of now we have compute and storage capabilities, the data sharing capabilities, and the model training capabilities that we didn’t have before.”

Key advances:

• Mass Adoption of Connected Devices: Wearables and smart devices now collect diverse physiological signals continuously, scaling far beyond the periodic “snapshot” of clinic visits.

• Deep Learning and Flexible Models: Machine learning systems can process massive streams of often noisy, real-world data, and are adaptable enough to make sense of incomplete or varying inputs.

• Interoperable Platforms and Federated Learning: The rise of secure, distributed training approaches means insights can be developed from diverse sources without centralizing sensitive data - addressing both utility and privacy.

A healthcare startup working with older adults, for example, used wearable step counters and local business transactional data to identify early declines in mobility and lifestyle. When a diner owner noticed a regular hadn’t visited in a week, it prompted a wellness check that uncovered a serious fall - an example of how subtle deviations can trigger life-saving interventions.

Closing the Loop: From Lab to Life, and Back

Digital biomarkers also pave the way for more inclusive, participatory research. Traditional clinical trials are expensive and often inaccessible to many due to travel or complexity. By gathering data passively via consumer devices - say, tracking heart rate or activity with smartwatches - studies can reach underserved populations and reflect the full diversity of real-world living.

Continuous metrics aren’t just useful for trials. They’re also transforming regular care: For example, FDA-cleared AI tools for diabetic retinopathy screening now review eye images quickly and safely, triaging only those patients who actually need to see an ophthalmologist - freeing human experts for more complex cases.

The Power - and Risk - of Proxy Data

Innovators are learning to make use of proxy signals—nonmedical data streams that offer early clues to vulnerability. Shopping behavior, online activity, or financial changes can flag cognitive decline or shifts in mental health, useful for aging populations who want to “age in place.” The nervous system, as Sessel noted, is affected by many diseases; trends in basic physiological metrics can offer fingerprints of emerging problems well before they manifest as symptoms.

Action for Executives: Seek partnerships across sectors—retailers, financial institutions, technology platforms—to responsibly incorporate new proxy data streams into predictive care models. Start small; pilot programs in high-need populations to demonstrate value and navigate early privacy concerns.

Privacy, Trust, and Data Stewardship: The Brand Imperative

Perhaps no topic looms larger for executives than trust. Digital biomarkers and the increasing breadth of personal data make privacy and data governance non-negotiable. “You don’t want sensitive information leaked out there,” said Dr. Sessel, emphasizing the growing public scrutiny around data practices.

Meaningful progress will require:

• Transparent patient consent, visible privacy engineering, robust de-identification, and interoperable controls.

• Federated learning—where models are trained locally on siloed data—allowing multiple institutions to collaborate without ever exposing raw records. Open source initiatives like ML Commons’ MedPerf and Nvidia Flare are quickly setting benchmarks in responsible health AI.

• Brand alignment with patient values: As consumer research in Harvard Business Review and Deloitte shows, organizations known for privacy leadership (such as Apple) earn more trust and, as a result, more willingness from users to share their health data.

Simply put: The ability to access and use sensitive health and behavioral data will increasingly depend on a company’s reputation for trust and ethical stewardship.

Synthetic Data: Fuel for Accelerated, Secure Innovation

Innovation in AI and healthcare shouldn’t come at the cost of real patient privacy. The emergence of synthetic data - algorithmically generated datasets that mimic real patient statistics without ever exposing true identities - solves a major bottleneck in model development and regulatory compliance. As Deloitte highlights, synthetic data now underpins many pharmaceutical R&D trials for rare and high-risk diseases, enabling robust model training and faster patient impact.

From Data Burden to Clinical Partnership

The explosion in available health data - while promising - is creating a risk of information overload for providers. Traditional dashboards often require clinicians to sift through large amounts of data manually, risking missed signals and fatigue.

The solution? Seamless, workflow-embedded recommendations powered by validated AI and digital biomarkers, with “human-in-the-loop” oversight. Dr. Sessel argued,

“A physician aided by these technologies will always outperform a physician who resists these technologies.”

The model is not to replace the expert, but to increase their productivity and reach, especially as diagnostic and treatment pathways become more granular and personalized.

Redefining Disease: From Labels to Personal Trajectories

With richer, more sensitive measurement, we are moving away from umbrella disease terms like “type 2 diabetes” and towards sub-classification, allowing targeted interventions and smarter resource allocation. Recent studies (Nature Medicine, 2018; JAMA 2023) have shown that within the “type 2 diabetes” cohort, certain subgroups are much more likely to develop specific complications; tailored care pathways—enabled by digital biomarkers and analytics—can both improve outcomes and reduce unnecessary interventions.

Five Leadership Imperatives for Executives

1. Move Beyond the Clinic: Build or partner for platforms that actively gather, integrate, and analyze real-world, patient-generated data alongside traditional health records.

2. Embed Privacy by Design: Invest in secure federated learning, clear patient consent journeys, and robust de-identification—aligning data strategy directly to brand promise.

3. Bridge Clinical and Data Science: Cross-train teams or build interdisciplinary groups that include data scientists and practicing clinicians. Encourage pragmatic validation and ongoing “human-in-the-loop” criteria for all system deployments.

4. Pilot for Outcomes, Not Just Insights: Launch tightly focused pilot programs where new models or biomarkers must demonstrate measurable benefit: improved early detection, reduced hospitalizations, or streamlined care pathways.

5. Earn and Safeguard Trust: Communicate openly about data use and benefit; involve patients in design; and regularly audit fidelity to values that matter most to your communities and customers.

Closing Perspective

Healthcare stands at a crossroads. The technical building blocks for precision medicine have arrived, but only those organizations that blend innovation with responsibility will lead. As Dr. Sessel put it,

Executives who make privacy, trust, and interdisciplinary teamwork part of their execution strategy will define the next era in health—one that is genuinely personal, equitable, and impactful.

For deeper insights, follow leaders like Dr. Gareth Sessel and connect with emerging interdisciplinary communities at innovahealth.ai and similar platforms. The future of health is not just about better data—it’s about delivering better care, with and for real people, at scale.

The Digital Twin Journey: From Hype to Strategic Asset

A decade ago, “digital twin” often felt more like a Silicon Valley buzzword than a business imperative. Today the landscape has changed. When I sat down recently with Ed Martin - formerly of Unity, Autodesk, and founder of Twin Site Consulting - we agreed: talk of digital twins is no longer about blue-sky future potential. It’s about present-day, competitive differentiation.

Ed brought decades of hands-on manufacturing, engineering, and digital thread experience, and together we explored how digital twins have evolved from 3D visualizations to dynamic, decision-driving virtual models.

A digital twin is more than a 3D model or a dashboard. It is a living, data-driven replica of a physical asset, system, or even an entire operation. Rather than simply visualizing, these twins pull real-time sensor data, business system information, and context from across the organization. They synchronize, separate (create abstraction from the asset), and synthesize - combining and analyzing disparate data into actionable insights.

The digital thread is, as Ed put it, “the virtual wiring.” It’s the infrastructure that enables the flow of data, ensuring the right context reaches the digital twin at the right moment. Think of the thread as the nervous system - and the twin as the brain.

Why Now? The AI, IoT, and Cloud Computer Surge

The acceleration of digital twin technology is fundamentally tied to three key trends: Internet of Things (IoT), artificial intelligence (AI), and cloud computing. IoT now connects billions of devices, providing a tidal wave of real-time data. AI and machine learning distill that data into patterns and predictions. Cloud platforms give us the muscle to simulate, collaborate, and analyze at scale.

According to Forbes, the digital twin market is projected to hit $110 billion by 2028, driven primarily by manufacturing and healthcare - but with use cases exploding across energy, construction, logistics, and smart cities. Gartner estimates that by 2027, 70% of businesses deploying IoT will also adopt digital twin technology. Capgemini reports an average 15% jump in operational efficiency and lower emissions among digital twin adopters, with more than half citing sustainability as a direct benefit.

Not All Digital Twins Are Created Equal

Ed and I often see “digital twin” claims that miss the point. A VR walk-through of an office, or even a 3D simulation, isn’t a digital twin unless it’s ingesting real-time operational data and genuinely abstracted from the physical system. The best digital twins do three things:

• Synchronize: They continuously pull and align external sensor and system data.

• Separate & Abstract: They operate independently from the asset, creating a safe layer for analysis and testing.

• Synthesize: They aggregate multiple sources to surface non-obvious patterns and actionable recommendations.

We agreed: accurate use of digital twin terminology matters. “Twin-washing” helps no one.

Practical Examples Across Industries

Digital twins are already proving their worth. Here’s how I’ve seen, and research confirms, real organizations deploy digital twins:

Manufacturing:
Firms with millions in daily production are using twins to reduce variability, minimize downtime, and optimize predictive maintenance. In real-world deployments, manufacturers caught early anomalies (before the failures) and cut costs through better scheduling and root-cause analysis. For example, Siemens leverages digital twins and AI-powered servers for real-time quality control, robotic path planning, and defect scrutiny - cutting both errors and energy usage.

Aerospace and Automotive:
Automakers design and test everything - from vehicle aerodynamics to autonomous systems - virtually before a single part is built. Twins allow teams to simulate road scenarios that are impossible or unsafe to try live, dramatically shortening development cycles, as in the aerospace sector where NASA first pioneered the twin concept in spaceflight safety and planning.

Healthcare:
The rise of AI-powered twins in healthcare is particularly exciting. Hospitals use them to model patient flow, resource needs, and process bottlenecks before changes are made. On the clinical side, doctors can simulate individual patient responses to treatments or surgeries, harnessing wearable, genetic, and historical data to craft hyper-personalized care plans and optimize outcomes. This has the dual effect of driving better patient results and operational efficiency.

Smart Cities and Energy:
Urban planners in technology-forward hubs like Singapore use city-scale twins to improve energy management, traffic monitoring, and disaster response. Utilities deploy them for predictive grid maintenance, optimizing crew dispatch, and preempting outages, unlocking both cost savings and societal value.

Enterprise & Service Industries:
Digital twins aren’t just for factories. Banks, insurers, and logistics companies use them to map human and automated processes for efficiency gains, workforce optimization, and customer experience transformation.

Beyond Real-Time: The Predictive Power of Twins

Real-time monitoring is valuable - but the game-changer lies in prediction and simulation. As Ed emphasized, the ability to “decouple” a digital twin from the live environment for what-if analysis lets companies test changes before risking disruption. This creates a controlled environment for AI to model scenarios, helping executive teams make confident, data-backed decisions.

Partnering generative AI with digital twins is the frontier: instead of limiting models to past data, AI can run countless “what if” scenarios inside the parameters provided by the twin. This dual approach supercharges both problem-solving and innovation, and McKinsey research notes that “AI-powered twins” help automate everything from complex scheduling to developing resilient supply chains.

Starting the Digital Twin Journey: Best Practices and Lessons Learned

Ed and I get asked all the time: “How do we begin?” The right path is both strategic and practical:

1. Define the Problem, Not the Technology. Start with a business challenge - not a checklist of features. If a twin cannot solve high-value, organizational pain points, it may not be justified. Do a rigorous gap analysis: where is your data today, and what pieces are missing?

2. Engage Stakeholders Across Silos. Digital transformation often exposes cracks in both process and culture. Bring in operators, IT, security, and business stakeholders early. You need buy-in at every level to change how work gets done.

3. Start Small but with Purpose. Initiate projects that are “big enough to matter, but small enough to win.” This validates value, tunes your approach, and creates momentum for scale-up. Think pilot, not proof-of-concept.

4. Invest Upfront in Data Quality and Security. Twins are only as good as their data. Build data foundations, ironclad security, and governance up front - you’ll avoid costly missteps later as privacy and data use regulations tighten.

5. Prioritize Education and Change Management. A sophisticated digital twin is only effective if the team knows how to use it - and why. Training and cross-functional communication are as critical as any technical feature.

6. Design for Growth. The best digital twins are modular and expandable - ready to absorb new data sources, AI tools, or changing business needs as you grow more ambitious.

The Road Ahead: Next-Gen Trends in Digital Twins

The next two years will see digital twins become smarter, more autonomous, and more sustainable. Several trends are converging:

• AI-Enhanced Twins: Machine learning is automating scenario analysis, anomaly detection, and real-time optimization - often finding root causes faster than human analysts ever could.

• Digital Twins of Organizations (DTO): Beyond assets, teams now virtualize entire business processes, surfacing hidden inefficiencies and opportunities in non-linear, “knowledge work” environments.

• Edge and Cloud Synergy: Moving twin computation to the edge enables more instant decision-making on-site, while the cloud powers deeper simulation and collaboration across the ecosystem.

• Sustainability at the Forefront: Digital twins are central to environmental goals, from energy reduction in buildings to net-zero commitments in industrial powerhouses. Over 57% of companies now cite twin-enabled sustainability as a competitive advantage.

• Immersive Interfaces: Technologies like VR, AR, and the industrial metaverse are fusing with digital twins, providing immersive, 3D windows into data - and bringing frontline operators and decision-makers together in new ways.

My Advice: Rethink Value, Rethink Scale

I’ve seen firsthand that you don’t need to be a global giant to benefit from digital twin technology. The essential shift is mindset: from seeing twins as a one-off project, to embedding them in your enterprise’s digital transformation DNA. The organizations winning with twins aren’t just saving money - they’re transforming resilience, customer value, and growth speed.

Ed’s parting advice resonates:

Think big about the outcome. Start small to prove value. Scale fast with confidence - but always stay anchored in your organization’s real needs.

If you’re considering digital twins, or are ready to scale, I’d love to hear from you. What challenges are you facing? What data is still “locked up” in organizational silos? And what could you do if you had a real-time, predictive playbook for your business?

Let’s continue the conversation.

Special thanks to Ed Martin, founder of Twinsight Consulting for sharing his expertise as guest on our live stream. For more examples, case studies, and the latest vendor solutions, reach out or drop your experiences in the comments.

Starting a business has always been an act of courage, resourcefulness, and a fair bit of uncertainty. The questions come fast: What do I need to form? How do I handle taxes? Is my business structure right? For years, the only way forward was to rely on expensive experts or muddle through forms and legalese, hoping not to make a costly mistake.

The landscape is shifting. Today, AI is opening a new front door for entrepreneurs. It’s not just about automating paperwork, but about empowering founders to ask better questions, make smarter decisions, and avoid the classic pitfalls that stall so many promising ventures.

In this episode of Facing Disruption’s: The Future of series, I sat down with Evin Wick, a tax lawyer who’s spent his career building fintech solutions for small businesses and startups. Evin’s experience spans everything from 401(k) plans for solo professionals to full-stack back office platforms. Together, we explored how AI is no longer just a buzzword but an essential tool for navigating the early stages of entrepreneurship.

We’ll walk through the core themes that emerged from our discussion:

• The three biggest hurdles for founders: knowledge, experience, and financial gaps

• How automation and AI-powered tools are shifting the paradigm

• The “traffic light” framework for when to trust AI, when to involve humans, and when to blend both

• Practical steps founders can take today to leverage AI in building their business

Let’s dive in.

The Three Gaps: Knowledge, Experience, and Finance

Every founder, whether launching a tech startup or opening a local coffee shop, runs into three big hurdles.

The Knowledge Gap

Most new business owners don’t know what they don’t know. When I started my first business, I remember staring at a stack of forms, unsure which ones mattered and which were just noise. Evin shared a story of a client who trademarked the name of his consulting firm simply because an online service suggested it, spending nearly a thousand dollars on something that added no real value. That money could have gone into customer outreach or better tools.

AI is starting to close this gap by making it easy to ask open-ended questions and get a framework for what matters. Instead of relying on someone else’s checklist, you can prompt an AI with the details of your business and get tailored guidance. For example, you might ask, “Does my business need to register a trademark?” and get a clear, contextual answer based on your industry and goals.

The Experience Gap

Even when you know what should be done, it’s hard to know when and how to do it. I’ve seen founders form the wrong type of entity because it was the easiest option online, only to discover months later they’d created unnecessary overhead. Evin recounted a time when a group formed a partnership, only to find out during dissolution that they’d never actually filed the right paperwork. They lost both time and money, but more importantly, learned the hard way that experience matters.

AI can’t replace real-world reps, but it can help simulate them. By reviewing documents, flagging mismatches (like using a Texas LLC template for a California single-member business), or surfacing relevant compliance steps, AI acts as a second set of eyes. It helps founders avoid rookie mistakes and focus on what really drives their business forward.

The Financial Gap

Expert advice is expensive, and most founders don’t have the budget to hire lawyers, accountants, and consultants for every decision. In the past, this led to either going it alone or overspending on services that weren’t actually necessary.

Today, AI-powered tools are democratizing access to foundational knowledge. You can get relatively far along with free or low-cost solutions, saving your budget for those moments when true expertise is needed. This shift is leveling the playing field, making it possible for more people to start and run businesses without breaking the bank.

The Evolution of Startup Tools: From Automation to Intelligence

We’re living through a transition from the “automation era” to the “intelligence era.” Early tools simply made it easier to fill out forms or automate basic tasks. For example, services like Stripe Atlas or LegalZoom can set up a Delaware LLC in minutes. But as Evin pointed out, this sometimes led to founders creating unnecessary entities, like opening a Delaware LLC for a local coffee shop, only to face extra fees and compliance headaches.

The new wave of tools is different. AI doesn’t just speed up the process, it helps you ask better questions and make more informed decisions. Imagine using ChatGPT or Claude to review your business plan, identify compliance requirements based on your location and goals, or even draft and review legal documents. The goal is not just to get somewhere faster, but to ensure you’re heading in the right direction.

The Traffic Light Framework: When to Trust AI, When to Bring in a Human

Evin introduced a practical framework for thinking about where AI fits in the business-building journey:

• Green Light: Tasks that are low risk, high frequency, and low consequence. For example, categorizing expenses in your accounting software. AI can handle these reliably, freeing you up for more strategic work.

• Yellow Light: Tasks that benefit from a human in the loop. Drafting contracts, preparing compliance checklists, or making decisions that have moderate consequences. Here, AI can take the first pass, but you should review and approve before moving forward.

• Red Light: High-stakes, low-frequency decisions that require human judgment. Choosing the right business structure, making major tax decisions, or signing significant contracts. AI can help with research and framing the issues, but the final call should always be made by you or a trusted expert.

For example, when I was setting up payroll for a new venture, AI helped me understand the options and draft the initial setup, but I still brought in an accountant to review everything before submitting. This blend of automation and human oversight saved time and money, while reducing the risk of costly errors.

Real-World Examples: Mistakes, Lessons, and Best Practices

The journey from idea to execution is rarely smooth, but each misstep is a chance to learn. Here are a few stories that stood out:

• Trademarking Too Soon: A founder spent nearly a thousand dollars trademarking a business name that didn’t need protection. That capital could have been invested in customer acquisition or product development.

• Entity Formation Missteps: I once filed paperwork for a partnership, only to discover months later that the business had never been properly formed. The lesson: always double-check the requirements for your state and business type, and use AI to review your documents for accuracy.

• Recurring Overhead: Evin described founders who set up Delaware LLCs because it was easy, only to pay annual fees for years without any real benefit. Dissolving an unnecessary entity is often harder and more expensive than creating it.

• Bookkeeping Automation: While AI can automate much of the bookkeeping process, it’s not truly “autopilot.” Manual review is still needed, especially for categorizing unique or complex transactions. In my own experience, having an AI-driven system flag unusual expenses made it easier to catch mistakes early.

Actionable Steps: How to Put AI to Work in Your Business Today

If you’re ready to harness AI as your front door advisor, here’s how to get started:

1. Open a Separate Business Bank Account: This is the single most important step for any new business. It keeps your finances clean, simplifies tax time, and makes it easier to track profitability. Services like Mercury or Rho make this process fast and painless.

2. Connect Your Accounting System: Use modern accounting platforms that integrate with your bank account. This automates much of the expense tracking and reporting, giving you real-time insight into your business health.

3. Leverage AI for Compliance and Planning: Use AI tools to map out your compliance needs based on your business type, location, and goals. Prompt the AI with specifics about your situation and ask for a checklist of requirements.

4. Trust, But Verify: Use AI to review documents, surface potential issues, and prepare questions for experts. But always double-check critical decisions with a qualified professional, especially when the stakes are high.

5. Ask Open-Ended Questions: Don’t be afraid to admit what you don’t know. AI is most powerful when you use it to explore possibilities and frame the right questions, not just to get quick answers.

Wrap Up: The Future of Founding Is Human Plus AI

AI isn’t here to replace founders or experts, but to make both more effective. By bridging the knowledge, experience, and financial gaps, AI is democratizing entrepreneurship and making it possible for more people to build successful businesses.

The key is to use AI as an advisor and accelerator, not as a substitute for judgment. When you combine the efficiency of automation with the wisdom of experience, you unlock new levels of clarity and confidence.

If you’re starting or scaling a business, now is the time to ask: Can AI help with this? The answer is almost always yes—just remember to bring your own expertise to the table, and don’t hesitate to call in a human when it counts.

Ready to take the next step? Open that business bank account, connect your accounting system, and start the conversation with AI today. The front door to your business has never been more accessible.

Special thanks to Evin Wick for sharing his expertise and to everyone pushing the boundaries of what’s possible for founders and small businesses today.

The Great AI Awakening: Where We've Been and Where We're Going

The phone calls started in late 2022. Every major enterprise leader suddenly wanted to know the same thing: "What does AI mean for my business?" As Jeff Sawyer, Managing Director - Digital & AI Transformation at Boston Consulting Group, puts it, we've witnessed something unprecedented in business transformation - a technology that's not just changing how we work, but fundamentally challenging what it means to be human in a world of artificial intelligence.

In our recent conversation, Jeff and I explored the three-phase evolution of enterprise AI adoption, the critical mistakes organizations are making, and why the companies that succeed will be those that master the art of human-machine collaboration rather than simply deploying the latest technology.

Jeff Sawyer brings a unique perspective to this discussion. With over a decade in consulting - first at Accenture where we worked together on IoT and digital mobility transformations in the cruise and hospitality industries, and now as a Managing Director at BCG - Jeff has been on the front lines helping Fortune 500 companies navigate digital transformation. His current focus on AI strategy and implementation gives him an unparalleled view into what's actually working (and what's failing spectacularly) in enterprise AI adoption.

Let’s dive in below!

The Unprecedented Nature of This Transformation

What makes the current AI transformation different from previous technological shifts is its scope, speed, and fundamental nature. As Jeff noted in our conversation,

"We've never before as a species invented something that's smarter than us".

This isn't just another tool—it's a technology that challenges basic assumptions about human cognitive advantage.

The implications extend beyond business transformation to questions about the future of work, education, and society. Unlike electricity or the internet, which augmented human capabilities in specific domains, AI has the potential to enhance or replace human intelligence across virtually every field of endeavor.

This creates what Jeff calls an "existential moment" for organizations and individuals. The traditional approach of gradual adaptation won't work when the technology is evolving at machine timescales rather than human ones. Success requires embracing uncertainty, investing in continuous learning, and maintaining the flexibility to pivot as new capabilities emerge.

The Three-Phase Evolution of Enterprise AI

The enterprise AI adoption story unfolds across three distinct phases, each presenting unique challenges and opportunities that mirror broader patterns in technology transformation but with unprecedented speed and scope.

Phase One: The Great Questioning (2023)

The first phase was characterized by what Jeff calls "strategy paralysis." Organizations knew they needed to respond to AI, but most conversations centered around existential questions rather than practical implementation. CEOs were asking consultants to predict an unpredictable future, while boards demanded AI strategies for technologies that were evolving faster than quarterly planning cycles.

During this phase, countless organizations fell into what I call the "silver bullet syndrome"—the same pattern we saw with IoT, AR/VR, and other emerging technologies. Leaders would ask for "the box of AI" without understanding the fundamental business problems they were trying to solve. The technology became the solution in search of a problem, rather than a tool to address clearly defined challenges.

This period was dominated by assessment projects and strategic discussions rather than implementation. Companies were essentially trying to understand the scope of the disruption ahead while grappling with a technology that was advancing at machine timescales rather than human ones.

Phase Two: The Efficiency Obsession (2024)

As we moved into 2024, organizations shifted from questioning to doing, but with a narrow focus on efficiency use cases. The promise was simple: deploy AI to cut costs and automate processes. The reality proved far more complex.

Jeff's experience with BCG's "10-20-70" framework reveals why so many efficiency-focused AI initiatives failed to deliver promised savings. While organizations invested heavily in the technology (10%) and algorithms (20%), they consistently underestimated the change management required (70%).

"You can't achieve efficiency savings without embracing fundamental change in how your organization works," Jeff explains. "But most companies charged down the path of building AI solutions while never bothering to focus on the human elements that actually drive value."

This mirrors my own experience helping organizations implement emerging technologies. During our work together at Accenture, we encountered manufacturers who wanted AR-enabled assembly instructions but had never digitized their paper-based processes. The technology wasn't the bottleneck - the lack of foundational capabilities was.

Phase Three: Strategic Transformation (2025 and Beyond)

We're now entering the third phase, where successful organizations are moving beyond efficiency theater toward genuine strategic transformation. Current data shows that 78% of global companies now use AI in their business operations, with 92% planning to increase their AI investment over the next three years. However, only 25% of companies are maximizing value from their AI investments, revealing a significant "AI Impact Gap"1.

This phase requires a fundamental shift in thinking about AI's role in business. The most successful implementations combine efficiency gains with growth opportunities while building the foundational capabilities that will support whatever AI developments emerge next.

The Portfolio Approach: Balancing Efficiency and Growth

Smart organizations are adopting what Jeff calls a "portfolio approach" to AI transformation. Rather than betting everything on a single use case, they're diversifying across multiple dimensions while building shared infrastructure that serves multiple objectives.

Efficiency Use Cases: The Foundation

While efficiency-focused AI projects face significant change management challenges, they remain important for several reasons. They provide immediate, measurable value that can fund more ambitious initiatives while helping organizations build AI capabilities and confidence. However, efficiency gains are inherently limited—you can only reduce costs so far.

The key insight is that successful efficiency implementations require embracing fundamental change in how organizations work. You cannot achieve efficiency savings without transforming underlying business processes, which is precisely where most organizations underinvest.

Growth Use Cases: The Multiplier

Growth-oriented AI applications offer theoretically unlimited upside potential. These initiatives focus on creating new revenue streams, entering adjacent markets, or fundamentally improving customer experiences. A compelling example comes from a Brazilian city that used Google's Veo3 to create a high-quality tourism commercial for $52—a project that would have cost $18,000 using traditional methods.

This represents not just cost savings but the democratization of capabilities that were previously accessible only to well-funded organizations. Such democratization enables entirely new business models and competitive dynamics across industries.

Experimental Initiatives: The Innovation Engine

Organizations also need low-risk, high-learning experiments that help them understand emerging capabilities. These projects may not deliver immediate ROI but provide crucial insights into future possibilities. The cost of experimentation has dropped dramatically—what used to require months of development can now be prototyped in days or weeks.

The challenge for executives is that there's no universal "golden ratio" for these portfolios. The right mix depends on industry context, competitive position, and strategic objectives. However, successful companies share one common trait: they treat AI transformation as a journey of continuous learning rather than a destination to reach.

The Human Element: Why Change Management Trumps Technology

Perhaps the most counterintuitive insight from our conversation is that successful AI transformation is primarily about people, not technology. Organizations that focus solely on technical implementation consistently underperform those that invest in human capabilities and change management.

Current research validates this approach. Companies that invest in change management are 1.5 times more likely to meet their AI goals than those that don't2. The most successful AI implementations require 70% of effort dedicated to change management, yet only 37% of organizations make significant investments in these activities3.

Consider the current talent landscape. We're seeing unprecedented disruption in professional services, where traditional moats of proprietary information and methodologies are being eroded by AI's democratization of knowledge. Law firms that built decades of expertise in specific jurisdictions now compete with AI systems that can access the same information instantly.

But this disruption creates opportunities for those who adapt. The most valuable professionals are becoming those who can effectively collaborate with AI systems - using technology to structure thinking, generate frameworks, and accelerate execution while applying uniquely human capabilities like contextual judgment, relationship building, and creative problem-solving.

The Speed Imperative: Why Experimentation Can't Wait

One of the most significant shifts in the current AI landscape is the collapse of traditional innovation timelines. Where innovation initiatives once required 6-12 months just to reach proof-of-concept, AI enables experimentation in days or weeks.

This creates both opportunity and pressure. Organizations can no longer hide behind lengthy development cycles or use resource constraints as excuses for inaction. The Brazilian tourism commercial example demonstrates how dramatically the economics of content creation have shifted—professional-quality output for the cost of a nice dinner.

I've experienced this acceleration personally in my content creation workflows. What previously took hours of manual work across multiple people now happens with a single button click, generating 30 pieces of content for review. The bottleneck has shifted from creation to curation, from doing to deciding.

This pattern is playing out across industries. The constraint isn't technological capability—it's organizational capacity to absorb and act on the outputs of AI-enhanced processes. Organizations must prepare for a future where decision-making speed, not just decision quality, becomes a competitive differentiator.

The Skills Revolution: Rethinking Professional Development

The rapid advancement of AI capabilities is creating unprecedented disruption in professional services and knowledge work. Traditional moats—specialized knowledge, proprietary methodologies, access to information—are being eroded by AI systems that can process vast amounts of information and generate insights at superhuman speed.

This disruption is particularly acute in consulting, legal services, and other knowledge-intensive industries. The value proposition is shifting from information access to problem definition, creative thinking, and human judgment. As Jeff noted in our conversation, AI doesn't live in the three-dimensional world that humans inhabit, which means human insight remains crucial for understanding real-world context and implications.

Organizations must invest heavily in upskilling their workforce. This isn't just about teaching people to use AI tools—it's about developing the uniquely human capabilities that complement AI systems. These include:

Strategic Thinking: The ability to define problems clearly and think through complex, multi-dimensional challenges.

Creative Problem-Solving: Moving beyond conventional approaches to explore novel solutions.

Emotional Intelligence: Understanding human motivations, concerns, and needs that AI cannot fully grasp.

Ethical Reasoning: Making decisions that consider broader implications and stakeholder impacts.

Systems Thinking: Understanding how changes in one area affect the broader organizational ecosystem.

The most successful organizations will be those that view AI as an amplifier of human capability rather than a replacement for human workers. This requires a fundamental shift in how we think about professional development and organizational design.

Building Sustainable Competitive Moats

While the AI landscape appears chaotic, Jeff identifies four foundational elements that will determine long-term competitive advantage: data, power (energy), compute, and capital1. Organizations that control these resources will have sustainable moats regardless of how specific AI technologies evolve.

Data Infrastructure: Companies with clean, accessible, well-governed data will consistently outperform those struggling with legacy systems and siloed information. This isn't just about having data—it's about having data that AI systems can effectively utilize.

Energy and Compute: The computational requirements for advanced AI are enormous and growing. Organizations with efficient access to both energy and computing resources will have fundamental advantages in deploying sophisticated AI capabilities.

Capital Allocation: While the cost of experimentation has decreased dramatically, scaling AI solutions still requires significant investment. Companies that can efficiently allocate capital across their AI portfolios will outpace those making scattered bets.

Human Capabilities: Perhaps most importantly, organizations that invest in upskilling their workforce and creating cultures of continuous learning will adapt faster to technological changes than those focused solely on technology acquisition.

Infrastructure and Data: The Hidden Foundation

While much attention focuses on AI applications and use cases, the underlying infrastructure and data architecture often determine success or failure. Many organizations discover that their legacy systems, siloed data, and outdated architectures cannot support the AI initiatives they want to pursue.

The crawl-walk-run approach becomes essential. Organizations cannot leapfrog the hard work of modernizing their technology stack, moving to cloud-based architectures, and rationalizing their enterprise data. However, AI can actually help with this modernization process, creating a virtuous cycle where improved infrastructure enables better AI implementations.

Data governance becomes particularly critical. AI systems require high-quality, well-organized data to function effectively. Organizations must address data location, privacy, security, and compliance considerations before they can fully leverage AI capabilities. This foundational work isn't glamorous, but it's essential for long-term success.

Practical Recommendations for Executive Leaders

Based on our conversation and broader industry trends, several strategic recommendations emerge for enterprise leaders:

Invest in Foundational Capabilities: Before chasing the latest AI applications, ensure your organization has modern data infrastructure, cloud capabilities, and governance frameworks. These investments will pay dividends regardless of how specific AI technologies evolve.

Prioritize People Over Technology: Dedicate at least 70% of your AI transformation effort to change management, training, and cultural adaptation. The organizations that master human-AI collaboration will consistently outperform those with superior technology but inferior adoption.

Embrace Experimentation: Create safe spaces for rapid experimentation with AI tools and applications. The cost of testing new capabilities has dropped dramatically—use this to your advantage by running many small experiments rather than a few large bets.

Think Portfolio, Not Projects: Develop a balanced portfolio of AI initiatives spanning efficiency gains, growth opportunities, quick wins, and strategic bets. Avoid the temptation to focus exclusively on any single category.

Prepare for Acceleration: Build organizational capabilities to handle the increasing pace of change. This includes decision-making processes, resource allocation mechanisms, and communication systems that can operate at AI speeds rather than traditional business timescales.

The Path Forward: Embracing Unprecedented Change

The conversation with Jeff reinforced a fundamental truth about the current moment: we're living through a transformation that has no historical precedent. The scope of change spans every industry and function, the speed of evolution exceeds human adaptation timescales, and the technology itself challenges basic assumptions about human cognitive advantage.

For executives, this creates both tremendous opportunity and existential risk. Organizations that successfully navigate this transformation will gain sustainable competitive advantages, while those that fail to adapt risk obsolescence.

The key insight is that success won't come from predicting the future or betting on specific technologies. Instead, it will come from building adaptive capabilities - in technology, in people, and in organizational culture - that can evolve with whatever changes emerge.

As Jeff concluded our conversation, "The best thing you can do, whether you're a corporate leader or an individual, is embrace it as much as you can, as fast as you can. Push your own envelopes, learn as much as you can, as fast as you can, be innovative in your approach. Fail fast, but don't stand still because the rest of the world's only going faster and the technology's going faster still."

The future belongs to organizations that can master the art of continuous transformation in an age of artificial intelligence. The question isn't whether AI will reshape your industry - it's whether you'll be leading that transformation or struggling to catch up.

Videos and cases studies mentioned in the conversation

• Genesis: Artificial Intelligence, Hope, and the Human Spirit https://g.co/kgs/qViRY9m

• 2 AI Agents talk to each other:

• Alpha Go Move 37:

• Veo3 Commercial: https://www.reddit.com/r/singularity/comments/1l2azl6/ulianopolis_city_hall_in_brazil_made_a_complete/

• CMII Event: https://www.linkedin.com/posts/louisgump_innovation-aiandmedia-activity-7336205012735324161-SLY-?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAARGbYBd05iVpLvNm4EoyPktMMayR_Ewno

The New Defense Frontier: Where Private Capital Meets National Security

When I sat down with Sam Moyer, a leading voice on private capital trends in defense, one statistic stopped me cold: global venture investments in defense tech surged to $31 billion in 2024, with U.S. firms like Anduril and European players like Helsing securing record-breaking rounds. This isn’t just about dollars—it’s a fundamental shift in how America’s defense industrial base operates.

Why Private Capital Matters Now More Than Ever

The Department of Defense (DoD) faces a dual challenge: escalating threats from peer adversaries and a defense budget that’s shrinking as a percentage of GDP. While Congress allocated $141 billion for RDT&E (research, development, test, and evaluation) in 2025, this alone can’t bridge what Sam calls the “valley of death”—the gap between prototyping and mass production.

Private capital brings more than money. Venture firms like RTX Ventures and Arlington Capital inject commercial discipline, mentorship, and networks that accelerate dual-use technologies. Take Helsing, which raised $487 million to develop AI-powered battlefield sensing systems. Their platform integrates data from drones, satellites, and ground sensors—technology that’s as valuable for disaster response as for military operations.

But alignment is critical. “Investors want ROI in 3–5 years; DoD procurement cycles often take 7–10,” Sam noted. This tension was clear when Anduril pivoted from purely defense contracts to selling its Lattice AI platform to border security agencies-a move that satisfied investors while maintaining national security impact.

Case Study: How eVAC Magnetics Became a Blueprint for Success

One breakthrough example emerged from Sumter County, South Carolina. eVAC Magnetics, a subsidiary of German firm Vacuumschmelze, secured a $94.1 million Defense Production Act grant paired with a $335 million private financing package to build the first U.S. rare earth magnet facility. Here’s why it worked:

1. Demand Signaling: A 10-year offtake agreement with General Motors guaranteed a market for magnets used in EV motors, while DoD commitments ensured defense applications.

2. Risk Mitigation: The DoD’s Title III grant de-risked the project for private lenders like BMO and MUFG Bank, who provided non-recourse financing.

3. Policy Leverage: A $111.9 million Qualifying Advanced Energy Tax Credit highlighted how clean energy incentives can dual-serve national security.

This model - government catalytic capital + private scale-up - is replicable. Sam emphasized similar successes with the Air Force’s AFWERX program, which matches venture dollars to accelerate startups like autonomous drone maker Shield AI.

Navigating the “Valley of Death”: Tactics for Leaders

1. Align Incentives with “Patient Capital”

Traditional VC’s 10-year fund cycles clash with DoD timelines. Solution? Blend capital stacks:

• Corporate Strategic Investors: Lockheed Martin’s Ventures arm co-invests in hypersonic startups, ensuring alignment with defense roadmaps.

• Sovereign Wealth Funds: Qatar’s $300 million stake in SpaceX Starlink secured priority access for Middle Eastern allies.

• Revenue-Based Financing: Companies like Epirus use future contract cash flows to secure working capital loans.

2. Master the Art of Dual-Use

Startups thriving today sell to both DoD and commercial markets. Boston Metal, which raised $120 million for green steel tech, supplies alloys to the Navy while serving manufacturers like Toyota. Sam’s advice: “Design for commercial scalability first, then adapt to defense specs.”

3. Leverage New Policy Tools

The DoD’s Office of Strategic Capital (OSC), launched in 2023, is a game-changer. Its $2.1 billion budget supports:

• Loan guarantees for critical mineral projects

• Co-investment funds with private equity firms

• Testbed Access: Startups like True Anomaly use OSC grants to trial space tech at DoD ranges.

The Road Ahead: A Call to Action

The stakes couldn’t be higher. China invests 43% of its GDP in industrial expansion versus America’s 20%. To compete, we need:

• Standardized Contracting: Expand OTAs (Other Transaction Authorities) to reduce legal overhead for startups.

• Talent Pipelines: Partner with universities like Purdue’s hypersonics program to build a skilled workforce.

• Metrics That Matter: Shift from “cost-plus” contracts to outcome-based incentives for faster fielding.

As Sam put it: “This isn’t about replacing government spending—it’s about multiplying its impact.” For executives, the message is clear: Engage early with DoD’s innovation hubs (NSIN, DIU), explore dual-use partnerships, and advocate for policies that de-risk private investment.

The future of defense isn’t just in Washington or Silicon Valley—it’s in the boardrooms and factory floors where public purpose meets private ingenuity. Let’s build it together.

AJ Bubb is a technology strategist and host of “Facing Disruption” webcast series. Sam Moyer is a Research Fellow at NDIA’s Emerging Technology Institute, where he advises on defense industrial policy.

Read the full report on Leveraging AI for Transformative Federal Acquisition

Artificial intelligence is rapidly emerging as a game-changer for federal acquisition, promising to tackle the persistent challenges of slow procurement cycles, overworked contracting officers, and the complexity of navigating regulatory environments. In our latest webcast, I sat down with Christopher Barlow from MITRE and Wilson Miles from the National Defense Industrial Association’s Emerging Technologies Institute (ETI) to unpack how AI is already making a difference-and what it will take to unlock its full potential in government contracting.

Chris brings deep expertise from MITRE’s Acquisition Innovation Center, where he’s focused on AI strategies and practical tools for transforming procurement. Wilson’s research at ETI zeroes in on the intersection of emerging tech, supply chains, and acquisition policy, making him a leading voice on modernization challenges and workforce issues.

Together, we explore the urgent need for speed in federal acquisition, spotlighting how AI can accelerate market research, automate contract drafting, and streamline compliance-while also addressing the critical barriers of cultural resistance, fragmented data, and the ever-present need for experienced human decision-makers. We also dive into the skills gap and workforce planning, discussing how agencies can upskill teams and ensure knowledge transfer as technology evolves.

Conversation Deep Dive

The integration of artificial intelligence into federal acquisition processes represents a transformative opportunity to address longstanding challenges of bureaucratic inefficiency, overworked contracting officers, and extended procurement timelines. This article explores how AI can augment human capabilities throughout the acquisition lifecycle, the real-world applications already showing promise, and the critical balance between technological advancement and human expertise.

The Acquisition Challenge: A System Under Strain

The federal acquisition system, particularly within the Department of Defense (DoD), faces mounting pressure to modernize in the face of rapid technological change and evolving threats. Traditional procurement methods-characterized by manual reviews, duplicative paperwork, multiple approval layers, and inconsistent data management-have resulted in extended acquisition timelines that undermine mission readiness.

According to the Government Accountability Office, the DoD takes an average of 309 days to award complex service contracts due to administrative bottlenecks and fragmented information systems. In some cases, as Chris Barlow noted during our webcast, the lead time for major system acquisitions can stretch to nearly two years.

"When mapping out those processes, we are seeing extended lead times," Chris explained. "When you talk to the program managers and they actually see what those lead times look like... some of them were very shocked."

This problem is compounded by workforce challenges. The DoD acquisition workforce, consisting of 157,594 members (both civilian and military personnel), is frequently described as overworked. Wilson Miles, emphasized this point:

"Contracting officers are incredibly overworked and understaffed. People often complain about how long the acquisition process takes, particularly for the Department of Defense."

The consequences extend beyond just bureaucratic frustration. Extended procurement timelines create strategic disadvantages compared to international competitors. As Barlow noted, "If you look at our adversaries and some other organizations internationally, they're able to acquire warfighting capabilities at a much quicker pace than we are."

The strain on the system also creates risk for vendors and contractors. Long timelines between contract pursuit and award create uncertainty for businesses-especially smaller, innovative companies-about whether they have the financial runway to sustain themselves through the process.

The AI Opportunity: Beyond Automation to Augmentation

Artificial intelligence offers a promising set of tools to address these challenges by shifting mundane work away from acquisition professionals to IT systems. For the purposes of this discussion, AI encompasses several technologies, including machine learning, generative AI, retrieval augmented generation, multi-modal systems, and robotic process automation.

The potential benefits are substantial. A 2023 MIT study found that using generative AI tools like ChatGPT substantially raised productivity: the average time to complete controlled writing tasks decreased by 40% while output quality rose by 18%. Applied to acquisition, AI can help deliver capabilities faster by streamlining processes and enhancing outcomes.

However, both Chris and Wilson stressed that the goal isn't merely speed, but enhanced quality. "Aside from speed to delivery, we're also looking at enhancing outcomes," Chris noted. "Leveraging AI from a background of being able to collect all of your organization's data, see what you've done in the past, analyze your previous outcomes-you should, in theory, be able to apply AI to enhance your outcome."

AI Across the Acquisition Lifecycle: Practical Applications

The acquisition lifecycle consists of several phases, each with opportunities for AI integration:

1. Needs Identification and Requirements Definition
   AI can help articulate program needs by analyzing previous contracts and identifying common patterns and language. It can map stakeholder interests and provide technically feasible starting points for requirements.

2. Market Research and Analysis
   This is often a lengthy step that AI could make more insightful by analyzing vast amounts of data to identify potential vendors, assess market trends, and predict future needs. Natural language processing can quickly extract relevant information from industry reports and databases.

3. Acquisition Strategy Development
   AI can analyze data from past program strategies and outcomes to help determine the best acquisition approach. It can also educate program managers on why a particular strategy was chosen and what's required to execute it.

4. Solicitation Creation
   AI tools can automate the development of requests for proposals (RFPs) into standardized formats, resulting in quicker solicitation times. They can develop evaluation criteria based on large datasets of previous solicitations and responses.

5. Evaluation and Source Selection
   AI can assist in evaluating proposals against predefined criteria, analyzing supplier capabilities and past performance data. It can predict the likelihood of a vendor's success based on historical data, helping to provide more informed decisions.

6. Contract Award and Negotiation
   AI tools can suggest optimal contract terms and conditions by analyzing similar contracts and outcomes. They can detect anomalies and potential fraud in contract awards by analyzing patterns and flagging suspicious activities.

7. Contract Management
   AI can continuously monitor contract performance using data analytics, alerting managers to potential issues before they become significant problems. It can generate reports on performance, compliance, and financials, reducing administrative burden.

8. Contract Closeout and Evaluation
   AI can streamline the review of contract documents to assess whether all obligations have been met and identify outstanding issues. It can capture lessons learned and best practices from closed contracts, providing valuable insights for future acquisition strategies.

Early Success Stories: Pilots Showing Promise

While widespread adoption is still pending, early movers in federal acquisition have begun to see tangible benefits from initial applications. The Defense Logistics Agency has employed AI to optimize inventory management through supply chain forecasting and demand planning. The Air Force has explored AI tools for personnel and resource management, including the recent launch of NIPRGPT, designed to assist users with correspondence, background papers, and code.

In late 2024, the Army announced a pilot program experimenting with a generative AI tool to assist with multiple acquisition activities. The Army AI Integration Center also developed CamoGPT, which is built to optimize equipment maintenance, logistics, and supply chain management.

Miles shared an example from his LinkedIn network: "Someone posted about NIPRGPT, which is an AFRL (Air Force Research Laboratory) tool, and they were saying that it helped them put together a list of questions to present to vendors who are making proposals on a topic. It's just a really small win, but these informal successes are happening."

Human in the Loop: The Critical Balance

Perhaps the most crucial aspect of AI integration into acquisition is maintaining the right balance between technological capability and human judgment. Both experts emphasized repeatedly that AI should augment human capabilities, not replace human decision-makers.

This human-in-the-loop approach is not just a philosophical preference-it's a practical necessity. As Wilson Miles noted, "In the DoD context, the level of review doesn't go away in acquisition just because there's AI. There's never not going to be a human in the loop at the end of the day."

The legal and regulatory framework reinforces this requirement. Certain functions are inherently governmental, which statutes and regulations define as tasks that must be performed by government officials. These include functions requiring discretion over governance areas such as policy decision-making, performance accountability, and execution of monetary transactions.

Chris Barlow illustrated this with a powerful example:

"If a contract officer leverages AI to build a contract, the contract becomes awarded, and upon award we realized that there were some data or IP clauses that were not included in that contract, and now the vendor owns government data-there's a huge risk to security within that system. Who's going to get in trouble? It's not going to be the AI."

The Skills Gap Paradox

One of the most thought-provoking discussions during the webcast centered on what might be called the "skills gap paradox." As AI takes over more routine tasks, there's a risk that junior professionals won't develop the foundational knowledge needed to eventually become experts.

However, he identified a significant risk in succession planning: "If we could supplement all of the junior-level work because we have the experts to refine all of that, then we don't need junior-level engineers. That's simply not true. You have a huge risk of succession planning if you are not bringing in junior people and getting them up to speed and training them and creating the opportunity for them to become those subject matter experts."

This highlights the need for a balanced approach to AI integration-one that automates routine tasks while still providing opportunities for professional development and knowledge transfer between generations.

Challenges to Adoption: Culture, Policy, and Technology

Despite the promising applications, several significant barriers stand in the way of widespread AI adoption in acquisition:

Cultural Challenges

The DoD's strong warfighter culture can make it difficult to justify spending on "back office" functions rather than weapons platforms. Additionally, there's often resistance from contracting officers who lack the time to learn new tools.

Fear of job loss and risk-aversion also contribute to cultural resistance. Approximately 50% of the DoD civilian acquisition workforce consists of individuals aged 40 and above, creating varying levels of comfort with new technologies.

Policy Challenges

While policy itself isn't necessarily a barrier-there's nothing in the Federal Acquisition Regulation (FAR) that prohibits using AI tools-there are important considerations around inherently governmental functions, classification of information, and intellectual property rights.

Current DoD policy restricts the use of national security information and controlled unclassified information (CUI) in publicly accessible AI tools. This necessitates the development of DoD-specific solutions that meet security requirements.

Technical Challenges

Data quality and availability represent significant technical hurdles. The DoD has struggled to collect and retain data about acquisition processes and program execution, and to share data across government and the private sector.

The Authority to Operate (ATO) process, which determines when new software can be installed and used on government systems, is often the longest step in deploying software. This process is particularly challenging for small and non-traditional businesses developing innovative AI solutions.

The Path Forward: Recommendations for Implementation

Based on the research and expert insights, several key recommendations emerge for successfully integrating AI into acquisition processes:

Understand Your Workflow

Organizations should identify painful parts of their process where hours are spent on mundane and repetitive tasks, map these workflows, and challenge assumptions about why particular steps exist.

"Pick a painful part of your process where hours are spent on this really mundane and repetitive task, map it, understand in your workflow why that particular step exists, and challenge it," Miles advised.

Pilot Programs with Clear Metrics

Agencies should establish pilot programs using commercial AI tools for select applications in the contracting lifecycle, with strict criteria to evaluate success and plans to scale successful tools.

"We're fully for more pilot programs," Miles stated. "The best way is for the Hill to push the department to use more commercially available AI tools. There needs to be strict criteria for evaluating success."

Focus on Process First, Then Technology

Organizations should streamline processes manually before adding AI enhancements. As Barlow emphasized, "These tools should be enhancements to systems, not problem solving inherently. If that process is not efficient to start with, throwing a tool at it is very likely to not be as successful as you're hoping it to be."

Build AI Literacy Through Training

Comprehensive training programs should be developed to ensure AI literacy across the workforce, with varying levels based on roles and responsibilities. This includes understanding both the capabilities and limitations of AI tools.

Invest in Data Infrastructure

Significant investments in infrastructure and robust data governance policies are necessary for AI adoption to succeed. This includes establishing standards for data collection, usage, and sharing.

Plan for Success, Not Just Failure

Organizations need to think beyond the pilot phase and plan for what happens if AI tools exceed expectations. As Barlow succinctly put it: "Plan for failure, but prepare for success."

Conclusion: A Human-Centered Technological Revolution

The integration of AI into federal acquisition represents not just a technological shift but a fundamental reimagining of how government procures goods and services. By automating routine tasks, enhancing decision-making, and improving outcomes, AI can help address the chronic challenges of overworked staff and extended timelines.

However, the most successful implementations will be those that recognize AI as a tool to augment human capabilities rather than replace human judgment. As our experts emphasized throughout the webcast, the goal is to free acquisition professionals to focus on strategic thinking and complex decision-making while leveraging AI to handle the routine and repetitive aspects of the process.

The path forward requires a balanced approach-one that embraces technological innovation while preserving the essential human elements of the acquisition process. With thoughtful implementation, clear metrics, and a focus on building both technical infrastructure and human capability, AI can help transform federal acquisition into a more efficient, effective, and responsive system.

As we navigate this transformation, the guiding principle should be that AI exists to serve human needs and objectives-not the other way around. By keeping humans in the loop and focusing on outcomes rather than just processes, we can harness the full potential of AI to deliver better value for taxpayers and enhanced capabilities for those who serve our nation.

A condensed version of this article was originally published on Forbes

Throughout this series, we've explored various barriers to AI adoption and strategies to overcome them. However, even with quality data, receptive users, and well-chosen use cases, AI initiatives can still falter without the right leadership support and organizational culture. In my experience as an Innovation Strategist, I've seen that sustained executive sponsorship and a culture that embraces innovation are non-negotiable elements for AI success.

In this final installment of our series, we'll explore how leadership and culture set the tone for AI success and examine strategies for fostering an environment where AI innovation can thrive.

Leadership and Culture: Setting the Tone for AI Success

In my experience as an Innovation Strategist, sustained support from top management is non-negotiable for AI success. Deloitte's research corroborates this, finding that 40% of organizations cite lack of leadership support as a top challenge in AI adoption.

Securing Executive Sponsorship

Securing an executive sponsor for AI projects is crucial. This high-level support provides visibility and prioritization for AI initiatives, access to necessary resources and funding, and a powerful voice to overcome organizational resistance. For example, at Microsoft, CEO Satya Nadella's "AI-first" vision catalyzed a company-wide transformation, leading to a 30% increase in AI-related revenue streams.

Fostering a Culture of Innovation and Experimentation

To build a culture that embraces AI, leaders should lead by example by actively engaging with AI tools and showcasing their potential. They should encourage experimentation by creating safe spaces for teams to test new AI-driven approaches without fear of failure, and recognize and reward innovation by highlighting successful AI implementations and the teams behind them.

Aligning AI with Business Strategy

To increase the sense of urgency and elevate the importance of AI initiatives, clearly articulate the AI vision by explaining how AI fits into the overall business strategy. Set measurable AI-related goals tied to specific business outcomes, and regularly communicate progress by sharing AI successes and ROI with leadership and the broader organization.

Bridging the Skills Gap

The AI talent shortage is indeed a significant challenge, as evidenced by IBM's Global AI Adoption Index 2022, which found that 34% of companies cite limited AI expertise as a barrier to adoption. However, the solution to this challenge is more nuanced than simply focusing on technical AI skills.

The Dual Nature of the AI Skills Gap

While there's a clear need for technical AI skills, we must recognize that the most valuable AI implementations often come from those with extensive experience in their roles. These domain experts understand what "great" looks like in their field, can better articulate and utilize the information generated by AI systems, and have the context to identify high-impact use cases for AI.

Strategies for Comprehensive AI Upskilling

To bridge the AI skills gap effectively, organizations should invest in upskilling their current workforce, focusing on both technical AI skills and AI literacy for domain experts. This involves creating tailored learning paths for different roles and expertise levels.

Partnering with academic institutions or AI consulting firms can leverage external expertise to design comprehensive training programs that cover both technical aspects and practical applications. Consider AI platforms that simplify implementation, looking for tools that empower domain experts to leverage AI without deep technical knowledge, focusing on user-friendly interfaces and no-code/low-code solutions.

Foster collaboration between AI specialists and domain experts by creating cross-functional teams to combine technical and domain expertise and encouraging knowledge sharing and mutual learning. This collaborative approach creates a multiplier effect where technical capabilities are enhanced by deep business understanding.

The Role of Domain Expertise in AI Success

It's crucial to recognize that domain expertise is not just complementary to AI skills – it's often the key differentiator in successful AI implementations. To leverage this, identify key domain experts within your organization, provide them with AI literacy training to understand AI capabilities and limitations, involve them in AI project planning and implementation from the outset, and create feedback loops between AI specialists and domain experts to continuously improve AI systems.

Conclusion: Leading the AI Transformation

Leadership and culture are the invisible forces that can either propel your AI initiatives forward or hold them back. By securing executive sponsorship, fostering a culture of innovation, aligning AI with business strategy, and bridging the skills gap, you create an environment where AI can flourish and drive meaningful business transformation.

As we conclude this four-part series on overcoming barriers to AI adoption, remember that successful AI implementation requires a holistic approach that addresses data challenges, human factors, use case selection, and leadership support. By tackling these barriers systematically, you can unlock the transformative potential of AI for your organization.

The journey to AI adoption may be challenging, but the rewards – enhanced efficiency, innovation, competitive advantage, and growth – make it well worth the effort. As your technology advisor, I encourage you to view AI not as a standalone technology but as a strategic capability that, when properly implemented, can fundamentally transform how your business operates and delivers value.

This concludes our four-part series on overcoming barriers to AI adoption. I hope these insights help you navigate your AI journey successfully. Remember, the goal isn't just to implement AI – it's to use AI to solve real business problems and create tangible value.

Image by Lukas from Pixabay

A condensed version of this article was originally published on Forbes

Even with quality data and a receptive workforce, AI initiatives can still fail if they're applied to the wrong problems. In my experience as an Innovation Strategist, I've seen organizations waste significant resources on flashy AI projects that delivered minimal business value, while overlooking opportunities where AI could truly transform operations.

In this third installment of our series, we'll explore how to identify and prioritize the right use cases for AI – ensuring that your investments in this powerful technology deliver meaningful returns and address real business challenges.

Identifying the Right Use Cases: The Cornerstone of Successful AI Implementation

Choosing where to apply AI can indeed make or break your initiative. As an Innovation Strategist, I've witnessed companies squander resources on flashy but low-impact AI projects while overlooking areas where AI could truly transform their operations.

The Pitfall of Technology-First Thinking

One of the biggest pitfalls I've observed is when organizations focus purely on a technology solution, developing it in isolation and then searching for a problem to solve – akin to a hammer in search of a nail. This approach often leads to misaligned solutions that fail to address real business needs or customer pain points.

The Danger of Assumption-Driven Development

Another common mistake is when organizations make assumptions about their customers' pain points (whether internal or external) and jump straight into solutioning without proper validation. This can result in developing AI solutions that miss the mark entirely, wasting valuable resources and potentially damaging trust in AI initiatives across the organization.

The Power of Working Backwards

To avoid these pitfalls, I'm a strong advocate for approaches like Amazon's "Working Backwards" methodology. This process emphasizes slowing down to validate the problem, ensuring that the problem you think needs solving is indeed the critical issue. It focuses on identifying the durable challenge - the underlying, long-term challenge that customers need solved, rather than jumping to a specific solution. The methodology also encourages solution divergence and convergence, spending time exploring multiple potential solutions before narrowing down to the most promising ones.

A Framework for Success

I recommend a framework for identifying the right AI use cases that starts with problem exploration, taking 4+ weeks to conduct deep customer research, gain insights through real-world observations, and clearly define the actual problem. This is followed by solution ideation over several weeks, brainstorming potential AI solutions, creating wireframes and prototypes, and gathering customer feedback iteratively. Finally, implementation planning ensures the solution integrates with existing workflows and validates that customers are willing and able to adopt the AI solution.

Best Practices for Identifying AI Opportunities

To effectively identify and prioritize AI use cases, conduct thorough analysis of business processes using data-driven approaches to understand where inefficiencies lie. Leverage process mining tools to objectively identify areas ripe for AI intervention, removing bias from the selection process. Develop a detailed AI roadmap with clear goals and Key Performance Indicators (KPIs) to measure success, and prioritize high-impact, feasible projects that offer significant ROI and align with your organization's capabilities.

Conclusion: Focus on Problems Worth Solving

The success of your AI initiatives depends largely on choosing the right problems to solve. By adopting a customer-centric approach, validating assumptions, and aligning AI with core business challenges, you can ensure your investments deliver meaningful value.

Remember that AI is a means to an end, not an end in itself. The goal isn't to implement AI for its own sake, but to solve real, impactful problems that drive business success. When you focus on identifying and prioritizing the right use cases, you set the foundation for AI initiatives that truly transform your organization.

In the final installment of our series, Part 4, we'll explore the critical role of leadership and culture in driving successful AI adoption across your organization. Stay tuned!

Image by Vasiliu Gabriel from Pixabay

A condensed version of this article was originally published on Forbes

While data forms the foundation of AI success, the human element often determines whether AI initiatives thrive or wither. Technology implementations don't fail because of technology – they fail because of people. In my experience as an Innovation Strategist, I've seen brilliant AI solutions gather dust because organizations overlooked the critical human factors in adoption.

In this second installment of our series, we'll explore how to ensure your team embraces AI technologies and how to manage the change process effectively. After all, AI is only valuable when it's actually used to improve how people work.

The Human Element: Adoption and Change Management

Technology is only part of the equation in successful AI implementation. The human aspect - getting your team on board and ensuring smooth adoption - is equally crucial. I've seen brilliant AI initiatives falter because of resistance to change or lack of understanding.

Aligning Technology with Business Needs

At the heart of any solution, what we're really trying to do is address a core customer challenge. It's crucial to ensure that whatever solution is created fits into our customers' existing workflows and provides positive impact. The goal should be to accelerate their ability to get their job done, rather than becoming another tool in a long chain that causes additional cognitive load.

I've seen this happen many times, especially when technology organizations begin to build solutions without tight alignment and cooperation with the business units they're solving problems for. The result? Multiple tools serving ambiguous goals, being forced on people who then push back on change.

Strategies for Successful Adoption

To overcome these challenges and ensure successful adoption, comprehensive training programs are essential. These programs should equip your team with the knowledge and skills they need to effectively use and benefit from AI tools, going beyond simple tool operation to include understanding of how AI fits into their specific roles and workflows.

Demonstrating early wins to build trust is another crucial strategy. By showing tangible results quickly, you generate enthusiasm and buy-in from stakeholders. These early successes should be highly visible and directly relevant to business goals, creating momentum for broader adoption.

Involving employees in the AI implementation process from the beginning ensures the solution meets their needs and fits their workflows. This collaborative approach transforms potential resistors into champions who feel ownership over the solution.

Aligning closely with business units ensures tight cooperation between tech teams and the business units they're serving to create solutions that truly address customer needs. This alignment should be established early and maintained throughout the project lifecycle.

Focus on workflow integration by designing AI solutions that seamlessly fit into existing workflows, enhancing rather than disrupting productivity. The best AI tools feel like a natural extension of how people already work, not a forced change in behavior.

Measuring Success and Continuous Improvement

Something that often comes too late in initiatives is consideration for how we're going to measure success. It's crucial to establish metrics for usage, performance, applicability, and impact. These metrics should be monitored consistently, allowing you to continue evolving the tools to ensure they are exceeding expectations for your customers.

Case Study: Salesforce's AI Adoption Success

Salesforce provides an excellent example of successful AI adoption. When implementing their Einstein AI features, they focused heavily on user experience and workflow integration. They designed AI features to work within existing Salesforce interfaces, minimizing the learning curve, and provided extensive training resources, including Trailhead modules specifically for AI features.

They showcased early wins, such as how AI-powered lead scoring improved sales team efficiency, and continuously gathered user feedback and iterated on features based on real-world usage. As a result, Salesforce reported that 84% of their customers were using at least one Einstein feature within a year of launch, demonstrating high adoption rates.

Conclusion: Putting People at the Center of AI

The human element is not just a consideration in AI adoption – it's the determining factor in whether your AI initiatives succeed or fail. By focusing on aligning with business needs, integrating with existing workflows, and providing comprehensive support, you can significantly increase adoption rates and realize the full value of your AI investments.

Remember that AI should serve people, not the other way around. When you design your AI initiatives with this principle in mind, you create solutions that people actually want to use because they make their work better, easier, or more fulfilling.

In Part 3 of our series, we'll explore how to identify the right use cases for AI – ensuring that you're applying this powerful technology where it can create the most value for your organization. Stay tuned!

Image by This_is_Engineering from Pixabay

A condensed version of this article was originally published on Forbes

Artificial Intelligence (AI) holds transformative potential for businesses, yet the staggering statistic that over 80% of AI projects fail underscores the challenges organizations face in realizing this promise. AI project failure rates are nearly double those of traditional IT projects, often due to misaligned expectations, poor data quality, and inadequate infrastructure.

As an Innovation Strategist, I've seen firsthand the transformative potential of AI in business. However, I've also witnessed the challenges many organizations face when implementing these solutions. In this first installment of our series, we'll focus on what I consider the foundation of successful AI initiatives: data.

The AI Adoption Landscape

The AI adoption landscape is rapidly evolving. According to McKinsey's 2023 global survey, 55% of companies report AI adoption in at least one function, up from 50% in 2022. This growth is encouraging, but it also means that nearly half of businesses are still on the sidelines.

More recent data from 2025 paints an even more optimistic picture. A survey reveals that 77% of companies are either using or exploring the use of AI in their businesses, and 83% of companies claim that AI is a top priority in their business plans. This significant increase in adoption and prioritization over the past two years demonstrates the growing recognition of AI's importance in the business world.

However, adoption doesn't always translate to success. Only 48% of digital initiatives meet or exceed business outcome targets, underscoring the urgent need for a more strategic approach to AI implementation. This statistic highlights the gap between AI adoption and realizing tangible business value.

Data: The Foundation and the Stumbling Block

Data Quality: The Achilles' Heel of AI

Poor or inconsistent data leads to unreliable AI models. This manifests in various forms including unstructured or poorly organized data, lack of appropriate meta-tagging and definition, and low accuracy percentages. As one CTO I worked with put it bluntly: "We have oceans of data, but it's more like a swamp than a clear lake."

Data Silos: The Political Quagmire

Information trapped in different departments hinders comprehensive AI initiatives. Interestingly, data silos aren't always a technology problem. Often, it's a political challenge where business lines view their data as proprietary and refuse to share access. This territorial approach to data ownership creates invisible walls within organizations, preventing the holistic view needed for truly transformative AI applications.

Data Privacy and Ethical Considerations: The Compliance Conundrum

Concerns about protection and compliance slow adoption. As AI becomes increasingly adept at combining various data sources, we're seeing situations where data scrubbed of personally identifiable information (PII) can be combined with other datasets, potentially allowing AI to back into identifying original sources or individuals. This creates a complex balancing act between leveraging data for insights and maintaining privacy and ethical standards.

Strategies to Address Data Challenges

To tackle these issues head-on, I recommend implementing robust data governance practices that establish clear ownership, quality standards, and usage policies. This includes investing in data cleaning and integration tools that can transform your data swamp into a clear lake of valuable information. Establishing clear data collection and quality assurance processes ensures that your AI models have reliable inputs from the start.

Remember, high-quality data is a primary source of competitive advantage in the AI world. Organizations that treat their data as a strategic asset gain a significant edge over competitors still struggling with fragmented, low-quality information.

Case Study: Federated Learning in Life Sciences

I've worked extensively with research teams in the life sciences space, specifically around developing digital biomarkers. A recurring challenge was the intentional withholding of specific datasets between research groups, hindering the development of new algorithms and digital biomarkers.

To combat this, we explored Federated Learning. This innovative approach allows organizations to maintain control and protect their data while providing external teams the ability to run models against the data and receive results without compromising data privacy. Federated Learning has shown promising results in healthcare, with studies finding that federated learning models among 10 institutions can have as much as 99% of the model quality compared to similar research with centralized data sharing.

Conclusion: Building Your Data Foundation

Data is the foundation upon which successful AI initiatives are built. Without high-quality, accessible, and ethically managed data, even the most sophisticated AI models will falter. As you embark on your AI journey, prioritize addressing these data challenges first.

In the next article in this series, we'll explore the human element of AI adoption – how to ensure your team embraces these new technologies and how to manage the change process effectively. Remember that successful AI implementation isn't just about the technology; it's about creating an ecosystem where data flows freely, securely, and accurately throughout your organization.

Stay tuned for Part 2 of our series, where we'll dive into the human element of AI adoption and change management strategies that ensure your AI initiatives succeed with your most important asset – your people.

Image by Tung Nguyen from Pixabay

The Evolution of Human-Machine Communication

The conversation began with Giuseppe sharing his unique background in theoretical physics and software development, where he initially used machine learning to simulate the collective behavior of particles at the nanoscale. This foundation in complex computational modeling eventually led him to Tonal, where he now applies AI to more customer-facing applications.

What makes Giuseppe's perspective particularly valuable is his experience on both sides of the technological equation – from the deep technical aspects of building and operating machine learning models to focusing on how everyday customers interact with that technology in practical applications.

Breaking Down Communication Barriers

One of the central themes we explored was how technology is evolving to overcome the traditional barriers in human-machine communication. As Giuseppe pointed out, despite the revolutionary advancements in AI, we haven't yet reached the point where these technologies are seamlessly accessible to everyone:

"Are we already at a point that we can promote AI, use AI, is AI accessible not only for tech people? Are we designing AI in a way that every person, every segment of the population can use AI? And the answer is not yet."

However, we're witnessing a significant shift as AI becomes more seamlessly integrated with our environment through wearables, glasses, and sensors that enhance human capabilities and break down these barriers.

Exciting Examples of Next-Generation Interfaces

Giuseppe highlighted several cutting-edge technologies that are revolutionizing human-machine interaction:

Smart Glasses: Projects like Meta's Ray-Ban glasses and their newer holographic glasses that are context-aware, can track eye gaze, and provide real-time information about what you're seeing.

Advanced Wristbands: Using surface electromyography (SEMG) to measure muscle activation through the skin, these devices can detect subtle hand movements to control interfaces with simple gestures.

Enhanced Earbuds: Apple's AirPods Pro already contain accelerometers and gyroscopes that can monitor posture, and patents suggest future versions might read electromagnetic waves from the brain to monitor brain, muscle, and heart activity.

These technologies are dramatically increasing the bandwidth of communication between humans and machines. While traditional interfaces like typing or speaking are limited to just a few words per second, visual interfaces and gesture controls enable much faster information exchange.

AI at Tonal: Personalized Fitness Training

Giuseppe provided fascinating insights into how Tonal is implementing these concepts. Rather than using traditional weights, Tonal employs electromagnetic resistance to create a compact home gym with an AI personal trainer. The system uses a network of sensors, smartwatch connectivity, and computer vision to:

• Measure range of motion, speed, and strength at specific points

• Provide feedback on posture and form

• Adjust weight automatically when struggling

• Create personalized progression toward fitness goals

This creates a fully immersive environment where the machine can communicate with the user through visual and audio cues without causing cognitive overload during workouts.

The Three Pillars of AI Evolution

Our discussion revealed three key pillars in the evolution of AI systems:

1. Predictive AI: Collecting more signals than users are consciously aware they're putting out

2. Reactive AI: Analyzing that information to make decisions and implement actions

3. Proactive AI: Shaping user behavior by providing guidance and feedback

As Giuseppe explained, these capabilities operate across different time scales – from real-time safety features that spot when someone is struggling with a weight to long-term progression planning that adapts to help users reach their goals.

Ethical Considerations and Guardrails

No discussion about AI would be complete without addressing ethical considerations. Giuseppe emphasized the importance of implementing proper guardrails, especially when AI systems are used in health applications. General-purpose models trained on the entire web can't simply be deployed for health predictions without careful controls.

He shared a simple but illustrative example of bias amplification: when you ask AI systems to generate an image of a watch, they almost always show the time as 10:10. This happens because watchmakers historically used this time in promotional materials as it creates a more aesthetically pleasing image. While this example is harmless, it demonstrates how AI can amplify existing biases in training data – a much more serious concern when those biases relate to race, gender, or other sensitive attributes.

Giuseppe advocated for open-source models that allow developers to see what's behind the algorithms and understand their chain of thought. He also highlighted the importance of emerging regulations like the EU's AI Act and California's AI regulations in establishing boundaries for AI applications.

Preparing for the Next Generation of Interfaces

For businesses looking to prepare for this new era of human-machine interfaces, Giuseppe offered several recommendations:

• Invest in R&D for AI-driven multimodal interaction

• Explore emerging technologies for voice, gesture, and gaze tracking

• Develop adaptive interfaces that are context-aware

• Form strategic partnerships with AI hardware startups and research institutions

• Create inclusive designs that work for all users, not just tech-savvy early adopters

• Prioritize privacy, security, and trustworthiness

• Stay ahead of regulatory changes

Final Thoughts

This conversation with Giuseppe provided a compelling glimpse into how the relationship between humans and machines is evolving. As interfaces become more intuitive and seamless, we're moving toward a world where technology enhances our capabilities without requiring conscious effort to engage with it.

However, this future also demands careful consideration of ethical implications and inclusive design principles to ensure these powerful technologies benefit everyone. As Giuseppe put it, we need to make these interfaces both easy to use and accurate in their functionality.

I'm excited to continue this conversation in future episodes. In the meantime, I encourage you to watch the full interview for deeper insights into how human-machine interactions are transforming our world.

If you found this article valuable, please consider watching the full video, where Giuseppe and I explore these topics in much greater detail. Don't forget to subscribe to our channel for more discussions on emerging technologies and their impact on our future.

Leave a comment

The Changing Landscape of HR

The demands on HR organizations have never been greater. In my conversations with CHROs and HR leaders, I've consistently heard about the challenges they face:

1. Attracting and retaining top talent in a cost-constrained environment

2. Addressing the rising demand for technical skills amid a talent shortage

3. Implementing organization-wide upskilling and reskilling initiatives

4. Continually redesigning roles, teams, and ways of working

5. Deepening focus on diversity, equity, and inclusion (DE&I)

6. Driving behavioral change and building trust in human-machine collaboration

These challenges are not just HR problems; they're business problems that require a strategic, data-driven approach. The BCG report on how Human Resources can unlock the power of AI highlights, HR is at an inflection point where the focus is on turning AI's potential into real value for the organization at large.

Leveraging Data, Technology, and AI

Forward-thinking HR leaders are harnessing the power of data, advanced analytics, and artificial intelligence to drive value creation. Let's dive deeper into some key applications:

AI-Powered Talent Acquisition

A double edged sword

AI and machine learning are revolutionizing talent acquisition, but this technological advancement has created a complex dynamic between candidates and recruiters. On one side, AI systems can analyze vast amounts of data to identify candidates with the right skills and cultural fit, potentially eliminating bias and promoting diversity. However, this has led to an arms race of sorts in the hiring process.

Candidates are increasingly using AI tools to tailor their resumes and applications to specific job descriptions, creating hyper-specialized documents designed to pass through AI screening systems. Simultaneously, HR departments are deploying ever more sophisticated AI filters to sift through these optimized applications, attempting to identify the truly best-fit candidates. This back-and-forth has created a challenging landscape where the line between genuine qualification and AI-enhanced presentation has become blurred.

As a result, the optimal use cases for AI in talent acquisition are still evolving. While AI can certainly streamline processes and uncover hidden talent, there's a growing recognition that human judgment remains crucial in the final stages of candidate selection. Organizations are now grappling with how to strike the right balance between leveraging AI's analytical power and maintaining the human touch necessary for effective hiring decisions.

Predictive Analytics for Workforce Planning

Predictive analytics in HR has evolved beyond simple turnover predictions to become a powerful tool for proactive workforce management. Advanced analytics can now identify intricate patterns in employee behavior, performance, and engagement, allowing HR to predict which employees are at risk of leaving and understand the underlying reasons. This deeper insight enables HR to implement targeted retention strategies and develop personalized career paths for high-potential employees.

However, the challenge of placing employees in roles that maximize their skill sets and value to the organization persists. Traditional linear career paths often fall short, leading to mismatched placements that result in employee dissatisfaction or underperformance. Amazon's approach offers an innovative solution to this problem. By evaluating new hires on two dimensions - cultural fit and role-specific fit - Amazon creates a foundation of trust in an employee's abilities and future potential. This trust then allows for more flexible internal mobility, with managers and colleagues working collaboratively to find the right role for each individual within the company.

This approach to workforce planning and talent mobility not only enhances employee satisfaction and performance but also unlocks hidden potential within the organization. By leveraging predictive analytics and adopting more flexible approaches to career development, HR can play a crucial role in creating a more dynamic and effective workforce that drives business growth.

Intelligent Automation for Administrative Tasks

The automation of routine HR tasks is not new, but the level of intelligence in these automations is rapidly increasing. AI-powered chatbots can now handle complex employee queries, freeing up HR professionals to focus on more strategic work. According to the BCG report, some organizations are seeing up to 90% efficiency boosts for certain administrative workflows

Advanced HCM Platforms for Enhanced Employee Experience

Modern Human Capital Management (HCM) platforms are leveraging AI to create personalized, consumer-grade experiences for employees. These platforms can recommend learning opportunities, suggest career moves, and even provide personalized wellness recommendations based on an employee's unique profile and preferences.

The impact of these technologies is significant. As Accenture’s report, The CHRO’s Role as a Growth Executive, notes, organizations that effectively connect data, technology, and people stand to gain a premium of up to 11% on top-line productivity

The Rise of the "High-Res" CHRO

The Accenture report introduces the concept of the "High-Res" CHRO, a new breed of HR leader who possesses an advanced skillset that combines deep HR expertise with strong business and financial acumen, systems thinking, and technology fluency

These CHROs are distinguished by their ability to:

1. Access and create talent in innovative ways

2. Connect new dimensions of data, technology, and people to unlock potential

3. Lead reinvention beyond the HR function

What sets these leaders apart is their ability to operate as true business partners, with strong relationships across the C-suite. They're not just HR experts; they're business strategists who understand how people strategies drive business outcomes.

Key Areas of Value Creation

Strategic Workforce Planning and Skill Development

In today's rapidly evolving business landscape, strategic workforce planning has become a critical value driver. High-performing HR functions are taking a data-driven, future-focused approach to talent. They're using AI-powered skills intelligence to map current capabilities against future needs, developing personalized learning and development pathways at scale, and implementing internal talent marketplaces to optimize skill deployment.

For example, Unilever has implemented a "Future Fit" program that uses AI to help employees assess their skills and suggest personalized development plans. This initiative has resulted in over 200,000 employees actively engaging in upskilling and reskilling activities

Employee Experience and Engagement

Leading HR functions are reimagining the employee experience, recognizing that engaged employees drive better business outcomes. They're deploying advanced HCM platforms to create consumer-grade employee experiences, using sentiment analysis and real-time feedback tools to continuously improve engagement, and leveraging behavioral science to design more effective interventions.

Airbnb, for instance, has implemented a "belong anywhere" philosophy that extends to its employee experience. The company uses advanced analytics to personalize benefits and development opportunities, resulting in a 90% employee satisfaction rate

Culture and Organizational Effectiveness

High-performing HR functions are playing a central role in shaping organizational culture and effectiveness. They're using network analysis and other advanced tools to optimize collaboration and information flow, leveraging data to identify and nurture key drivers of innovation and agility, and developing new approaches to performance management and rewards that align with evolving business needs.

Microsoft's HR team, for example, has been instrumental in driving the company's cultural transformation under CEO Satya Nadella. By leveraging data and AI, they've been able to measure and enhance collaboration, innovation, and employee growth mindset across the organization

Challenges and Considerations

While the potential for HR to drive value creation is immense, realizing this potential is not without challenges. Some key considerations include:

1. Data quality and integration issues

2. Ethical use of AI and data

3. Skill development within HR teams

4. Change management across the organization

Addressing these challenges requires a strategic approach and investment in foundational capabilities. As the BCG report suggests, organizations need to invest in data readiness and GenAI solutions in parallel

The Path Forward

For organizations looking to unlock the full value-creation potential of HR, several key steps are crucial:

1. Invest in foundational data and technology capabilities, including cloud-based HCM platforms and advanced analytics tools.

2. Develop a clear vision for HR's role in driving business value, aligned with overall corporate strategy.

3. Upskill HR teams and leaders, focusing on critical capabilities like data literacy, business acumen, and design thinking.

4. Foster strong partnerships across the C-suite, positioning HR as a strategic advisor on all people-related aspects of the business.

5. Implement robust governance frameworks to ensure ethical use of data and AI in HR processes.

6. Take an agile, iterative approach to transformation, starting with high-impact use cases and scaling based on learnings.

Conclusion

The future of HR is one of immense potential. By leveraging data, technology, and a more strategic mindset, HR has the opportunity to drive unprecedented value creation. However, realizing this potential requires bold leadership, significant investment, and a willingness to fundamentally reimagine the role of HR within the organization.

As Francine Katsoudas, Chief People, Policy and Purpose Officer at Cisco, puts it: "Today, all of us across the C-suite are trying to bust siloes and spot the 'gray spaces' where opportunities or challenges lie. Bringing our unique perspectives on the data enables us to solve issues or seize opportunities more rapidly."

For those organizations that get it right, the rewards - in terms of business performance, innovation, and talent advantage - promise to be substantial. The transformation of HR from a cost center to a value creation powerhouse is not just a possibility; it's an imperative for organizations looking to thrive in the digital age.

The Growing Security Threat of IoT

The Internet of Things (IoT) has ushered in an era of unprecedented connectivity and convenience, but it has also opened up a Pandora's box of security vulnerabilities. As demonstrated by the recent Jeep Cherokee hack, the risks associated with unsecured IoT devices are no longer theoretical – they have become tangible and potentially life-threatening.

Vulnerabilities Across the IoT Stack

IoT security challenges span multiple layers of the technology stack:

Device Level: Many IoT devices lack basic security features like strong authentication or encryption. Default passwords and outdated firmware make them easy targets.

Network Level: Insecure communication protocols and lack of network segmentation allow attackers to move laterally once they gain access.

Application Level: Poorly secured APIs and inadequate access controls create openings for data breaches and unauthorized control.

Data Level: Insufficient encryption and improper data handling practices put sensitive information at risk.

Key Security Concerns for Enterprises

1. Data Privacy: IoT devices collect vast amounts of personal and operational data, making them attractive targets for cybercriminals.

2. Device Hijacking: Compromised devices can be used to launch DDoS attacks or as entry points into corporate networks.

3. Supply Chain Risks: Vulnerabilities in third-party components or software libraries can introduce hidden backdoors.

4. Regulatory Compliance: Failure to secure IoT deployments can lead to violations of data protection regulations like GDPR.

Addressing IoT Security Challenges

To mitigate these risks, organizations must adopt a security-first approach to IoT:

1. Implement Zero Trust: Treat every device and connection as potentially compromised, requiring continuous authentication and authorization.

2. Secure by Design: Build security features into IoT devices and applications from the ground up, rather than as an afterthought.

3. Network Segmentation: Isolate IoT devices on separate network segments to limit the potential impact of a breach.

4. Regular Updates: Establish processes for timely patching and firmware updates across all deployed devices.

5. Encryption: Use strong encryption for data in transit and at rest, especially for sensitive information.

6. Security Audits: Conduct regular security assessments of IoT deployments to identify and address vulnerabilities.

The Imperative for Proactive Security

The Jeep Cherokee hack serves as a stark reminder that IoT security can no longer be an afterthought. As we continue to connect more devices and systems to the internet, the potential attack surface grows exponentially. From smart home devices to industrial control systems, every unsecured IoT endpoint represents a potential entry point for malicious actors.

Organizations must recognize that IoT security is not just about protecting data – it's about safeguarding physical assets, critical infrastructure, and even human lives. As we've seen with compromised webcams and vulnerable medical devices, the consequences of lax IoT security can extend far beyond the digital realm.

Conclusion

The ugly side of IoT is its potential to amplify existing cybersecurity threats and create entirely new categories of risk. However, by adopting a proactive, comprehensive approach to security, organizations can harness the transformative power of IoT while mitigating its inherent dangers. As we move forward in this connected era, security must be woven into the very fabric of our IoT ecosystems – from the smallest sensor to the largest data center.

The time for complacency has passed. Whether you're just starting an IoT project or managing an existing deployment, a thorough security audit is not just advisable – it's essential. The future of IoT depends on our ability to build trust through robust security practices, ensuring that the benefits of this technology can be realized without compromising safety or privacy.

Original Article

A few weeks ago, we watched as two hackers took control of a Jeep Cherokee remotely through the wireless info-tainment center. Not only could they control the radio, but the door locks, steering, brakes, and practically every other system fell powerless under their commands.

Between connected devices, and automation systems which depend on accurate data, there are numerous points of vulnerability where a malicious attack could take place.

IoT presents a unique challenge, with multiple standards in all layers of the IoT stack, simply addressing this issue in a single layer, say gateway to gateway communication (think MQTT), there is still the possibility of attacking other layers, say sensor to gateway communication, which many times rely on simple electrical signals communication with an edge device.

In this article from Windriver , we are taken through various layers of the IoT architecture, and presented with the different key concerns facing enterprises (and all companies implementing connected solutions), as well as some of the way’s these vulnerabilities are being addressed.

The bottom line is, security should be on the forefront of any IoT project plan. Even in a proof of concept, critical information can be leaked, whether it is personally identifiable information, protected information, or even proprietary data (re: mixing procedures during chemical fusion processes).

The Jeep Cherokee hack won’t be the last, and it definitely wasn’t the first, just do a Google search for unsecured web-cams and you’ll get a taste for just how lax we are with securing our devices. If your company is embarking on an IoT project, or already has one in place, it’s most likely time for a good physical and network security audit.