The argument in 30 seconds
- The constraint is the chassis, not the engine. Models already outrun what most organizations can absorb. The limit is organizational design.
- AI does not automate tasks; it transforms decisions. Value compounds only when you redesign around decisions, not workflows.
- AI-enabled and AI-native differ in kind. One bolts AI onto the old model; the other rebuilds the model around intelligence. Only the second is defensible.
- Knowledge is the new infrastructure, and cost hides below the waterline. Hardware is a third of the true bill; the CEO designs the system, the board measures capability.
Executive Summary
Three years into the enterprise AI era, the verdict is in: adoption is nearly universal and value is vanishingly rare. McKinsey’s 2025 survey finds 88 percent of organizations now use AI, yet only about 6 percent capture material earnings from it, and barely a third have scaled it across the enterprise at all. The gap is not a technology problem. It is an organizational one.
The prior edition argued that competitive advantage is consolidating around Enterprise Intelligence, the capability to continuously sense, reason, decide, execute and learn. This edition answers what that provokes: what organization can actually build that capability? Today’s enterprise cannot, because it was designed for a world where humans did nearly all the reasoning. Its silos, hierarchies and periodic planning are not flaws; they were the right answers to a now-obsolete question.
The answer is a different organization, designed from first principles around intelligence rather than retrofitted with it. The AI-Native Enterprise is an organization intentionally designed so that human expertise, artificial intelligence, enterprise knowledge and governance continuously improve every important decision. Deploying AI improves a company incrementally. Redesigning the company around intelligence is what redefines competition.
Opening
The company that fired its own org chart
In the spring of 2025, one of the best-known names in biotechnology did something that should have been a footnote and instead was a signal. Moderna, whose mRNA vaccine reached billions of arms, quietly merged two departments every management textbook keeps rigidly apart: Human Resources and Information Technology. A newly minted Chief People and Digital Technology Officer now runs both, because at Moderna the distinction between managing humans and managing machines had stopped making sense.
Here is why. The company had built more than three thousand custom AI assistants across legal, research, manufacturing and HR, one of which now does the work of a junior HR analyst. So its head of people stopped planning the workforce and started, in her own words, planning the work: looking at each task and asking whether a person, a machine, or the two together should do it. Headcount was no longer the unit of organization. The task was.
Most companies are not doing this. Most companies are doing the opposite. They are buying the same AI Moderna uses and bolting it onto an org chart designed a century ago, then wondering why the promised transformation never arrives. And here is the uncomfortable truth the data has begun to confirm: bolting a jet engine to a stagecoach does not produce a faster stagecoach. It produces wreckage. The engine was never the problem. The chassis was.
This edition is about the chassis. Because the models are already far more capable than the organizations deploying them can absorb, the binding constraint on the AI era is no longer the technology. It is the shape of the enterprise itself. And you cannot buy a new shape. You have to design one.
“I’ve shifted from workforce planning to work planning.”Tracey Franklin, Chief People and Digital Technology Officer, Moderna
The Argument · I
Every enterprise you know was designed for human intelligence
To see why the modern enterprise struggles with AI, you have to see what it was optimized for. For two centuries, one resource performed virtually every act of sensing, reasoning, deciding and learning inside a company: the human mind. Every structural feature of the organization we take for granted is, at root, a workaround for the limits of that single resource.
Humans have finite attention, so we built departments to divide cognitive labor. Humans cannot hold an entire enterprise in their heads, so we built hierarchies to aggregate judgment upward and push commands down. Human reasoning is slow and costly, so we encoded repeated decisions into policies and approvals. Information was scarce and hard to move, so we accepted silos and sequential workflows as the price of coordination, and planned periodically, because gathering information any faster was simply not possible.
None of this was irrational; it was a brilliant adaptation to a real constraint. The ERP system, the org chart, the KPI cascade: each is a monument to the scarcity of human cognition. But adaptations become liabilities the moment the constraint disappears, and artificial intelligence does exactly that, making a form of reasoning abundant, fast and cheap for the first time in history. The structures built to ration human thinking now obstruct the abundant thinking the enterprise could be doing.
This is the deep reason that bolting AI onto the existing structure disappoints. The structure’s entire purpose was to economize on the very thing AI now makes plentiful.
The Argument · II
AI changes the unit of work
Ask most executives what AI does and they will tell you it automates tasks. They are not wrong. They are just describing the least valuable thing it does, and that misunderstanding is precisely what produces the disappointing results the data keeps recording. AI does automate tasks. But its real power is that it transforms decisions. And the decision, not the task, is where enterprise value is actually made or lost.
Consider the progression. A task is a unit of activity: process this invoice. A process is a sequence of tasks: the procure-to-pay cycle. But what creates or destroys value is the decision inside the process: extend this supplier’s terms, flag this transaction, reprice this contract? When AI merely speeds the tasks, the decisions are unchanged and the value is marginal. When AI improves the decisions, value compounds, because better decisions made faster and fed back is the definition of Enterprise Intelligence.
This reframes enterprise architecture itself. The organizations chasing value through AI have built workflow-centric architectures: systems designed to move work through predefined steps. The AI-Native Enterprise builds a decision-centric architecture: systems designed so that every important decision draws on the best available human judgment, machine reasoning, proprietary knowledge and governance, and improves with each cycle. The workflow becomes the servant of the decision, not the other way around.
The Argument · III
AI-Enabled is not AI-Native
Almost every large company today is AI-enabled. Very few are AI-native, and the distinction is not a matter of degree. It is a difference in kind, and it is the difference that will separate the next decade’s leaders from its laggards.
An AI-enabled organization treats AI as a portfolio of projects layered onto an unchanged operating model: departmental pilots, purchased tools, automated existing workflows, counted initiatives. Its mindset is a technology programme, acquire and deploy and measure usage. This is where the 88 percent live, and why so few see earnings move. McKinsey is unambiguous that the strongest differentiator of value is not the technology but the fundamental redesign of workflows, exactly what AI-enabled firms do not do. High performers are nearly three times as likely to have redesigned how work happens.
An AI-native organization treats intelligence as the organizing principle itself. Knowledge is centralized as an asset; learning is continuous, not periodic; governance is built into the flow of decisions, not bolted on; humans and machines collaborate by design, not through hand-offs. Its mindset is operating-model design, not technology procurement. It asks not “where can we apply AI?” but “how should this enterprise be structured now that intelligence is abundant?”
The Framework
The AI-Native Enterprise Model
If the enterprise must be redesigned, what is the design? An AI-native organization is best understood as seven interlocking layers, each of which must be deliberately built. Most companies have some layers and not others, which is why their AI efforts stall: a powerful intelligence layer starved of a knowledge layer produces confident nonsense; a governance layer without a learning layer produces safety without improvement. The layers only deliver as a system.
Leadership sits at the top because the redesign cannot be delegated; only the chief executive owns the operating model, and therefore the intent. Strategy translates that intent into the few decisions where intelligence is concentrated first. Knowledge is the proprietary memory of the firm, structured to reach the point of decision. Intelligence is the engine: the capability that combines people, AI, knowledge and governance into better decisions. Operations is where those decisions become action. Governance is the trust layer that lets the organization act on machine-shaped decisions with confidence. And Continuous Learning closes the loop, so every outcome refines the next decision, turning the structure from a machine into a flywheel.
Definition
The AI-Native Enterprise
An organization intentionally designed so that human expertise, artificial intelligence, enterprise knowledge and governance continuously improve every important decision.
The Argument · IV
Knowledge becomes infrastructure
Every economic era has had a defining form of capital, the asset around which advantage was organized. In the industrial era it was the factory: physical capacity to produce at scale. In the digital era it was software: the systems that ran the business and scaled at near-zero marginal cost. In the AI era, the defining asset is knowledge, specifically the proprietary, structured, continuously updated knowledge unique to a single firm.
The logic is direct. When the models are available to every competitor, the model cannot be the differentiator. What differentiates is what you feed it: the institutional memory, the context, the hard-won operational understanding no rival can buy. Satya Nadella frames this as companies needing not only human capital but token capital, and the real opportunity being for every organization to own the learning loop that encodes its institutional knowledge. Knowledge moves from a byproduct of work to the infrastructure the enterprise runs on.
This has a sharp strategic implication that most boards have not yet internalized. If knowledge is the asset, then the way an organization captures, structures, governs and routes its knowledge is no longer an IT housekeeping matter. It is core strategy, as consequential as a factory location was in 1950 or a software platform choice was in 2010. The AI-native enterprise treats its knowledge architecture as infrastructure, funded and governed accordingly.
The Architecture
What it actually costs, and what it is actually made of
A redesign this deep raises two questions every board will, and should, ask. What does it cost to run? And what are its structural parts, so we can tell whether we are building it or merely talking about it? Both deserve straight answers, because both are where AI-native ambitions quietly die.
Begin with cost, the more dangerous blind spot. The visible price of AI is the compute, and leaders anchor on it. But compute is the tip of the iceberg. Independent 2025 and 2026 analyses put hardware at only about a third of the true five-year total cost of ownership; power, cooling, specialized talent, data pipelines, model drift, security and sheer underutilization make up the submerged remainder, pushing the real total to three or four times the hardware line. No wonder most organizations misestimate their AI costs by more than ten percent, and a quarter by half or more. The AI-native enterprise treats cost as an architectural discipline, not an invoice that arrives later.
The single most useful reframe here is to stop measuring cost by the chip and start measuring it by the token, the unit of intelligence actually produced. A cheaper GPU that delivers fewer tokens per second is not cheaper; it is more expensive per unit of the only thing you are buying. This is the discipline that separates enterprises whose AI economics improve as they scale from those whose costs spiral until, as one study found, they cancel projects. Cost per token, not cost per chip, is the metric an AI-native board should ask to see.
Now the structure. An AI-native enterprise is not a model with software around it; it is a stack of building blocks, each deliberately owned and funded, and two of them decide whether the whole is defensible. The knowledge layer, governed and provenance-tracked, is the proprietary asset that is the new infrastructure. The security and identity layer, which the next edition takes up in full, determines whether the intelligence can be trusted enough to act on. Around them sit the model layer, orchestration, data and retrieval, observability, governance and the learning loop. Miss one and the stack underperforms: intelligence without knowledge produces confident nonsense, and the whole thing without security is a moat with the drawbridge left down.
The Argument · V
The enterprise becomes a knowledge network
Redesigning around intelligence does not abolish the functions. Marketing, sales, finance, HR, legal and operations do not disappear; they remain the homes of deep domain expertise. What changes is how intelligence moves between them. In the inherited design, information traveled vertically, up and down each functional column, and crossing columns required meetings, hand-offs and delay. In the AI-native design, intelligence flows horizontally, through a shared layer of enterprise knowledge that every function both feeds and draws upon.
The practical effect: a decision in one function is instantly informed by the knowledge of all the others. A pricing decision in sales reflects finance’s margins, legal’s constraints and operations’ capacity in real time, not after three rounds of email. The organization stops behaving like a stack of silos and starts behaving like a network, where value lives in the connections as much as the nodes. This is Enterprise Intelligence made structural: knowledge that compounds because it is shared, not hoarded.
For the Chief Executive
The CEO becomes the enterprise designer
Each edition of this series has returned to the same shift in the chief executive’s role, because it is the hinge on which everything else turns. The first edition argued the CEO’s real job had moved from managing technology initiatives to redesigning the enterprise. The AI-native model makes concrete what that redesign consists of.
The CEO of the AI era is increasingly the designer of five systems rather than the manager of five functions: the decision systems that determine how the most important choices are made and who, human or machine, makes them; the knowledge systems that determine what the enterprise remembers and how that memory reaches the decision; the learning systems that set how fast it improves; the governance systems that let it act safely at speed; and the human-AI collaboration model that determines how the two forms of intelligence work together rather than around each other.
One system the redesign lives or dies on is the human one. A CEO can redesign decision flows on a whiteboard in an afternoon; the organization will resist it by Friday. This is the quiet reason so many transformations stall, and why Moderna’s merger of HR and technology under one leader was not a curiosity but the point: the people question and the machine question had become the same question. The redesign asks people to give up tasks they have owned for years, trust a machine’s judgment, and move from doing the work to improving it. None of that happens by decree. It happens through reskilling, visible senior sponsorship, and redesigning roles rather than eliminating them. The firms that treat this as change management, not a technology rollout, are the ones whose redesign survives contact with their own workforce.
This is a profound change in the nature of the job. Managing functions is about running the machine that exists. Designing systems is about building the machine that should exist. The evidence suggests this cannot be delegated: high performers are roughly three times more likely to have senior leaders who personally own and visibly use AI, and fewer than a third of companies have CEOs directly sponsoring the AI agenda, which is a large part of why so few succeed.
“The real opportunity is not in picking the best model but in building a learning system… where every organization can own the learning loop that encodes its institutional knowledge.”Satya Nadella, Microsoft
For the Board
The board’s new dashboard
Boards govern what they measure, and most are still measuring the wrong thing. Asking management how many AI projects are underway is the governance equivalent of judging a research lab by its number of experiments rather than its discoveries. It rewards activity and tells the board nothing about whether the enterprise is becoming more intelligent. The AI-native board retires the project count entirely and adopts a dashboard of capability.
Eight signals matter. Decision quality and velocity: are decisions getting better, and faster? Knowledge reuse: is the firm’s memory compounding? Learning rate: how fast do outcomes improve cycle over cycle? Governance maturity and AI trust: can the organization act on machine output safely and confidently? Business value confirms the capability reaches earnings. And Enterprise Intelligence itself, the health of the sense-reason-decide-execute-learn loop, is the summary metric the rest compose into.
The Map
The AI-Native Enterprise Maturity Model
No organization becomes AI-native overnight, and pretending otherwise is how transformation programmes fail. There is a path, and it helps to name the stages honestly so a leadership team can locate itself without flattery.
At Level 1, Digital, processes are digitized but humans make every meaningful decision. At Level 2, AI-Assisted, individuals use AI ad hoc, producing scattered gains that never aggregate. At Level 3, AI-Enabled, functions run pilots on an unchanged operating model; this is where most large enterprises sit, and where value refuses to scale. At Level 4, AI-Native, the enterprise is deliberately redesigned around intelligence, the seven layers built as a system, and the moat begins. At Level 5, Self-Improving, the organization systematically improves its own decision-making, the flywheel turning largely under its own power.
The uncomfortable truth the model surfaces is that the leap from Level 3 to Level 4 is not incremental. It is a redesign, and it is exactly the leap most organizations are avoiding by staying busy with pilots.
What This Means For You
Where you are, and your next move
The model is only useful if you place yourself on it honestly, without the flattery that kills most transformations. If humans still make every meaningful decision, you are at Level 1, and the next move is not a platform but a single high-value decision redesigned around intelligence. If people use AI ad hoc, you are at Level 2: standardize access and capture the knowledge those interactions generate. If functions run pilots on an unchanged model, you are at Level 3, and the move is the hard one this series keeps returning to, redesigning one end-to-end workflow rather than launching a tenth pilot. Level 4 builds and instruments the learning loop so it compounds; Level 5 extends it across the enterprise.
The value of naming your level is that it converts a vague ambition, “become AI-native,” into a single, fundable next step. That is the difference between a strategy and a slogan.
Looking Ahead
Five predictions for the AI-native decade
These are offered not as speculation but as consequences that follow once the logic of the AI-native enterprise is accepted.
- The performance gap between AI-native and AI-enabled firms will widen, not converge, because intelligence compounds and structural advantages cannot be bought late.
- “AI strategy” will disappear as a separate document, absorbed into enterprise strategy itself, just as “internet strategy” did two decades ago.
- Proprietary knowledge architecture will become a board-level asset class, funded and governed like infrastructure rather than treated as IT overhead.
- Agentic systems will force the operating-model question, because autonomous agents amplify a well-designed enterprise and accelerate the dysfunction of a poorly designed one.
- The CEO mandate will formally include the enterprise’s learning rate, the way it includes the profit and loss statement today.
Conclusion
The revolution that redesigns the enterprise itself
Every great transformation redesigns something. The Industrial Revolution redesigned production: the workshop became the factory. The Digital Revolution redesigned information: the filing cabinet became the database. The AI Revolution redesigns something that has not been touched in two hundred years, and that most boardrooms have not yet noticed is even on the table. It redesigns the enterprise itself, the very structure through which a company senses, decides, learns and acts.
This is why deploying AI and becoming AI-native are not points on the same line. An organization that deploys AI into its existing structure will improve, modestly, and wonder why the promised transformation never arrives. An organization that redesigns itself around intelligence will not merely improve. It will operate on a different basis of competition, one that the deployers cannot reach by deploying harder. The first treats AI as something the enterprise has. The second treats intelligence as something the enterprise is.
The last edition argued that Enterprise Intelligence is the final moat. This one has answered the question that follows. The moat is not built by buying better technology. It is built by becoming a different kind of organization. The companies that understand this will spend the coming decade not deploying AI, but redesigning themselves around it, and in doing so they will not just win the next round of competition. They will redefine what the contest is.
Key Takeaways
- The constraint is the chassis, not the engine. Models already exceed what most organizations can absorb. The binding limit is organizational design.
- Today’s enterprise was built to ration human cognition. Silos, hierarchies and periodic planning were adaptations to a scarcity AI has now removed.
- AI changes the unit of work from the task to the decision. Value compounds only when AI improves decisions, which calls for decision-centric, not workflow-centric, architecture.
- AI-enabled and AI-native differ in kind. One adds intelligence to the old model; the other rebuilds the model around it. Only the second is defensible.
- The AI-Native Enterprise is seven layers as one system: leadership, strategy, knowledge, intelligence, operations, governance and continuous learning.
- Knowledge is the new infrastructure, the CEO is its designer, and the board should measure capability, not count projects.
A question for the boardroom
If your enterprise had been designed from a blank sheet, today, knowing that intelligence is now abundant, would it look anything like the organization you currently run?
And if not, who owns redesigning it?
Selected Sources
- McKinsey & Company, The State of AI in 2025: Agents, Innovation, and Transformation, November 2025 (88% adoption; ~6% high performers; 55% vs 20% workflow redesign; human-in-the-loop 65% vs 23%).
- McKinsey & Company, Seizing the Agentic AI Advantage, June 2025 (the “gen AI paradox”; rewiring the organization; Moderna HR-IT merger; <30% CEO sponsorship).
- McKinsey & Company, The State of AI: How Organizations Are Rewiring to Capture Value, March 2025.
- McKinsey & Company, Building the Foundations for Agentic AI at Scale, 2025 (fewer than 10% scale agents; data as foundation).
- Satya Nadella, “A frontier without an ecosystem is not stable,” and Microsoft Build 2026 remarks on token capital and the institutional learning loop, 2026.
- Stanford Institute for Human-Centered AI, AI Index Report 2025 and 2026.
- NVIDIA, COMPUTEX 2025 keynote, Jensen Huang on AI factories and AI as infrastructure.
- Peter F. Drucker, Managing in Turbulent Times and collected works.
- Michael E. Porter, Competitive Strategy (1980); Clayton M. Christensen, The Innovator’s Dilemma (1997); Rita Gunther McGrath, The End of Competitive Advantage (2013).
- Marco Iansiti & Karim R. Lakhani, Competing in the Age of AI, Harvard Business Review Press (2020); Andrew McAfee & Erik Brynjolfsson, Machine, Platform, Crowd (2017).
- World Economic Forum, Future of Jobs Report, 2025; PwC and Deloitte enterprise AI surveys, 2025.
- CXO Intelligence Series: Edition 01, The CEO’s Real Job in the AI Transition; Edition 02, The Last Moat.
Statistics reflect the cited primary research at time of writing. Figures from annual reports such as the McKinsey State of AI and Stanford AI Index should be read against the most recent edition. Quotations are reproduced from public records of the speakers’ remarks and writings.