The deeper potential took a decisive step forward with the announcement from Moderna and Merck that their Phase 3 trial of intismeran autogene, an individualized mRNA cancer therapy, met its primary endpoints in patients with high-risk melanoma. Administered in combination with Merck’s immunotherapy PD-1 checkpoint inhibitor Keytruda, the treatment demonstrated statistically significant reductions in both cancer recurrence (Recurrence-Free Survival) and distant metastasis (Distant Metastasis-Free Survival). While peer-reviewed publication and formal regulatory reviews remain ahead, the clinical milestone is substantial.

This is not an incremental enhancement to standard oncology; it represents a structural shift in how therapeutics are conceived, designed, and manufactured.

Overcoming The Historic Population-Level Compromise

Cancer has never been a monolithic, population-level disease. Every malignant tumor carries a mutational profile unique to an individual patient. But, the sheer combinatorial complexity of human genomics made individualized molecular tailoring impossible. Medicine was forced to accept a pragmatic compromise: standardized therapies formulated for population averages. They worked for some, failed others, and inflicted debilitating side effects along the way. We accepted this as the inevitable trade-off between what medicine could aspire to and what it could actually do.

The convergence of high-throughput sequencing and predictive machine learning dissolves this historic constraint. Here is what intismeran actually does. When a patient undergoes surgery to remove a melanoma tumor, scientists sequence its DNA and compare it against the patient’s healthy tissue. That comparison isolates the precise mutations driving the cancer. A suite of five machine learning algorithms then run autonomously and sequentially across this dataset. With no human intervention, these models scan and prioritize up to 34 neoantigens: abnormal surface proteins expressed exclusively on that patient’s tumor cells. The algorithms select the targets most likely to trigger the strongest immune response and encode that selection directly into a custom mRNA sequence.

No human designer intervenes in the process between raw genomic data and drug design.

Once injected, the patient’s own cells translate the synthetic mRNA sequence into harmless replicas of the tumor’s mutated proteins. The immune system recognizes these as foreign, and trains its T-cells (the body’s primary cancer-fighting cells) to hunt down and destroy any cells carrying those exact mutations.

However, tumor microenvironments frequently evade destruction by exploiting immune checkpoint pathways—effectively disabling incoming T-cells before they can mount an attack. This is where combination therapy becomes essential. Keytruda blocks the PD-1 pathway on T-cells, preventing the tumor from suppressing immune activity. Intismeran supplies the precise immunological target; Keytruda removes the regulatory brake that shields the tumor from destruction.

The Orchestration Architecture: Mass Individualization at Scale

Beyond molecular design, the operational hurdle of personalized medicine is manufacturing orchestration. Producing a customized, single-patient lot within a global multi-center clinical trial presents extraordinary supply chain and scheduling complexity.

To execute this, Moderna engineered an automated orchestration engine called Maestro. The system coordinates the seven sequential manufacturing stages required for each bespoke batch in real time. If a clinical site adjusts a patient’s biopsy or infusion timeline, the platform dynamically reschedules that individual’s production sequence while simultaneously rebalancing capacity and throughput across every other active patient in the global pipeline. From tissue biopsy to formulated vial, Moderna has reduced the end-to-end design and manufacturing cycle to approximately six weeks.

As my readers and students know, I have long described AI as a technology of discovery, one akin to a microscope. The optical microscope did not merely sharpen human vision; it revealed an invisible cellular universe that founded modern bacteriology and medicine.

Applied AI operates under the same mechanism: it is an instrument of discovery that reveals pattern and order within combinatorial complexity far beyond human cognitive capacity.

Why This Time Is Different

I have written about AI breakthroughs before: protein folding and math proofs, for instance. They are genuinely important. But they are also, for most people, somewhat abstract. Few of us have an intuitive feel for what it means to predict the three-dimensional structure of a protein.

Melanoma is visceral. More than two percent of the population will confront it during their lifetime. Most people reading this know someone who has faced it. When caught early, it is very treatable. When it spreads, the five-year survival rate can fall below 20 percent. The gap between those two outcomes, and the fear that lives in that gap, is something most of us understand without needing an explanation.

That is why this breakthrough matters beyond its medical significance. The clinical breakthrough in intismeran grounds the value proposition of artificial intelligence in an immediate, human reality. It makes the promise of AI concrete and human in a way that few prior announcements have managed.

And it signals something even more important: if this works for melanoma, the platform works. The same mRNA technology, the same algorithmic pipeline, the same organizational capabilities that Moderna has built are already being tested across lung cancer, bladder cancer, renal cell carcinoma, and other tumor types. Melanoma is not the destination. It is the proof of concept.

What Comes Next: Reframing The Value Proposition Of AI

Global investment in computational infrastructure for AI is often scrutinized through the lens of incremental cost savings and workflow automation. Yet the ultimate economic and societal return on these investments will not be measured by automated back-office tasks. It will be determined by whether organizations can harness these discovery engines to solve structural problems previously deemed intractable.

That promise is no longer hypothetical. Moderna’s achievement demonstrates that this transition is underway. The question this raises for business leaders, researchers and academic institutions, and policy-makers alike is what we aim at next.

The strategic imperative for institutional leaders and researchers is to look past superficial application layers and examine the fundamental organizational and architectural capabilities required to build discovery-driven operating models.

And in my next post, I will explore what it actually took for Moderna to get here, because the answer has profound implications for every organization trying to harness AI for something that genuinely matters.

Of all the dimensions of the Transformation Flywheel, the one I am asked about most is Know-how and IP: the proprietary knowledge that makes competitive advantage defensible and compounding. But to understand how that know-how is built, we first need to answer a more fundamental question: what is AI actually for?

As an academic who has spent four decades studying technological change, I am seeing something in my own research that I believe many business leaders are missing. I’m now able to go from a preliminary idea to hypothesis generation, data collection, analysis, and a reasonable first draft of a research paper within hours rather than years. The pace of discovery is accelerating across domains in ways that are genuinely unprecedented. And the organizations that grasp what this means, not just for science but for business, will be the ones that define the next era.

The Real Promise

AI’s real promise is not efficiency. It is discovery.

It is the possibility of solving problems that have eluded humanity for generations. Treatments for diseases we cannot yet cure. Education personalized to every student, at a fraction of current cost. Climate models precise enough to guide real decisions. Prevention of vector-borne diseases like dengue fever that infect millions every year. Faster discovery of the critical minerals the energy transition demands. Drug repurposing that finds new lifesaving uses for medicines already approved. These are not incremental improvements. They are the kinds of advances that could fundamentally change the human condition.

This is not a distant aspiration. It is already happening. Google DeepMind’s AlphaFold solved protein folding, one of biology’s grand challenges, opening doors in biology that weren’t previously visible. KoBold Metals is finding critical mineral deposits that decades of traditional exploration had missed. Every Cure is repurposing drugs to treat diseases that have long been neglected. AI is not merely helping us do familiar things faster. In domain after domain, it is helping us see what we could not see before.

And when we can see what was previously invisible, we can begin to solve what once seemed unsolvable.

That is why the most important question for leaders is not simply: where can AI make us more efficient? It is: what can AI help us discover that we could not discover before?

And yet, most organizations are settling for far less. They are focused on the easy wins: automating existing tasks, cutting costs, speeding up processes that were already working. These gains are real. But they are small, and they distract leaders from the harder, more consequential question: what problems that seemed permanently beyond reach are now solvable in your industry?

The distinction is fundamental. Efficiency moves an organization closer to the existing frontier. Discovery moves the frontier itself. Efficiency improves the current business. Discovery makes new businesses, new capabilities, and new sources of advantage possible. Efficiency can often be copied. Discovery, when embedded in proprietary data, domain expertise, and organizational learning, can compound.

That compounding is the mechanism that matters.

“Efficiency moves an organization closer to the existing frontier. Discovery moves the frontier itself.”

The Mechanism: The Compounding Loop

Getting to the big promise requires a specific architecture. I call it the compounding loop: a continuous cycle with four stages. Human knowledge seeds the machine. The machine surfaces discoveries that exceed any individual’s reach. Human expertise, judgment, and vision translate those discoveries into new value. And the knowledge generated through that process becomes the foundation for the next cycle — richer data, sharper models, deeper expertise. Each cycle makes the next more powerful. The loop accelerates as it runs.

AlphaFold is the textbook illustration.

Phase 1: Human Knowledge Seeds the Machine

For more than half a century, one of biology’s grand challenges was protein folding: predicting the three-dimensional structure of a protein from its amino-acid sequence. The problem mattered because structure helps determine function. To understand how proteins work, how diseases progress, and how drugs might intervene, scientists needed to understand the shapes proteins take.

But mapping protein structures was slow, expensive, and painstaking. Experimental methods such as X-ray crystallography, nuclear magnetic resonance, and cryo-electron microscopy produced extraordinary knowledge, but they required enormous human effort. Determining the structure of a single protein could take months or years.

By the time DeepMind began its work, humanity had built a remarkable archive, mapping roughly 110,000 protein structures. That archive represented decades of human labor: the accumulated work of chemists and structural biologists around the world. It was not simply a dataset. It was the product of human curiosity, expertise, failure, correction, and discovery. This human knowledge seeded the machine.

AlphaFold’s achievement was not that it bypassed science; it was that it learned from science at a scale no human could match.

The lesson for any enterprise is immediate: AI algorithms are becoming widely accessible. Your operational data, your institutional memory, and the tacit understanding embedded in your people are the indispensable raw materials that anchor the technology in something uniquely yours.

Phase 2: Machine Discovery Elevates the Human Baseline

Once trained on the cumulative record of protein structures, AlphaFold revealed an entirely hidden world. It did not merely make an existing process marginally faster; it transformed the scale of the problem entirely. Within a remarkably short window, AlphaFold predicted the structures of over 200 million proteins, virtually every cataloged protein known to science. In a single leap, the machine handed humanity a radically sharper, complete map of biological reality. What once required extraordinary effort could now be looked up in seconds.

Phase 3: The Expanded Loop Captures Economic Value

But discovery alone is not enough; the map is not the destination. What the machine reveals creates a new starting line, a fundamentally richer understanding of the world. The question that follows is always the same: now that we can see what was previously invisible, what becomes possible? And that is where human expertise, judgment, and vision come in again.

The machine expands the field of vision. Human expertise determines what to do with what becomes visible.

This is why AlphaFold is so important for business leaders to understand. It is not merely a story about biology. It is a story about the new logic of competitive advantage.

Human knowledge trained the machine. Machine discovery expanded human understanding. Human expertise began translating that expanded understanding into new scientific and economic possibilities. And the knowledge generated through that process now becomes the basis for the next cycle of discovery.

Google DeepMind had both the understanding of what AlphaFold had produced, and the vision to recognize that translating protein structure predictions into therapeutic drugs required a completely new organizational vehicle. To capture this new economic value, Alphabet launched Isomorphic Labs, combining world-class AI experts with deep human domain expertise in medicinal chemistry and biology. Isomorphic is recoding the drug discovery process from first principles, designing novel drug candidates computationally at a speed and precision that traditional lab-based methods cannot match, then moving the most promising into physical validation. The company has already secured major research partnerships with Novartis and Eli Lilly, raised $2.1 billion in 2026, and expects its first AI-designed drugs to enter clinical trials by the end of this year. What began as a scientific breakthrough is on the verge of treating real patients.

Phase 4: Knowledge Capital Accumulates

Each cycle leaves the organization knowing something it did not know before. New data from drug candidates in validation. New models trained on real-world molecular interactions. New expertise in the scientists and engineers who ran the experiments. New intellectual property in the designs and discoveries. This accumulated knowledge capital does not reset between cycles. It compounds. It becomes the foundation on which the next cycle is built, making every subsequent loop faster, richer, and harder for any competitor to replicate.

“The machine expands the field of vision. Human expertise determines what to do with what becomes visible.”

This is the loop. But running it well is not automatic. Many organizations stall in Phase 1 or 2, for a predictable reason: they lack vision. Vision is where the loop begins: the conviction that if you can see something that was previously invisible, you can create solutions to problems that seemed unsolvable. It was this vision that led Google DeepMind to invest in mapping protein structures in the first place: not because the technology was ready, but because they could see what it would make possible. AI fluency is what allows leaders to recognize which problems AI is genuinely equipped to solve, and which it is not. And when the machine delivers its discoveries, it is domain expertise and accumulated judgment that determine what they mean and where they lead.

By feeding this expanded human understanding back into the technology, the loop accelerates once again. Human intelligence informs the machine, the machine informs the human, and the cycle repeats at an entirely new production frontier. This is the aspiration for every organization: not just to operate closer to the existing frontier, but to define a new one, achieving a level of performance that resets the standard for your entire industry.

The Human Imperative

The efficiency narrative misses something fundamental: it treats AI as a replacement for human intelligence. The compounding loop depends on exactly the opposite. Remove the human from the loop and you get efficiency. Keep them in, and you get discovery. Discovery is where the big promise lives.

AI without human expertise can generate output. AI with human expertise can generate knowledge. That distinction is the difference between a productivity tool and a compounding advantage.

There is a moral dimension to this argument. But there is also a strategic one, and for most business leaders, the strategic case is what matters. The defensible asset is never the raw AI model. The moat is everything around it: your proprietary operational data, your organizational design, and above all, the accumulated, tacit knowledge of your people. A competitor can license the exact same baseline algorithm, but they cannot download the thousands of iterative, mutual-learning cycles that have occurred inside an organization running its own compounding loop. They are not just behind a technology curve; they are locked out of a learning loop. The moat is not the algorithm alone. The moat is the learning loop.

“AI without human expertise can generate output. AI with human expertise can generate knowledge.”

Nobel laureate and MIT economist Simon Johnson has argued in the Financial Times that the true promise of AI is found not in automating existing work to use fewer people, but in enabling humans to do entirely new things, creating new fields and a renewed demand for deep human expertise. As I see it, the defining question for every organization is not how many tokens you spend, but whether you deploy them to amplify your human capital or substitute for it. The organizations that win will be those that use AI to deepen the expertise, judgment, and accumulated knowledge of their people, not those that use it to replace them. Satya Nadella, CEO of Microsoft, has arrived at a similar framing, arguing that the frontier lies in the optimal combination of token capital and human capital. True economic value, the kind that compounds over time and perpetually widens the gap between a firm and its competitors, lives in the loop where human intelligence and machine capability make each other smarter, cycle after cycle.

Starting the Loop

Building this loop does not require a scientific moonshot. It begins with a strategic choice: targeting AI at the pivotal processes that define how your firm creates value, rather than peripheral ones. Spreading limited resources across dozens of back-office tasks may generate short-term accounting value, but it fails to build the kind of compounding know-how that creates durable advantage.

Every organization has a small number of processes where its proprietary data and deepest domain expertise intersect. For a pharmaceutical company, it may be drug discovery. For a bank, it may be pricing risk or advising clients. For a manufacturer, it may be designing novel products. For a retailer, it may be better understanding customer demand. For an energy company, it may be finding resources or managing the transition to cleaner power. That intersection is where the loop should begin.

Find it, and point the machine there. That is where distinctive, difficult-to-replicate knowledge lives. And once the loop begins, every cycle makes the next one more powerful, until the distance between you and your competitors is no longer a gap in technology, but a gap in accumulated wisdom that no competitor can shortcut.

The Dual Scale of the Promise

The business case and the civilizational case are not in tension. They are the same argument at different scales.

When a firm runs the compounding loop (human knowledge seeding the machine, machine discovery elevating human capability, human judgment driving the next cycle, and accumulating knowledge capital) it builds competitive advantage. It generates proprietary know-how that compounds over time and becomes increasingly difficult to replicate. That is the business case.

But when many firms run this loop across many industries simultaneously, the aggregate effect is something far more significant. Better healthcare. Cleaner energy. More abundant food. Resilient financial systems. Safer cities. The production frontier of human possibility itself moves forward. That is the civilizational case.

“It is the pursuit of competitive advantage, the drive to discover what rivals cannot, that cumulatively changes the world.”

The two are inseparable. It is the pursuit of competitive advantage, the drive to discover what rivals cannot, that cumulatively changes the world. This is why the choice organizations make about how to deploy AI matters so much, and why the efficiency trap is not just a strategic mistake. It is a missed opportunity of historic proportions.

The organizations that will define the coming decades are those that aim at the hard problems, build the organizational architecture to pursue them, and have the courage to recode their underlying business logic to make it possible.

The future depends entirely on what you choose to recode today.

I call it the Transformation Flywheel.

Why a Flywheel?

History teaches us that general-purpose technologies cannot simply be bolted onto existing processes. Steam, electricity, and the digital computer each demanded complementary innovations — new organizational structures, new skills, new ways of working — before they delivered their full economic potential. AI is no different. The enterprises that treat it as a plug-in will capture a fraction of its value. The companies that invest in the complementary innovations around it will capture all of it.

In evaluating the organizations navigating this transformation journey, I have found that competitive success hinges on a balanced, simultaneous execution across five distinct dimensions: vision and strategy, know-how and IP, the technology ecosystem, talent and work, and education. Many companies approach them as independent initiatives: an education program here, a technology investment there, a talent strategy somewhere else. They treat each dimension as a standalone workstream with its own budget, timeline, and sponsor.

This is a fundamental strategic error. These dimensions are not independent; they reinforce each other, and they follow a compounding logic. Education to drive AI fluency is the non-negotiable prerequisite. Without it, leaders cannot accurately evaluate opportunities, set strategic direction, or make sound capital allocation decisions. Once fluency is in place, you can generate a clear Vision of the future: a definitive view of how your company will compete, not just today, but in the economy that is emerging. This vision reveals your distinctive know-how, because every successful enterprise wins on its ability to know how to do something central to its value proposition better than its rivals. Know-how tells you where to point your AI: at developing and sustaining your competitive edge. Your Technology Ecosystem provides the capabilities that enable you to build and operate the AI, data, and software that bring that edge to life. And your Talent and Work models are what make it all real. The skills, roles, and ways of working must continually evolve alongside the technology.

When these dimensions operate in harmony, they trigger a self-reinforcing flywheel: each investment makes the next one more valuable, and the entire system generates a compounding economic advantage that no standalone initiative could ever produce. Conversely, when they are left to languish in isolation, you get what most companies suffer from today: a fragmented collection of AI pilots that never scale, and a growing frustration that the technology isn’t delivering on its promise.

Integrating all five is your Operating Model, which sits at the very center of the system. To demonstrate how this framework moves an organization to the frontier in practice, we look to the transformation of Reuters.

The Five Dimensions

Vision and strategy

The flywheel begins with vision: the ability to look across your business and ask, “What is now possible that was unimaginable before?” This is a strategic question, not a technological one. And getting it right requires disciplined imagination: a deep strategic fluency with what the technology can achieve, combined with the foresight to see where it can redefine your industry’s production frontier.

Reuters’ vision is anchored on a single competitive advantage that has defined the company since its inception: having the news that no one else has. In 1850, Paul Julius Reuter famously used carrier pigeons to carry market messages between Brussels and Aachen because they were faster than the regional trains. Speed has been coded into Reuters’ DNA ever since.

We see AI as an incredible accelerator, not a strategy. We need to be very clear where we are heading, or we will quickly end up in the wrong place.

Paul Bascobert, President, Reuters

Today, AI has replaced the pigeons, but the underlying strategic logic is identical: be first, or be irrelevant. Sometimes that edge comes from embedding journalists in high-stakes locations where competitors are absent. Sometimes it means utilizing AI to sift through mountains of complex data faster, asking the right questions to be the first to uncover a pattern no one else saw. Either way, the result is the same: Reuters has the story, and its rivals don’t.

Critically, Reuters’ leadership is clear that AI is not a short-term experiment but a long-term business transformation. Their goal is not incremental efficiency, but a redefinition of the production frontier of journalism. They are implementing a recoded business model designed to shift human capital to the high-value front end of reporting—getting the first version of history—with more people breaking news than moving it through the system, more journalists in the field, poring through documents and data sets, and holding institutions to account.

AI-generated cost savings are being reinvested into expanding these core capabilities. This distinction matters enormously. A vision aimed solely at efficiency will generate fleeting accounting value, cost savings that show up in the financial statements but are effortlessly replicated by competitors. A vision aimed at shifting the production frontier will generate true economic value, the kind that compounds over time because it is fundamentally defensible.

Know-how and IP

The second dimension consists of the proprietary knowledge and intellectual capital that render your competitive position defensible. Raw AI capability itself is rapidly being commoditized; foundation models are available to everyone, and powerful open-source algorithms proliferate. The algorithm itself is rarely a durable moat. What remains excludable is your accumulated know-how, proprietary operational data, and institutional knowledge that you train it on. This is the unique raw material upon which AI must be trained, and it determines whether your AI produces generic, commodity outputs or something no competitor can match.

At Reuters, this know-how runs extraordinarily deep. It is forged from the decades of insight into how the world’s disparate interests connect. A missile strike in Iran becomes an oil supply story, which becomes a global energy story, which becomes a China rare-earth minerals story, which becomes an Australian mining M&A story. Reuters unique cognitive advantage is its ability to see these connected interests all at once, and AI helps them to capitalize on that hidden web of knowledge at unprecedented speed.

It is also the editorial expertise built over decades: an understanding of what constitutes a story, how to rigorously verify information under extreme time pressure, how to provide context that readers cannot get elsewhere, and the source relationships that give Reuters access to information others simply do not have.

The AI systems can scan 100,000 press releases a day and generate a first-draft news alert in seconds. But it is the seasoned judgment of experienced journalists—knowing what is genuinely newsworthy, identifying what demands verification, and uncovering what requires deeper context—that makes the output valuable.

A competitor can license the same foundation models, hire a capable AI team, and mimic the software tool. But they cannot download the decades of knowledge that Reuters’ journalists possess. That human know-how is the ultimate moat.

Technology ecosystem

The third dimension is the overarching technology infrastructure that powers and scales your AI capabilities. The critical strategic question is never a narrow technology choice like “what technology or vendor model should we use?” Rather, it is a structural architectural map: “What must we build to own our unique edge, what should we buy to accelerate deployment or save money, and what capabilities should we partner for?” The answer depends entirely on where your complementary assets are.

Reuters made a deliberate strategic choice to co-locate its dedicated AI Labs team with editorial, product, and development teams. By embedding the engineers with the domain experts, they built proprietary tools like Fact Genie (which scans incoming press releases, government and corporate disclosures, identifies the most newsworthy elements, and suggests draft alerts for journalists to review and publish) through rapid, agile iteration with journalists in the loop.

The result of this integrated ecosystem was staggering. Fact Genie went from concept to production deployment in just four months. They discovered quickly that different underlying large language models suited entirely different needs. While early algorithmic iterations took minutes to generate a news alert; newer, optimized models do it in under 30 seconds.

These technology choices were dictated not by a desire for the most computationally sophisticated technology, but by what best served the editorial mission of speed and accuracy. Since then, Reuters has built a set of proprietary tools that systematically accelerate the execution of a wide array of high-frequency everyday tasks including automated story drafting, instant headline and summary generation, intelligent video editing, and near real-time language translation.

The advisory rule for the technology ecosystem is clear: Build what is core to your distinctive competitive advantage. Partner or buy for everything else.

Talent and work

The fourth dimension is the workforce and the organization of work: who you recruit, how you systematically prepare existing employees for the transformation ahead, and how you reconfigure your work processes.

In deploying AI, Reuters uncovered an invaluable lesson: frontline domain experts, the journalists, know best what kinds of AI tools can unlock value and alleviate friction. Of course, this requires a level of fluency with AI. Building on this realization, Reuters created a secure, sandboxed AI playground. Within this protected digital environment, journalists could freely experiment with building, coding, and testing tools without the risk of their work escaping into the public domain.

Rather than tell journalists what they had to do, leaders put the technical building blocks in the room and invited them to invent tools that would directly help them do their own work. This bottom-up initiative caught fire across the enterprise. It yielded more than specific productivity tools; it catalyzed genuine operational breakthroughs that a traditional top-down directive could not have conceived. It resulted in a new framing: AI for journalists, created by journalists. The ultimate payoff was a profound culture shift—an environment of continuous invention, where building a tool your colleagues use became a source of professional pride.

Of course, none of this cultural transformation occurred without friction. When AI-driven workflow changes were first introduced, they triggered anxiety among the staff, sometimes voiced through resistance, and other times expressed through cynicism.

Reuters leadership addressed the friction head-on. They coupled transparent communication regarding the necessity of embracing change with the right deployment strategy. They didn’t pilot the new AI systems on low-risk, peripheral tasks. Instead, they deliberately embedded it with their most senior, battle-tested journalists on their highest-stakes beat: the 10 senior reporters covering US financial markets. Once those senior reporters demonstrated safe, reliable results on a high-stakes beat, adoption expanded rapidly across the newsroom. However, as the deployment widened, Reuters uncovered a highly subtle behavioral pattern: different tiers of operational experience produced entirely different results when interacting with identical algorithmic systems.

When an automated AI tool engineered for headline and synopsis generation was introduced, junior editors worked faster. Conversely, senior editors actually slowed down. They were meticulously analyzing the AI’s generative choices and rereading the original source texts more carefully.

A traditional manager might have misdiagnosed this slowdown as a failure of technological adoption or some other bottleneck. But true strategic fluency recognizes it for what it truly is: deep human expertise and seasoned judgment acting as the ultimate quality gate for protecting organizational trust. Rather than viewing this friction as a metric problem, Reuters leveraged the strategic insight to rethink how it deployed and tiered AI tools for different roles across the enterprise.

Education

The fifth dimension is the catalyst that makes all the others possible: cultivating baseline AI fluency across the entire organization. True fluency is the essential prerequisite of modern leadership. It is not a superficial, casual familiarity with trending tools, but a rigorous, comprehensive understanding of what AI can and cannot do, how it evolves, and where it breaks.

Without this baseline fluency, leaders cannot formulate the right corporate vision or make sound investment choices. Without it, managers cannot accurately evaluate AI outputs or exercise the seasoned judgment that separates good decisions from dangerous or systemic risks. And without fluency, frontline employees cannot work effectively alongside AI tools, and they certainly can’t participate in the development of new tools.

At Reuters, fluency is being built directly through active, hands-on use. Reuters instituted explicit operational metrics to track AI adoption for every single employee. As adoption deepened, Reuters strategically shifted its metrics from capturing weekly datapoints to measuring daily AI usage, incentivizing employees to incorporate AI in everyday workflows.

This method of organic adoption, where fluency propagates through daily execution and practice, rather than through classroom training alone, is generating a multiplier effect. The more your frontline experts experiment, the more your core workflows improve, and the more the entire organization learns what AI can and cannot do in the context of their work.

The Operating Model at the Center

At the absolute center of the flywheel sits the operating model: the way a company actually organizes itself to execute its strategy. The operating model serves as the hub of the flywheel; it is where all five dimensions converge, integrate, and translate into real-world performance.

A critical, defining element of the operating model is governance, and the most consequential governance question in the AI era is deceptively simple: where do you invest? Because AI is a general-purpose technology, it offers an almost unlimited number of potential enterprise use cases. This sheer versatility is precisely what makes the investment decision so difficult for management. The natural executive temptation is to spread precious resources across dozens of disparate applications. Strategic discipline, however, demands the exact opposite: concentrating investment on the pivotal processes that will structurally define how your company competes in the future, not on how it increases efficiency today.

Reuters maintained this strategic discipline. They targeted their investment directly at the systems that ingest breaking news, execute complex contextual analysis, and accelerate the cadence at which journalists write and publish. The company then redesigned its operating model around that competitive priority, integrating AI into publication workflows to compress cycle times.

Throughout this transition, the guiding principle has remained absolute: Journalists must remain at the center of AI workflows. Every single operational variable—how journalists work, how editors review, how technology supports the newsroom, how AI-assisted content is governed—was evaluated around the guiding principle. The people who do the work shape how the technology serves them.

But governance is not only about how and where to invest capital; it is also about what boundaries you set. Reuters’ celebrated Trust Principles, originally forged during World War II, have stood as the moral and operational anchor of its global journalism for decades. As AI entered the newsroom, those principles became the framework for governing its use.

Culturally, we were inspired by our history as an early adopter. Operationally, we did our best to coordinate experimentation with guardrails and investment with strategy.

Paul Bascobert, President, Reuters

Under this model, the human professional, never the algorithm, bears final responsibility for what gets published. Every piece of AI-generated content is transparently labeled. As Reuters’ Head of AI Strategy, Jane Barrett, has articulated, the goal is not merely to keep humans in the loop, but to keep them in control.

In an automated world where AI can produce content at massive scale, robust governance is the only factor that ensures speed never comes at the cost of trust. Paradoxically, establishing clear boundaries makes the flywheel spin faster. When your frontline professionals know exactly where the hard legal, ethical, and brand guardrails are, they can push right up to the ragged edge of innovation without hesitation.

Like any mechanical flywheel, the system starts slow. The early capital investments are heavy, the near-term returns are uncertain, and the temptation to spread resources thin is immense. But as these five core dimensions begin to lock together and reinforce each other, the flywheel picks up speed and the engine shifts into a state of self-sustaining momentum. Know-how compounds. Talent deepens. Workflows sharpen. And with every turn, the system spins faster, generating advantages that are increasingly difficult for any competitor to replicate. By the time a rival recognizes the structural game you are playing, your flywheel is already moving far too fast to catch.

Building for a Future We Cannot Yet See

We do not yet know what the AI economy will look like when it matures, just as no one standing in the agricultural economy could have imagined the industrial age, and no one in the industrial economy could have foreseen the hyper-connected digital world that followed. What we do know with absolute certainty is that the future will look fundamentally different from today.

Navigating this transition demands a continuous, self-reinforcing engine. A clear strategic vision tells you where to aim the technology. The right talent, properly educated, executes that vision. Execution generates proprietary operational data and know-how. And that know-how, embedded in a redesigned operating model, becomes the competitive advantage that rivals cannot replicate. Each dimension feeds the others, and the entire system accelerates with every turn.

The purpose of the transformation flywheel is never to execute a static plan. It is to keep steering the enterprise dynamically toward that emerging horizon, pivoting as the technology evolves, as competitors shift their positions, and as new possibilities and entirely new production frontiers reveal themselves. The companies that will break under the weight of the AI era are those that invest in one or two dimensions in isolation. The enterprises that build durable, compounding economic advantage are those that build the full flywheel.

The firms that win this era won’t be those that deployed the most sophisticated off-the-shelf models, but those that possessed the courage to restructure their underlying business logic.

So if AI-driven value leaks, what exactly is defensible?

The Defensibility Question

The answer is rarely, if ever, the AI itself. Raw AI capability is becoming widely accessible. Foundation models are available to everyone. Powerful open-source algorithms proliferate. If your strategy depends on having better AI than your competitors, you are building on a foundation that erodes continuously; the models that give you an edge today will be available to everyone tomorrow.

What is defensible are the things that surround the AI: the complementary assets that are difficult to observe, difficult to replicate, and that compound over time.

Proprietary data is the most obvious. An AI model is only as powerful as the data it learns from. If your data is unique, continuously generated by your operations, and unavailable to competitors, then the insights your AI produces are inherently difficult to copy. The model can be replicated. The data cannot. Consider John Deere. For years, their tractors and combines have been collecting granular, field-level agricultural data across millions of acres. When Deere applies AI to precision agriculture, it is working with a data foundation that no competitor can assemble from scratch, generating new know-how that raises the productivity of its customers. A rival can build a comparable AI model. They cannot conjure decades of proprietary field data into existence. This is the new business logic in action.

Or consider Waymo, which has moved well beyond its origins, scaling its commercial robotaxi service to an expanding roster of cities. Today, acquiring a highly capable self-driving algorithm is neither difficult nor particularly expensive. As with Deere, what is not available is the data captured from millions of miles of driving in real-world conditions: the edge cases, the unpredictable behaviors of pedestrians and other drivers, the endless variation of weather, road surfaces, and urban environments. It is this continuous training on proprietary data that improves the algorithm, sharpens the decision-making, and compounds the advantage with every mile driven. A competitor can build an autonomous driving system. They cannot shortcut the miles.

Accumulated knowledge of a firm is equally important, and often overlooked. Every company possesses a deep reservoir of institutional knowledge: not just what is stored in databases or even in its training data, but the expertise embedded in its people, the hard-won lessons of decades of operations, the tacit understanding of how things actually work that lives in the judgment and experience of the workforce. Much of this knowledge has never been written down. It exists in the way a seasoned engineer diagnoses a problem, in the intuitions a portfolio manager brings to an unfamiliar market, in the institutional memory of what has been tried before and why it failed. AI can amplify this accumulated knowledge enormously.

I compare AI to a microscope, an instrument that reveals what was previously invisible. When that microscope is trained on a company’s accumulated knowledge, it surfaces patterns, connections, and insights within it that no individual could see. A competitor can license the same AI model, but they cannot license the decades of experience your organization has built. And it is this accumulated knowledge that underpins something AI cannot replace: human judgment. AI can surface an insight, but it takes experienced leaders and professionals to look at that insight and say, “yes, this is right, let’s go for it.” That judgment, the ability to evaluate what AI produces, to know when to trust it and when to question it, is forged through years of accumulated expertise. It cannot be automated, and it cannot be copied. This is why the most important complement to AI is not more technology. It is talented people whose expertise, judgment, and institutional knowledge make the technology worth having.

Workforce composition is another. KoBold Metals, a mining startup focused on making mineral exploration scientific and repeatable, made headlines in 2023 when it announced the largest copper discovery in over a decade, outperforming its established rivals. The Mingomba deposit, buried roughly a mile beneath the surface of the earth, in Zambia, is expected to yield more than 300,000 tons of copper annually. What made the discovery possible was not just the AI; it was who built it. KoBold structured its workforce in equal thirds: geoscientists, AI experts, and software engineers. That combination is not something a competitor can quickly replicate. The integration of domain expertise with AI and software capabilities creates organizational knowledge that lives in the interactions between people, not in any single system.

Organizational design and operating models matter as well. How a company structures its decision-making, how it integrates AI into its workflows, how it governs the use of AI: these are deeply embedded capabilities that cannot be observed from the outside or downloaded from a vendor.

And the quality of the questions leaders ask may be the most underappreciated source of advantage of all. Consider CathWorks, which Medtronic has agreed to acquire for up to $585 million. For decades, cardiologists who found a blockage in a patient’s coronary artery faced a costly next step: to determine whether that blockage was actually restricting blood flow, they had to thread a thin pressure wire into the heart, an invasive procedure that added time, cost, complexity, and risk. The established players in this space focused on refining that wire-based approach, making it more precise and more efficient. They were optimizing within the existing production frontier.

CathWorks reframed the problem entirely. Instead of asking how to improve the wire, it asked whether the wire was necessary at all. Could AI extract the same clinical insight from the imaging that cardiologists were already capturing during routine procedures? The answer was yes. Their system achieves approximately 93% diagnostic accuracy compared with the invasive standard, without the wire, without the added risk, and without the added cost. The breakthrough was not a better device. It was a different question about what existing data could reveal.

This illustrates something business leaders should pay close attention to. AI does not always create value by requiring entirely new inputs or data sources. Often, its greatest impact comes from extracting far more insight from assets you already own: data, workflows, and processes that have long been treated as fixed. By acquiring CathWorks, Medtronic is not just adding a product to its portfolio. It is making a strategic bet that AI can unlock far more value from data and processes that already exist. That is a frontier shift.

The ability to ask the right question, at the right time, with the conviction to pursue the answer, is a form of organizational capital that no competitor can easily replicate.

Value Creation vs. Value Capture: A Framework

Let me bring this together into a simple framework.

Value creation is about pushing the production frontier outward, finding new possibilities, delivering better outcomes for customers, reducing costs, discovering new insights. AI is extraordinarily good at this, and it gets better continuously.

Value capture is about your ability to hold onto the economic value you’ve created, to prevent it from leaking to competitors, being competed away, or becoming the new industry baseline. This depends not on the AI, but on the complementary assets that surround it.

The firms that fall into the value trap are those that focus entirely on creation. They deploy AI, see genuine improvements, and assume the job is done. But without defensible complementary assets, every improvement they make is available to their competitors within a short cycle of imitation.

The firms that escape the value trap are those that invest simultaneously in AI and in the proprietary data, workforce composition, organizational design, and leadership capability that make their AI-driven value difficult to replicate. They understand that the technology is necessary but not sufficient. The moat is not the AI. The moat is everything around it.

And there is a practical imperative here: target AI on your pivotal processes. Every company has a limited budget for AI investment and a limited capacity to absorb change. Consider Reuters. In news, being first to publish confers competitive advantage; being second matters far less. Reuters has focused its AI initiatives squarely on speed: compressing the time between when a story breaks and when it reaches readers. That is not a peripheral use case. It is the core of how Reuters competes.

Leaders who spread that budget across dozens of peripheral use cases will generate a portfolio of accounting value and little economic value. Leaders who concentrate it on the processes that sit at the heart of their value proposition, the ones that define how the company wins, will build advantages that are far harder for competitors to replicate. The question to ask of every AI initiative is not just “will this save money?” but “will this strengthen the core of how we compete?”

The Leadership Imperative

This has profound implications for how leaders should evaluate AI investments. The first question should rarely be “will this create value?” Almost any well-implemented AI initiative will create some form of accounting value. The more important question is: will we be able to capture the economic value this creates, and for how long?

That question changes everything. It shifts the conversation from technology to strategy. It forces leaders to think about defensibility before deployment. And it reveals one of the most underappreciated reasons why AI investments deliver disappointing returns: not because the technology failed, but because the value it created was never captured.

But the leaders who will define this era won’t stop at capturing value. They will use it to keep pushing the production frontier outward, into territory their competitors cannot follow. Every defensible advantage, every proprietary data asset, every accumulation of institutional knowledge and judgment, becomes fuel for the next leap. The firms that win won’t be those that used AI to do the same things more efficiently. They will be those that used it to do things no one else could, because they built the complementary assets that no one else has.

In the AI era, the danger is not that you fail to create value. It is that you create value brilliantly, and watch it become everyone else’s baseline. The opportunity is to create value, capture it, and use it to reach a frontier that only you can see.

This is the value trap of the AI era. And escaping it requires a distinction that many leaders are not making.

Two Kinds of Value

When executives talk about the “value” of their AI investments, they are usually talking about accounting value: measurable improvements in cost, speed, or efficiency that show up in the financial statements. Consider software development, where AI coding agents are now widely deployed. As research at my Center, and by many others, has shown, companies are writing and maintaining software with significantly less human effort. Many report producing cleaner code as well. That’s real. It’s measurable. And it feels like progress.

But accounting value and economic value are not the same thing. Economic value is the advantage you hold over competitors, the gap between what you can do and what they can do. And that gap allows you to earn abnormal profits that are not competed away quickly. If every company in your industry has access to the same AI coding tools (and they do), then writing the same software faster is not a competitive advantage. It is the new baseline. You haven’t pulled ahead. You’ve just kept up. The value you created hasn’t disappeared — it has simply flowed to your customers, who get more for less, and your vendors, who collect their fees. Everyone benefits except your competitive position.

The real question is different, and far more consequential: if the unit cost of software development drops substantially because of AI, what new valuable software are you going to build? Put differently, are you codifying new know-how — new ways to do new things — into software, or are you sticking to your original plan? The accounting value lies in doing the same things more efficiently. The economic value lies in doing new things that were previously impossible, or impossibly difficult to do well. This is the defining characteristic of a Frontier Firm, the companies that redefine the art of the possible. Most companies are focused on the first. The firms that will pull ahead are those asking the second question.

And where you aim AI matters as much as how you use it. Economic value comes from targeting AI at the pivotal processes that define how your firm creates and delivers its value proposition, not the peripheral ones. Applying AI to speed up back-office tasks may generate accounting value, but it will rarely change your competitive position. Applying it to reimagine how you develop products, serve customers, or make decisions at the heart of your business is where economic value lives.

This is the distinction that separates firms that thrive from firms that spend millions on AI and wonder why their competitive position hasn’t changed. Creating accounting value with AI is relatively straightforward. Creating economic value, the kind that compounds over time and widens the gap between you and your competitors, is far harder.

The Intangible Economy and the Defensibility Gap

To understand why, we need to look at the nature of the assets involved.

U.S. companies spend $800 billion annually on software. I argue that this codebase is the modern firm’s most valuable asset. But software is only part of a much larger shift. When you add R&D, design, branding, organizational know-how, and other forms of intellectual capital, the picture becomes even more striking. In their influential book Capitalism Without Capital, Jonathan Haskel and Stian Westlake documented this quiet revolution: for the first time, developed economies were investing more in these intangible assets than in physical ones like machinery, buildings, and equipment. The most valuable things a modern company owns are things you cannot touch.

This matters enormously for AI investments, because intangible assets behave fundamentally differently from physical ones. You can patent a machine. You can restrict access to a factory. You can lock a competitor out of a physical supply chain. These tangible assets have natural boundaries that make them excludable, and therefore defensible.

Intangible assets have no such boundaries. As Haskel and Westlake showed, they are prone to spillovers: the value leaks. A competitor can observe what you’ve built, study how it works, and figure out how to replicate the underlying logic. Code, by its very nature, can be studied, reverse-engineered, and reproduced. The logic it embodies can be observed in the products and services it produces, and a capable competitor can infer enough to build their own version. And they can hire away your key employees. When your competitive advantage lives in software alone, imitation is not a matter of years. It can happen quite quickly. This is why the advantage can rarely be the algorithm itself (other than in a few exceptional cases). The defensible advantage lies in the algorithm applied to your proprietary data, IP, and know-how that competitors cannot access.

AI makes this defensibility gap dramatically worse. AI doesn’t just create intangible value — it also makes it easier for competitors to replicate it. What once took a team of engineers months to reverse-engineer can now be approximated in weeks. Think of how quickly the frontier LLMs catch up to each other. The very technology that creates value also makes it harder to hold onto.

This raises an urgent question for every leader deploying AI: if the value you create leaks so readily, what exactly is defensible? The answer is more specific — and more actionable — than most executives realize. That is the subject of my next post.

#EconomicValue #AccountingValue #AI #SustainableCompetitiveAdvantage #RoI

In my previous post, I argued that AI is not just a technology of automation but a technology of discovery, one that pushes the production frontier outward for everyone simultaneously. That raises an urgent question: if every company now has access to this power, what separates the firms that will thrive from those that will decline?

The answer is not who adopts AI first. It is who uses it to do things that were previously impossible, or impossibly difficult to do well.

Beyond the Faster Horse

Most organizations are still treating AI as an optimization tool: a way to shave costs, accelerate existing workflows, and extract incremental productivity from legacy processes. These companies are working harder within the boundaries of their current production frontier. They are getting better at yesterday’s game.

But there is a different category of firm emerging. These companies are not using AI to do things faster. They are using it to see things, build things, and solve problems that were previously out of reach. They are not optimizing within the frontier. They are redefining it.

I call these Frontier Firms.

The Economics of the Frontier

The concept draws on a foundational idea in economics: the production frontier, which represents the maximum output an organization can achieve given its existing technology, resources, and knowledge. Within any given technology regime, firms compete by trying to operate as close to this frontier as possible, reducing waste, improving processes, squeezing out efficiency.

But every so often (and it is rare) a new general-purpose technology emerges that gives astute firms the opportunity to define an entirely new production frontier. Steam did this. Electricity did this. The digital computer did this. Each time, the firms that recognized the moment didn’t just compete more effectively within the old frontier. They rendered it obsolete and defined a new one. Because AI is a general-purpose technology, this is precisely where we are now.

Frontier Firms are the ones seizing this moment. They use AI to reinvent their business logic on two fronts simultaneously: what they deliver to the customer and how they deliver it. They don’t just move closer to the existing frontier. They push the frontier itself outward, into territory that legacy logic could not even map.

History tells us what happens next. When Henry Ford harnessed electricity, his insight was not simply to automate artisanal car manufacturing. It was to recognize that the correct organizing logic for electricity was not the craft workshop but the moving assembly line. That was the new business logic, and it redefined the production frontier for the entire industry, delivering a superior, standardized product at a price point that was previously unthinkable. His competitors who remained tethered to the old logic didn’t just lose market share. Many of them ceased to exist.

The AI era is accelerating this same dynamic. Like Ford, the Frontier Firms of today understand that the technology demands a new organizing logic, not just faster execution of the old one. But unlike Ford’s era, where the new logic ultimately centered on a single breakthrough in production, AI touches every dimension of the firm, and we are still in the early stages of learning what that new business logic looks like. And AI offers something Ford never had, not just the power of automation, but the power of discovery.

The General-Purpose Microscope at Work

In my first post, I compared AI to the microscope, an instrument that didn’t just make existing things clearer, but revealed an entirely hidden world. The Frontier Firms I study are pointing that microscope in radically different directions, and what they are finding should fill us with a sense of possibility.

For centuries, human progress has been driven by science discovering new knowledge, each breakthrough opening doors that the previous generation could not even see. AI is now accelerating that process of discovery across every domain, simultaneously. We are on the verge of being able to do extraordinary things: find scarce minerals essential to the energy transition, give patients who have exhausted every treatment option a new chance at health, predict the behavior of the atmosphere with a precision that physics alone could never achieve. These are not incremental improvements. They are leaps, and they are happening now.

Consider what is happening beneath the earth’s surface. Finding critical minerals now requires ten times more capital than it did thirty years ago because the easy-to-find deposits are exhausted. KoBold Metals, a company whose workforce is split equally among geoscientists, AI experts, and software engineers, took a century of mining reports and combined them with muon sensor data to “see” deep into geological formations that traditional prospecting had abandoned. The result was the discovery of the Mingomba copper deposit in Zambia, valued at over $50 billion. Legacy logic had looked at that same ground and seen nothing. But notice something equally important: that one-third, one-third, one-third workforce is radically different from the talent pool of a traditional mining company. It is this new workforce composition, domain scientists working shoulder to shoulder with AI and software professionals, that complements the technology and makes both value creation and value capture possible. The AI alone doesn’t find the copper. The geoscientists alone can’t process the data. It is the combination that moves the frontier.

Now consider what is happening inside the human body. There are roughly 18,000 known diseases but only about 4,000 approved drugs. Every Cure is using AI to process 75 million potential drug-disease matches simultaneously, identifying repurposed uses for existing medications that no amount of traditional research could surface at this speed. They have already repurposed fourteen drugs, including a life-saving treatment for POEMS syndrome, a rare disease that might have waited decades for attention under conventional drug discovery logic.

And the frontier is moving in atmospheric science. Traditional climate models are notoriously computer-intensive and constrained by incomplete physical theories. Google’s NeuralGCM integrates AI into physics-based models to generate millions of accurate ensemble forecasts in under a minute, achieving higher accuracy with far lower compute costs than the previous gold standard. The frontier of prediction itself has moved.

What connects these examples is not the technology. It is the logic. Each of these firms used AI to ask a question that their industry’s existing business logic was not equipped to formulate, let alone answer. KoBold asked: what if we could see through the earth? Every Cure asked: what if every approved drug is also an unapproved treatment for something else? Google asked: what if we could predict the atmosphere faster and more accurately than physics alone allows? In each case, the discovery was not hidden by nature. It was hidden by the limitations of the old logic.

From Fluency to Vision

In my first post, I argued that fluency, a deep understanding of what AI can and cannot do, is an essential prerequisite for leading in the AI era. I stand by that. But fluency alone is not enough. Fluency without direction is just literacy. You can understand the technology perfectly and still fail to see where it should take you.

The defining leadership capability of the AI era is vision: the ability to look at your industry and ask, “What is now possible that was unimaginable five years ago?” This is not blue-sky thinking. It is disciplined imagination, grounded in fluency, aimed at a specific production frontier, and backed by the conviction to make large, staged investments whose returns materialize only when the complementary capabilities are in place.

That last point is crucial. The leaders I work with who are building Frontier Firms understand that AI does not deliver value in isolation. The returns come when data, talent, operating models, and governance are assembled around it. This demands a different kind of courage: not the certainty of a fixed destination, but the conviction to commit to a direction while remaining ready to adapt as the technology advances, competitors move, and customers respond. As George Harrison once sang, if you don’t know where you’re going, any road will take you there. Vision is what keeps you from wandering.

Perhaps no company illustrates this better than Waymo. For over fifteen years, Waymo has been investing in a bet that autonomous vehicles could redefine urban transportation, not incrementally, but fundamentally. That bet required far more than AI. It demanded complementary investments in sensor technology, high-definition mapping, regulatory partnerships, fleet operations, and the slow, painstaking work of building rider trust. For years, the returns were invisible, and along the way, many competitors lost their nerve. Uber sold its autonomous vehicle unit. GM walked away from Cruise. Others quietly shelved their ambitions. Today, Waymo’s service is live in multiple cities, and millions of riders have experienced a production frontier that most of the automotive industry said was decades away. Waymo didn’t just automate driving. It reimagined mobility and had the conviction to stay the course while the complementary capabilities caught up with the vision.

The organizations that defined the last era of business were those that mastered efficiency, operating as close to the production frontier as possible with the technologies of the time. The organizations that will define this one will be those that master discovery, have the vision to act on what they find, and the execution to make it real. The opportunity before us is not merely commercial. It is the chance to solve problems that humanity has struggled with for generations, at a speed and scale that were unimaginable until now. That is what makes this moment so extraordinary, and why the courage to push the frontier has never mattered more.

After forty years of studying how firms win and lose from technological change, I have never seen a wider gap between technological promise and the business logic required to exploit it. In every conversation I have with business leaders and board directors about AI, it comes down to the same question: Where is the value?

Companies are hiring AI teams, licensing foundation models, and launching pilot programs by the dozen. The noise surrounding AI has reached a deafening pitch. But beneath it, executives are confronting a quieter, more uncomfortable truth: most of these investments aren’t translating into competitive advantage.

Over a decade ago, I began describing these technological shifts as a “digital tsunami”—a relentless force capable of erasing old competitive boundaries in an instant. With generative AI, the force and speed of that wave have intensified beyond anything I anticipated.

But the tsunami metaphor, useful as it was to inspire action when the threat wasn’t clearly visible, no longer captures what is actually happening. A tsunami is something you survive. What AI demands is something far more fundamental: that you rewrite the logic your business runs on.

The $800 Billion Question

In the United States alone, companies spend over $800 billion annually on software. For most modern firms, this software is their most valuable asset—the literal “code” that defines how a company operates, serves customers, and competes.

But here is the critical distinction many leaders are missing: this massive codebase reflects a business logic that is now obsolete. The rules, assumptions, and decision frameworks that were coded into those systems over decades—how to price, how to segment customers, how to allocate resources—were designed for a world where information was scarce and expensive to process. They represented the best we could do at the time. AI means we can now do profoundly better—and that means this logic must be recoded accordingly.

Most organizations are responding by bolting AI onto these legacy processes, hoping for incremental productivity gains. They are looking for a software update when they need a completely new map. If you treat AI as just another IT investment, you aren’t building a strategy. You are funding an expensive distraction.

The Technology of Discovery

To find that new map, we must recognize that AI’s true power is not automation. It is discovery.

Think of the microscope. It didn’t just make existing things look “clearer”—it revealed an entirely hidden world of biology that changed the course of medicine forever. Resistance to disease, the behavior of pathogens, the mechanisms of infection—none of it was visible until someone built an instrument powerful enough to see it.

AI is that kind of instrument. Its superpower is the ability to uncover patterns, knowledge, and insights that were previously invisible—not just to the human eye, but to the entire apparatus of traditional business logic.

Here is what makes this so disruptive. Economists talk about the production frontier—the outer boundary of what is possible given existing technology and resources. As a general-purpose technology, AI doesn’t just push that frontier outward for one company or one industry. It pushes it outward for everyone, simultaneously. Every company can now serve customers with greater precision, develop products with deeper insight, and operate with intelligence that was unimaginable five years ago.

That sounds like good news, and it is—for customers. But for incumbents, it means something unsettling: your current value proposition, no matter how strong it looks today, is almost certainly inadequate measured against what is now possible. If AI enables every competitor to raise their game, then standing still isn’t stability. It is decline.

This is what I mean by discovery. AI doesn’t just improve what you already do. It reveals how much better everything could be done—and in doing so, it renders the old logic of your business insufficient. When the underlying nature of information changes—its cost, its availability, its granularity—the logic of the business must change with it.

Every industrial revolution reshapes the competitive landscape and creates new winners and losers. The firms that thrive are those that recognize what each general-purpose technology — the steam engine, electricity, the digital computer — actually demands: not just adoption, but a fundamental rethinking of how business is done. The AI revolution follows the same pattern, with one critical difference. Like every general-purpose technology before it, AI automates. But unlike any of them, it also discovers.

The Great Recoding

Because software is the bedrock of the modern firm, succeeding with AI requires us to rewrite the fundamental business logic that governs our organizations. This is why I call this era The Great Recoding.

We must move beyond the hype of chatbots and productivity hacks to answer the harder, structural questions. Among them:

Value Creation vs. Value Capture. Why do some companies turn technology into a compounding advantage while others spend millions and see no bottom-line impact? In my experience, the answer almost always lies in a distinction that deserves far more attention: creating value and capturing value are two fundamentally different activities, and AI changes the economics of both. A company can use AI to create extraordinary value for its customers and still watch every dollar of profit evaporate—because the technology also made it trivially easy for competitors to do the same thing. The firms that win will be those that understand where AI-driven value is defensible, not just where it is possible.

The Shift in Advantage. There is a popular claim that AI makes “intelligence” a commodity. I think this is both too simple and too comforting. Raw AI capability is indeed becoming widely accessible—but the ability to apply it in ways that reshape a business is not. The real question is not whether intelligence is cheap, but what new forms of advantage emerge when it is. The answer, I believe, lies in the complementary assets that surround the technology: proprietary data, organizational design, talent, and the quality of the questions leaders know to ask. These are the things that cannot be downloaded.

Leadership and Fluency. AI, as a general-purpose technology driving an industrial revolution, demands a fundamentally different model of leadership. The essential prerequisite is fluency: not a superficial familiarity with the tools, but a comprehensive understanding of what AI can and cannot do, how it evolves, and where it breaks. Without fluency, leaders cannot evaluate opportunities, assess risks, or distinguish genuine potential from hype. But fluency is only the foundation. Leaders must then develop the vision to see where AI can redefine their production frontier — and the conviction to make large, staged investments whose returns materialize only when complementary capabilities — data, talent, operating models, and governance — are in place. This is a different kind of courage than most executives have been trained for. It requires patience with ambiguity and impatience with incrementalism, at the same time.

These are the questions this publication will take on.

Why This Substack Exists

I am launching Recoding Business at the Speed of AI because there is a massive gap between the technical capabilities of AI and the strategic leadership required to harness them.

This will not be a publication for breaking news or tool reviews. Here, we will focus on the economics of AI and the quality of leadership judgment. We will strip away the jargon and build the structural roadmap needed to turn this technological shift into a durable source of advantage.

The organizations that thrive in the coming decade will not be those with the most sophisticated models. They will be the ones that learned to ask better questions—and had the courage to rewrite their answers.

I invite you to subscribe and join me as we recode the logic of business for the AI era.

AI is not just another software technology. It is a general-purpose technology, akin to electricity and railroads, that is forcing companies to rethink strategy, operating models, and capital allocation. Many organizations will invest heavily in AI and see little return—not because the technology fails, but because they didn’t make the necessary organizational transformation.

Earlier general purpose technologies (GPTs) and AI are similar in that they lead to shifts in the optimal organizational form, especially around scale and scope. However, AI is alone in that it is also a technology of discovery. Just like a microscope allowed us to see more than we could with the naked eye, allowing us to uncover hidden patterns, identifying new theories, new knowledge, and new insights. This is its superpower.

Still, as a business leader, the central question that concerns you is: How can I create and maintain sustainable competitive advantage in the age of AI. As a result, I focus on the economics of AI, value creation in particular, examining issues like productivity versus profitability, capital intensity, complementary investments, and how AI reshapes competition and market structure. The central question I address is simple: why do some companies turn AI into a compounding advantage while others are fundamentally unable to capture its value?

My name is Vijay Gurbaxani. For more than four decades, as the Taco Bell Endowed Professor of AI and Technology Management, and the Founding Director of the Center for Digital Transformation at the Paul Merage School of Business at the University of California, Irvine, I have worked at the intersection of technology, strategy, and economic value creation, advising and working with senior executives and boards across industries. AI stands apart from prior waves of digital transformation because it challenges some of the most basic assumptions leaders make about the sources of value creation and competitive advantage.

This is not a publication about breaking AI news or the latest tools. It is about something more enduring - leadership judgment in the age of AI—the decisions that determine whether AI becomes a durable source of advantage or an expensive distraction.

If you are responsible for strategy, capital, and long-term performance in the AI era, Recoding Business is written for you.

People