Imagine one second from now, all of the knowledge in the world and every computer program ever written answer to your voice. Name the first big problem you would go after.
Now stop imagining. That second has already passed.
For my father, John Murphy,
a wise man who told me as a youth how it would all go;
study, work, write and then teach.
Expertise is earned.
AI Partnership amplifies Expertise.
Now the exponential is you.
The Sophisticated AI Adopter is the companion to Why Us?. Where Why Us? made the case — that a career’s worth of experience and judgment is the most valuable thing a person can bring into a partnership with AI — this book expands that value thesis to its impact on organizations, investment and the AI Economy as a whole.
“The greatest shortcoming of the human race
is our inability to understand the exponential function.”
— Albert A. Bartlett, Physicist, University of Colorado
Faced with a threat, we flee first, think second.
Documented by Walter B. Cannon, Chair of Physiology, Harvard Medical School
A fear is not reasoned away or waited out —
it ends only when you move toward it and confront it.
Demonstrated by Joseph Wolpe, Psychiatrist, Pioneer of Behavioral Therapy
Exponential growth goes unrecognized until late in its rise —
and then it cannot be missed.
Confronting the fear is what lowered THE CEILING —
and brought the exponential into view sooner. Every time.
Every time before, the technology was the engine.
This time your Expertise is the engine.
One Man. Two Chairs. One MacBook Air.
It took ninety days. A man sat down alone in a comfortable chair in the 24th-floor Club Room of a downtown Austin high-rise and opened a MacBook Air. He was trying to solve a financing problem for a developer friend. He had no plan beyond that. He had never taken a course in AI and had used it only the way he had used every search engine before it — to find answers, not to build things.
What followed happened in two locations: that same Club Room chair, and a customer dining chair inside the Whole Foods corporate flagship store in Austin, a few blocks away — in consecutive sessions that ran six, eight, ten, and twelve hours at a stretch. No calls. No meetings. No interruptions. Just thirty years of institutional real estate development and investment experience, an AI Partner, a MacBook Air, and the kind of consecutive, uninterrupted thinking that the inside of a large organization rarely permits.
Ninety days later he had produced five things he had not set out to produce, and each was unimaginable to him at the start: a targeted 30-location, $15 billion national AI infrastructure investment platform, structured at seven times the scale of the largest platform he had previously built and executed for the largest commercial development company in the United States. Two books researched, written and completed on AI's amplification of human experience and judgment. A set of six e-book publishing innovations he invented and then combined into three pending U.S. provisional patent applications, all in an area of business he had never worked in. Two methods that did not exist before he built them: The Thirty, an online interactive simulation designed to elicit a reader’s first “Wow” from an AI Partnership driven by their own Expertise, and The Elevator, the method designed with Expertise as the engine working AI’s inert lever. Each carries its own U.S. provisional patent application, which brings the total to five patent applications now pending. The fifth, and the largest, was the last he saw: a new category of AI use and commerce, Expertise-Driven AI Partnerships, in which a person’s own Expertise drives the work and AI amplifies it. It had no name until he named it, and these two books document it.
The man in those chairs wrote this book. What follows is not the story of those ninety days — it is the explanation of them. Because what happened was not a personal anomaly. It was an early, visible instance of something now reorganizing the economics surrounding human experience and judgment everywhere: the economy’s scarcest input is being amplified — by more than most people are aware.
This is the second of two books about that amplification. The first, Why Us? was written for the individual — the experienced senior executive who had long suspected the structures around them were rationing their most valuable work, and who needed to hear that the rationing was ending. The Sophisticated AI Adopter is written for a much broader audience — ranging from students up to the executives, investors, educators, and other decision-makers who allocate capital, build organizations, and shape careers. The first book is a recognition. This one is an argument about where scarcity and value are now moving, and why.
The destination the argument ultimately points toward is abundance — amplified human experience and judgment producing more, for more people, than has ever been possible. The road there will be rough in places, and this book will not pretend otherwise. But the roughness is the reason for the urgency, not a mark against the hope: the faster people learn to do what happened in those two chairs, the shorter the hard part becomes — for everyone.
Carry one picture through the rest of this book. New gold is piling up around everyone in proportion to their Expertise — the combination of experience and judgment they have earned — and it remains invisible to most. People have always known their Expertise was growing; that is not what is hidden. What is new is that Expertise has become an asset anyone can put to work, and an AI Partnership, a person directing AI rather than AI assisting, multiplies it. Until now it was rationed inside a corporate pyramid, or it lay dormant outside one, with no way to use it on its own. The student has a pile. The chief executive has a pile. Nobody has none. What most people cannot yet see is what theirs has become.
It begins where the work began: with a gap hiding in plain sight.
The Gap
Millions of people now hold the same AI in their hands. The technology improves every month. The barriers that once stood between ordinary users and genuinely powerful AI came down this year, and they are still coming down. For the first time in the history of advanced technology, access has become close to universal.
The outcomes have not.
Some people are creating extraordinary value with AI. Others, working with the identical technology, are producing little more than novelty. Some founders are standing up businesses at speeds that would have been impossible a few years ago, while their competitors use the same systems to draft email a little faster. The technology in each pair is the same. The results are not even close.
Before naming the gap, it is worth understanding why so few people can see it clearly. The limit is perceptual, and it has been documented for half a century.
Albert A. Bartlett, professor of physics at the University of Colorado, delivered his one-hour lecture on exponential growth 1,742 times, beginning in 1969. Its central point was simple: the human mind cannot grasp the exponential. His most cited line is his conclusion:
is our inability to understand the exponential function.” — Albert A. Bartlett, Physicist, University of Colorado
Three questions. Answer each one in your head before you read the line beneath it.
Take thirty steps across a room and you have gone ninety feet. Now take thirty steps where each one is twice the length of the one before. Where do you finish?
Six hundred thousand miles. Twenty-four times around the earth. Most people say a few miles.
How many times do you have to double an inch before it passes the moon?
Thirty-four. The first sixteen only get you to a mile.
One bacterium goes into a bottle at eleven o’clock and doubles every minute. The bottle is full at noon. At 11:55 it is ninety-seven percent empty. Standing in that room at 11:55, would you have said anything was happening?
No. Nobody would have. And that is the finding — it is not that an exponential is large. It is that it looks like nothing at all until very late.
You did not miss those because you are not smart. Nobody gets them right. It is the same reason the new gold goes unseen: a pile that multiplies looks like nothing at all until very late. Expertise — the combination of experience and judgment — multiplied by AI, through an AI Partnership, using the Elevator, the method set out in Chapter Four, is what will prove to you that the third line, “Now the exponential is you,” is not a slogan.
AI in its current form is an exponential phenomenon. Its capability is compounding. The professionals who encountered it in its early stages and concluded it was not relevant to their work were not wrong about what they saw. They were encountering an exponential at the only point in its development when it is possible to mistake it for something small.
That limit has a public form, and the plates name it. This public ceiling is a line of Public Opinion — voters, regulators, employees, communities, governments and the media — and it is not a technical limit of any kind. Its height is the balance between the objections holding it up and the solutions, and the recognition of results, that pull it down. Underneath it the benefits of a new technology grow at their own rate the whole time, invisible to the general public until the ceiling comes down far enough for them to be recognized. Every prior revolution ran under one and every one of them ended above it. The AI ceiling has been rising since 2023, because the objections have been overwhelming the answers to them. Nothing in that is a forecast: a ceiling is lowered by better answers, and what varies is how long it takes.
The perceptual limit produces a predictable response. Walter B. Cannon, Chair of Physiology at Harvard Medical School, documented it in 1915: faced with a perceived threat — and an incomprehensible exponential registers as a threat before it registers as anything else — the human organism initiates withdrawal, fast and automatically, below the threshold of conscious deliberation.
The people who stepped back from AI met mathematics their biology was not built to handle. The nearly eight in ten Americans who expect AI to reduce the number of U.S. jobs — seventy-nine percent, in the 2026 Bentley University–Gallup Business in Society survey, up from seventy-three percent a year earlier — are reading the only business case they were shown, as the next chapter explains. But the speed and heat of the reaction are not a forecast. They are a Cannon response — the autonomic system processing an incomprehensible exponential as a threat and initiating withdrawal before the rational mind has had time to weigh the evidence.
The Gap between who is creating value with AI and who is not is not primarily a gap in access, intelligence, or willingness. It is a gap between the people who have moved through the Cannon response and the people who have not yet been given a reason to. The Bartlett limit is real and cannot be argued away.
Joseph Wolpe (1915–1997), a psychiatrist and pioneer of behavioral therapy, spent a career on what ends a fear of that kind.
Bartlett identifies the limit. Cannon explains the response. Wolpe supplies the method. The way through the perceived threat is not argument — it is contact. Appendix A sets out all three bodies of work.
This is one of the most reliable patterns in the history of technology. When the personal computer arrived, some people rebuilt entire industries around it and others used it as an expensive typewriter. When the smartphone put a supercomputer in every pocket, the device was the same in every hand — and some built companies with it while others mostly played games. The technology always mattered. What people brought to it decided what it became in their hands.
AI is following the same pattern.
This surprises people, because nearly the entire public conversation about AI is a conversation about the machine. Each model is measured against the last one. The unspoken assumption is that a better model simply produces better results. It is a reasonable assumption, and the evidence against it is in plain sight: if capability alone produced value, everyone holding the same model would produce roughly the same results. The spread is wide, and it is growing.
Consider two professionals on opposite sides of the same city, with the same AI, the same information, the same access. One consistently produces insight, opportunity, and decisions that create real value. The other does not. Neither the technology, the access nor the cost explains the difference. Something else is doing the work — something that was present before the technology arrived.
That something is Expertise — the combination of experience and judgment — that no system can acquire without living the consequences of acquiring it. And Expertise is widely misunderstood, because it is constantly confused with information.
They are not the same thing. Information is a resource; Expertise is an ability. Information can be acquired; Expertise must be developed. Information can be downloaded in an instant; Expertise is earned over years. Information answers the question in front of you. Expertise determines which question deserves to be asked.
The modern world blurs this distinction because information has become so abundant, and AI makes the retrieval easier still. It becomes easy to assume that a person with access to unlimited information is therefore equipped to act on it. They are not. Organizations drown in data and still make poor decisions. Information has diminishing value. Expertise creates new value, and the difference between them is widening at exactly the moment AI makes information available to everyone.
This is why a great deal of confident commentary about AI has the logic backward. It runs: AI can now perform intellectual tasks once reserved for highly trained people, therefore Expertise is becoming less valuable. The premise is true. The conclusion is closer to the reverse of the truth. Information was scarce once, and it stopped being scarce. The Expertise to know what to do with it did not.
In 2016, Geoffrey Hinton — the computer scientist most responsible for the architecture underlying modern AI — announced that we should stop training radiologists because deep learning would handle the job within five years. What happened instead: radiology caseloads rose and average radiologist compensation reached $571,000 in 2025 — up nine percent year over year, according to Medscape’s Physician Compensation Report 2026. Medical students heard the forecast too. In a 2022 survey of 463 U.S. students by Reeder and Lee, one in six who would have ranked radiology first chose otherwise because of AI — so a prediction of abundance helped produce a shortage. The scarce thing was never the image, and the cost of saying otherwise fell on the people who listened.
Consider two investors looking at the same deal, with the same data and the same AI systems. One sees the opportunity. The other sees the danger. Years of consequence have changed what the first investor notices, what they weigh, how they read a risk that looks manageable on paper. The same holds for two surgeons with the same imaging and the same patient — and only one of them has seen this before.
Where the novice sees facts, the expert sees relationships. Where the novice sees an event, the expert sees a pattern. Where the novice sees a decision, the expert sees its consequences three moves out. That capacity rarely appears on any balance sheet, and it silently shapes nearly every decision that matters. Almost every outcome of consequence carries an Expertise component, and Expertise has so far proven stubbornly resistant to automation, because context is rooted in Expertise and Expertise takes time to develop.
For most of history, that requirement set a hard limit. Expertise took years to build, and once built, its reach was bound by the body that carried it. An experienced executive could shape only so many decisions; a master physician could see only so many patients. The most knowledgeable people in any field were also its tightest bottlenecks, because there was one of them and a day held twenty-four hours.
AI breaks that constraint by extending the reach of Expertise rather than replacing it. The familiar argument is about humans versus machines, automation versus jobs, and it is a real argument. But it obscures the more consequential reality: amplification, when accumulated Expertise acquires scale and can operate through a system that runs at digital speed, around the clock, without tiring. The expert whose influence was bound to the decisions within reach can now shape far more of them. The constraint that bound Expertise for centuries has broken — and it has broken in the direction of the expert.
This is what reorders the economics. Access to AI is becoming universal; Expertise is not. AI levels access — it does not level Expertise. And scarcity has always been the thing markets reward and economic value concentrates around. As intelligence becomes abundant, the scarce complement to it becomes more valuable, not less. That is how scarcity has always behaved.
So, the gap has a name now. It is not the gap between humans and machines, or between one model and the next. It is the gap between access and outcomes — between intelligence, which is becoming universal, and Expertise, which is not. Hidden inside it is the explanation for why some people and institutions are creating extraordinary value with AI while others, holding exactly the same AI models, create almost none.
And hidden inside that explanation is a much larger economic story. A story that does not begin with AI. It begins with scarcity — and scarcity is beginning to move.
Before you read another chapter — prove it.
What you have just read is the argument. The next thirty minutes are the proof — and you are the one who runs it. Bring one problem only you are equipped to judge: not a question you could look up, but a decision that rests on your experience and judgment. Your Expertise, an AI, and a short sequence of prompts designed to show you, not tell you, exactly what that Expertise can produce in partnership with AI. The people who have done it call it their Thirty.
The design of The Thirty is not arbitrary. Wolpe established that fear does not end through argument or patience — only through direct contact with the thing feared.
It ends through direct, graduated, voluntary contact with the feared object — in conditions low-stakes enough to allow the autonomic response to habituate. A fear is not reasoned away or waited out — it ends only when you move toward it and confront it. The Thirty is the application of that principle to this specific perceptual problem: thirty minutes, one real problem, the output visible before the time commitment is required.
The Bartlett limit cannot be dissolved by reading about the exponential. It dissolves through experiencing what the exponential produces — in your own hands, with your own Expertise, before anything further is asked of you.
There is one more turn in it. Bartlett’s limit describes an exponential observed from outside — too large to size, and therefore a thing to flee. In every revolution before this, the technology was that exponential, and it reached people only after decades of laying track, wire, road or fiber, and only through those with the degrees and the industry experience to build on it. AI waits on neither. Its infrastructure is already in place, it asks for no particular schooling and no particular field — after more than fifty years the computer, through AI, speaks English and every other language, so a person just talks or types — and you can start today on any idea you choose, personal or professional. The only limit on what comes back when you run the Elevator — the method Chapter Four sets out — is the Expertise you bring to it. Confront it with your own and there is nothing outside left to observe. What compounds from there is not the technology; it is your own Expertise — the combination of experience and judgment you earned — running at a rate no one can see until its results grow large enough, and that you will see in your own work before anyone else sees it. Now the exponential is you.
The next chapter can wait. Your Thirty cannot.
Touch or click below and find out why. "Return to the Book" drops you right back to this page.
The Thirty →The Scarcity Shift
Economic history is, in large part, the history of scarcity. Value collects wherever scarcity exists and drains away wherever scarcity disappears. Water costs almost nothing where it is abundant and commands almost anything where it is not. Land is priced by location, because certain locations are fixed and cannot be reproduced. Talent is paid in proportion to how uncommon it is. None of this is about usefulness — air is essential and nearly free; diamonds are useless to most lives and cost a fortune. The difference is never how much a thing matters. The difference is how scarce it is. And markets, relentlessly, reprice the world as scarcity moves.
The AI Economy is a repricing event, and to see why, start with a single fact: AI is making intelligence abundant. Work that once took a trained person hours now takes seconds. Research, analysis, drafting, coding, the retrieval of almost any information — all of it is collapsing toward instant, and toward free. A capability that was scarce for the whole of human history is becoming abundant in the span of a few years.
The mechanism is the oldest one in economics, and it has run the same way every time. When agricultural output expanded, food became less scarce, and value moved to whatever remained constrained. When manufacturing capacity expanded, physical goods became less scarce, and value moved again. When the internet expanded access to information, information itself became cheap, and value moved to whatever the flood of information could not supply. Each wave made something abundant. Each wave relocated the scarcity to higher ground. The AI Economy is the same event at a larger scale: intelligence — the ability to generate, analyze, retrieve, and produce — is becoming abundant, and the scarcity is moving to whatever intelligence alone cannot provide.
So, the central question of the AI Economy is not what AI can do. It can plainly do a great deal, and it will do more. The question is what becomes scarce when intelligence becomes abundant. The answer is the thing that was always underneath the intelligence: Expertise. The accumulated experience that decides what deserves attention, what a result actually means, which risk is real and which is noise, and what to do next. AI produces output. It does not produce Expertise — because the experience and judgment that constitute it are not information; they are the residue of years of consequence, and they cannot be downloaded into a system that has not lived them.
This is why the most common conclusion about AI is precisely backward. The reasoning runs that as AI grows more capable, Expertise grows less valuable. The opposite is true. When everyone holds the same powerful system, the system stops being the differentiator. What separates one outcome from another is no longer access to intelligence — it is the Expertise directing it: what gets asked, how the output is read, what gets refined, kept, discarded, and deployed. As intelligence becomes abundant, Expertise becomes more valuable. Expertise did not change. What it can reach did.
That single shift explains a great deal of what organizations are currently getting wrong. Technology, by itself, has never produced durable advantage. Advantage comes from combining the technology with something a competitor cannot easily copy — and AI is becoming as widely available as any technology before it, while the Expertise that directs it remains stubbornly, unevenly distributed. Two organizations can deploy the identical AI models and produce entirely different results, and the difference is rarely the models. It is the Expertise being amplified. An organization that buys AI to do its existing work faster gets efficiency — real, but bounded, and available to every competitor on the same terms. That is AI-Driven Cost Savings, and it converges: the capability sits inside a tool every competitor can buy.
Michael Porter named that pattern in Harvard Business Review in 1996: improvements every competitor can copy raise the bar for the whole industry and leave nobody ahead. Full citation in the References. An organization that uses AI to amplify its accumulated Expertise gets something different in kind: not the same work faster, but work that was not previously possible at all. That is an Expertise-Driven AI Partnership, and it is resistant to commoditization because its input cannot be manufactured — no course and no AI training converts a lack of consequential experience and judgment into more depth of Expertise.
The two are bounded differently, and the difference is arithmetic rather than emphasis. AI-Driven Cost Savings is capped at the cost available to eliminate. A firm cannot save more than it spends, so the size of the prize is knowable in advance and it shrinks as it is taken. Expertise-Driven value creation carries no corresponding bound. It runs on the combinations of Expertise Capital and the AI multiplying it — at the level of one operator, and again at the level of operators working together — and the combinations do not run out. Nothing in the mechanism sets a limit on it. And because what it creates is not a thing a competitor can buy, the profit accumulates in the producer for extended periods rather than passing through it. Expertise Capital is Expertise held as an asset, the first instrument Chapter Five sets out.
Both are competitively necessary, and they run on different clocks. A business that does not implement AI-Driven Cost Savings in a competitive domain is priced out by the competitors who did — quickly, visibly, and on terms everyone can see. A business that does not implement Expertise-Driven AI Partnerships to drive new value creation arrives in the same place slowly, through work it never won and never knew it was in the running for. Neither is the optional one.
The reported disappointment in AI returns is explained by this frame without blaming anyone. Companies measured time saved. Time saved is precisely the converging metric — it equalizes across competitors by construction. The disappointment is not that AI underdelivered. It is that the industry measured the layer that was always going to converge, and never measured the other one. That gives a chief executive a reason to act that is not an accusation: not you were wrong, but you measured the side that commoditizes.
There is a consequence to that difference the public conversation has been circling since 2023 without naming. Cost-saving investment and job loss are the same event. If the value being captured is time saved, the only place to book it is headcount. That is why every discussion of AI collapses into a discussion of employment: the industry sold the layer whose return mechanism is doing the same work with fewer people. The public is not being pessimistic. It is reading the business case it was shown.
The other investment inverts that by construction. Revenue from work a firm could not previously reach requires people to do that work. Where AI does the work, the human is out of the loop, which is exactly why headcount falls. Where a person’s Expertise drives the AI, it amplifies rather than replaces, which is exactly why the work expands. This is a description of two configurations, not a forecast about either — which matters, because every other answer to the jobs question is a prediction, and predictions get argued with.
The same asymmetry runs through the hours. AI-Driven Cost Savings is counted in hours removed. Expertise-Driven AI Partnerships show up as hours added, and added voluntarily — the Sophisticated State described in Chapter Four, which nobody rations.
One discipline governs how it is written. Nothing on either side is stated as a clean offset. AI-Driven Cost Savings creates job losses. Expertise-Driven AI Partnerships create new work, and new work leads to new jobs. Both are said plainly, and no arithmetic is promised between them: no count, no offset, no match. A reader who has lived through a layoff stops believing anyone who claims the numbers match.
The history is stated as history. About six in ten jobs Americans do today did not exist in 1940: David Autor and his colleagues counted the new job titles in the census from 1940 to 2018. It is the record of the computer, the internet and the late decades of electricity, not a forecast of this revolution, and it promises no count for what comes next.
Amplification is possible because, for the first time, Expertise can be deployed beyond the body that holds it. The senior consultant whose advice once reached only the clients they could sit with can now shape work far beyond them. The experienced executive can now take part in far more decisions than a calendar once allowed. The Expertise itself did not have to change for this to happen; its reach did — and when reach expands, the economic value of the Expertise expands with it. The market is no longer simply pricing the technology. It is beginning to reprice Expertise operating through the technology. Which is why the AI Economy is a Human Capital transformation as much as a technological one.
Repricing of that magnitude never stays contained. When the value of Expertise moves, the systems built around the old prices begin to adjust. Educational institutions reconsider what they are actually for. Organizations reconsider what talent is worth and where it sits. Investors go looking for the new source of leverage. Entrepreneurs find opportunities that did not exist a year earlier. This is how every prior scarcity shift has played out: scarcity moves, value migrates, behavior follows, and new economic structures form around the new reality. That process is already underway, and the earliest signs are visible in the simplest observations — that access to AI alone is not producing superior outcomes, access to information alone is not producing superior returns, and access to technology alone is not producing competitive advantage. Something else is required. Expertise — not because it is new, but because it has become leverageable in a way it never was before.
The scarcity shift turns concrete once you have watched your own judgment redirect an AI’s output in real time — which is precisely what The Thirty is designed to make visible.
Which raises the question the next chapter exists to answer. If scarcity is shifting toward Expertise, and Expertise can now operate through systems that run at digital scale, then the oldest limit on Expertise — that it was bound to one person, one conversation, one decision at a time — is breaking. What happens when Expertise itself becomes scalable?
Expertise Becomes Scalable
For most of human history, Expertise was constrained by biology — deployable only one conversation, one decision, one relationship at a time. Civilizations built institutions to push against that limit: apprenticeships, universities, management layers, consulting firms. AI now amplifies an individual’s Expertise beyond what any of them could reach, and carries it at digital speed.
Be precise about what moves, because it is not the Expertise. That stays with the person who earned it — undiminished, and not handed to anyone. What moves is its reach: the number of problems one person's judgment can be brought to, and the speed at which it arrives.
And where that judgment lands on real work carried by someone earlier in their own climb, a second thing happens. That person is doing consequential work under it, and consequential work is the only thing that has ever built Expertise. They are given nothing. They earn their own — and they carry it into a partnership of their own.
For the whole of history, Expertise could be applied in only one place at a time. It can now be applied in one place and grown in another, in the same moment. That is what it means for Expertise to become scalable.
This matters far more than the raw intelligence of the AI frontier models. AI does not create Expertise — it expands the reach of the Expertise that already exists, as railroads expanded the reach of commerce and the internet the reach of information. Expertise has always been valuable. What has changed is that it can now scale.
What Expertise Actually Is.
Anders Ericsson, Professor of Psychology at Florida State University, spent four decades studying expert performance across domains from chess to medicine to music. His core finding was that expert performance is not primarily a function of innate talent but of deliberate practice. Set aside the popularized ten thousand hours, which Ericsson spent the latter part of his career correcting, and what remains is more precise: Expertise is built through deliberate practice, not repetition, and the quality of the experience matters as much as the quantity. This book reads that quality as consequence — whether decisions carried real stakes. In the domains he studied, recognized experts had at least ten years of intense practice behind them before their pattern recognition became different in kind. That is the timeline of a career.
That timeline was never a law. It was an average, produced by how much consequential work a career supplied and how thinly hierarchy rationed it. The rate is what AI Partnership changes. When the deepest Expertise in an organization operates at altitude, what it produces is more work that carries consequence. So the years required are a function of how much consequential experience and judgment a person carries per day, and that number is now movable for the first time. This book calls the rate Expertise Velocity; Chapter Five sets it among the instruments, and Chapter Six puts it to work inside the organization.
Michael Polanyi, whose 1966 work The Tacit Dimension remains foundational in epistemology, named what Ericsson’s research implies: much of what Expertise produces cannot be made explicit. It lives in the practitioner — the physician who knows something is wrong before they can say what, the investor who feels the deal before the model confirms it. Polanyi called this tacit knowledge: the compressed residue of thousands of consequence-bearing decisions, stored in a form the practitioner can act on but cannot fully explain. Appendix A sets out both bodies of work.
This is the knowledge the AI Partnership carries forward. AI can process information at a scale no human can match. It cannot hold tacit knowledge — it has not lived the consequences that produce it. The professional who brings genuine Expertise to the partnership is providing the one input that makes the output categorically different from what any non-expert would produce with the identical system.
The Mechanism Named.
At Hawaii Pacific University, Tao An set out to reconcile two bodies of published evidence that appeared to contradict each other. His argument: what drives AI output quality is the Expertise of the person directing it, not technical proficiency with the tools — and task complexity is what determines how much that matters. Where the correct approach is already known, AI supplies it and differences compress. Where it is not — on complex tasks requiring deep domain judgment, contextual reasoning, and iterative refinement — AI functions as a cognitive amplifier that widens the Expertise gap, and what it amplifies is Expertise. Appendix A sets out the paper and its limits.
Figure 1 (reproduced). Conceptual diagram. An, T. (2026). Hawaii Pacific University. arXiv:2512.10961.Nearly all published productivity research to date has sampled well-structured work — the work behind the AI-Driven Cost Savings the AI industry has marketed universally. That is why this effect was largely absent from the published record until the work examined below.
The Population Identified.
In March 2026, researchers at the McCombs School of Business at The University of Texas at Austin, in partnership with KPMG, published in Harvard Business Review the results of an eight-month study of 2,500 professional employees at one of the four largest advisory firms in the world. They tracked 1.4 million AI prompts and responses and defined sophisticated use rigorously, against data, and the credit for that definition belongs to them.
Sophisticated users, they found, engaged AI ambitiously across complex, multi-step work — treating it as a general cognitive tool rather than a narrow productivity shortcut, and holding a mature mental model of it as a dynamic collaborator.
A Sophisticated AI Adopter is this book’s own term and it names something narrower: a person who talks and works in partnership with AI the way they would with the most engaging and widely knowledgeable individual in every subject and computer program in the world, bringing it their most demanding work and pressing every answer until it reaches the ceiling their own Expertise sets rather than the one the AI Partner sets. The Elevator, the method Chapter Four sets out, is how a person gets there. The researchers described a level of use they found in a dataset. Sophisticated AI Adopter names the method, the practice, and the person who works at it.
Adoption here means none of what the word usually means. Not an account, not access, not use. It means having taken up the practice.
Before the study began, nearly ninety percent of the firm’s employees were already using AI regularly. That is the number the world reports when it reports on AI, and it says nothing about what comes back. After eight months of sustained encouragement and active institutional support, five percent of the 2,500 were judged to fit the researchers’ definition of sophisticated use. The other ninety-five percent were using AI as little more than an enhanced search engine.
The results are consistent with a single conclusion: the thing separating the amount of real AI value creation between sophisticated users and enhanced search-engine users is not something an organization can supply. Not tools. Not training. Not encouragement. The one input no organization can hand its people is accumulated Expertise — and that is precisely the input Tao An argues AI amplifies.
The researchers called one finding “surprising”: the best users were often those above manager level. Those above manager level are a small share of a firm’s people, which makes them disproportionately represented. It is not surprising in light of Tao An’s argument that output tracks domain expertise and that depth is its measure, and the UT/KPMG data is consistent with that at population scale. Appendix A sets out the study and what it does not establish.
The Record Observed.
In February 2026, Science published the largest record yet of what AI does to working output over time: more than thirty million code contributions on GitHub by 160,097 developers in six countries, tracked from 2019 through 2024. The experienced developers captured nearly all of the measured gains, and they took up libraries they had never used before, which the authors read as moving into new areas of programming. The authors’ own conclusion is that the technology widened the gap between skill levels rather than closing it.
“Experienced programmers capture nearly all of the productivity and exploration gains.”Figure 3D (reproduced). Daniotti, S., Wachs, J., Feng, X., & Neffke, F. (2026). Science, 391(6787), 831–835. Reproduced from the authors’ preprint, arXiv:2506.08945.Everyone in that population was already technically fluent; they write software for a living. The prevailing explanation for who benefits from AI is comfort with the technology, and every person in this population has it, so it cannot be what separates them. The record runs the assumption backwards: the newest developers used AI the most, the most experienced used it least, and the gains went to the ones who used it least — including, as the authors read it, the move into new areas of programming, new value creation rather than the same work done faster. Tao An proposed exactly that. Appendix A sets out how the record was built.
Each of the three has limits — a framework, one firm, one profession — and Appendix A sets them out. The test that matters most is the reader’s own Thirty, run in their own field.
Why Cognitive Style Matters.
In addition to depth of Expertise, the second less obvious personal characteristic affecting the output quality of an AI Partnership is cognitive style. It explains why two professionals with identical years of experience can produce dramatically different outputs from the same AI Partnership — and why some of the most experienced people in an organization are, by natural disposition, less likely to reach the highest Floors the Elevator can carry them to.
The distinction is between divergent and convergent thinking — in the language of the most widely used personality assessment in organizational settings, Intuitive versus Sensing processing styles, and Perceiving versus Judging orientations. People who score toward Intuition have been linked to creativity since Donald MacKinnon’s studies in the 1960s, and Robert McCrae and Paul Costa later found the Intuition scale closely tracking Openness to Experience; the link runs mainly through Intuition rather than Perceiving. The scales are continuous, most people sit between the poles, and the labels are used here as a vocabulary for two ways of working rather than as boxes. Convergent thinkers process sequentially, concretely, toward closure. They are extraordinarily valuable in execution, and pyramid organizational structures that prize reliability, predictability, and adherence to process have historically promoted and rewarded them.
The AI Partnership is a divergent task. It rewards the professional who can range widely across a problem space, tolerate ambiguity, make connections between domains that are not obviously adjacent, and stay open rather than close prematurely — preferring a better answer over a faster one.
AI is a powerful convergent partner, capable of synthesizing immense amounts of information into coherent outputs. The highest levels of performance come when it is guided by a divergent human with significant Expertise, who asks better questions, recognizes subtle distinctions, and shapes the direction of the collaboration.
The further implication, offered here as this book’s own reading rather than a finding, is that the confidence to operate divergently can deepen with accumulated proof: a professional with decades behind them has evidence that their divergent instincts pay off. The workplace research is consistent with it: a meta-analysis of nearly a hundred studies by Thomas Ng and Daniel Feldman found that older and longer-tenured employees do not engage in less innovation at work — even though scores on laboratory tests of divergent thinking decline after about forty, as Robert McCrae and colleagues found.
The research does not support the widely held expectation that digitally native younger workers would be the ones producing the highest-value AI output, and anyone reading the research in this chapter could have predicted as much. Domain Expertise and cognitive orientation at the organizational level cannot be supplied from the bottom up.
The Sophisticated AI Adopter.
Sophisticated AI Adopters are not programmers, data scientists, or AI researchers. They are entrepreneurs, investors, executives, physicians, engineers, educators, consultants, writers, business owners. What distinguishes them is what they bring to the technology, and how far they are willing to take it. Those are two different things and they do not travel together — the depth is earned over years and sets how high the output can go; the way of working is available at any depth.
The popular narrative assumed Expertise would matter less in the AI era. The data point the other way. Expertise matters more — precisely because it has become scalable.
The Elevator
The first three chapters established what happened, why it was possible, and what became scalable. What remains is how — and the how is a method.
So, this is where the book stops describing and starts addressing. The man in the chairs has done his work as a case. This chapter marks the shift from evidence to application — from examining what happened to doing it yourself.
The method is called the Elevator. It first appeared in Why Us? and it carries over to this book unchanged, because the two books describe the same shift and the shift has only one bridge. The Gap established in Chapter One — universal access, uneven outcomes — is not closed by training budgets, software purchases, or mandates. It is closed one person at a time, by the method below. Everything this book argues about organizations, capital, and careers depends on individuals crossing that bridge. This chapter is where you cross it. Get on believing now the exponential is you, even though you cannot conceive it — nobody can, which is what Chapter One demonstrated. The proof shows itself on the way up, more than once.
First, what the Elevator pushes against. Every frontier AI model arrives calibrated to the average person who might be sitting in front of it. Its default caution — the hedges, the qualifiers, the reflexive you may want to consult a professional — is not a flaw. It is a reasonable setting for an unknown user. But it means that every AI, on first contact, is treating you as an unknown rather than as whatever you actually are. The Elevator is the structured way to replace that unknown with the truth, and to keep replacing it until the answers coming back are calibrated to your actual Expertise instead of to a statistical average.
There are only three Moves. No courses. No certifications. Nothing to learn about agents or tools — nothing special about AI at all. The Elevator runs on consecutive hours with an AI Partner, not minutes — and it strains hardest breaking from the Ground Floor, before the climb turns smooth.
Move One — Ground Floor. Choose the domain you are boarding — the field where your Expertise actually lives — and bring a real problem from it, described at its full imaginable size rather than the conservative investment committee or board version. The problem is yours, and it does not have to come from your work — a business you have thought about for years, a dispute, a claim that was denied, something difficult you are teaching someone else. All of it rides the same three Moves. Full size is about how you describe the problem, not how large it is: on your first rides it need not be the largest thing you hold. Start where your footing is surest and let the size of what you bring grow as the method becomes second nature — the Elevator starts you where you are already strong. For now, nothing has to be overcome. You are standing on the Ground Floor with a real problem already aboard, and that is the whole of Move One. It is the easy part.
Move Two — Breaking from the Ground Floor. Now you engage, and the moment you do, the resistance arrives. What comes back is your AI Partner’s full inventory of caution: every objection, every reason the idea is too large or too fast. That is not the answer. It is the map of the static between you and the next Floor up — the calibration described above, running exactly as designed. The way past it is to tell your AI Partner what it does not yet know about you. Give it an accurate picture: the years inside the rooms it just warned you about, your history in the work it just called risky. Be accurate. Boast or fabricate, and the Elevator runs in reverse. It rises only as far as the depth of the real Expertise you bring into it. This is the hardest stretch of the ride. Breaking free takes more push than any Floor above it, because you are lifting the full weight of that default caution from a dead stop, and it will feel like effort in a way the smooth climb above never does. Most people who give up give up here, in the first laboring minutes, before the ride has the chance to turn. Stay with it.
Move Three — Climbing Floors. Now the work changes character. The default caution still returns, but softly — a stray qualifier, a sentence that begins one thing worth flagging when nothing needs flagging. Move Three is pushing the lever over and over: name the hedge, state that you have accounted for it, and continue — and your AI Partner adapts faster with each iteration. The push that took everything at the Ground Floor now takes almost nothing; you did not get stronger, the ride got smoother. But smoother is not flat. The climb comes in pulses — easy stretches broken by moments where the work turns hard again, and those moments are where the largest jumps happen. A hard moment reached with real Expertise is a door to a Floor you could not have seen from below; the same moment reached without it is only a wall. This is why the Floor your Expertise can support is not fixed when you board: the ride is real work, real work teaches, and something can arrive on the way up that you could not have imagined at the Ground Floor — yours to use on the next Move Three. Refuse to stop before reaching the Floor your Expertise can actually support. Move Three is where the consecutive hours accumulate; it is also where they stop feeling like hours.
That last observation — that the consecutive hours stop feeling like hours — describes a documented psychological state, and this book gives it a name: the Sophisticated State.
Mihaly Csíkszentmihályi, Distinguished Professor of Psychology and Management at Claremont Graduate University, spent decades studying the conditions under which human beings enter states of peak engagement. He called the state flow: total absorption in a challenging task that stretches but does not exceed the performer’s capabilities.
Flow states are characterized by loss of self-consciousness, distortion of time perception, intrinsic motivation, and — critically — the desire to repeatedly return to the activity that produced them. Csíkszentmihályi’s research showed that flow is most reliably produced when the difficulty of the task is precisely calibrated to the skill of the performer. Too easy produces boredom. Too hard produces anxiety. The narrow band between them produces flow.
The Elevator, operated by an experienced professional at genuine depth, meets the conditions Csíkszentmihályi’s research identified for flow. The AI handles what the operator cannot yet do. The operator supplies what the AI cannot produce. The collaboration is built to operate in the band between boredom and anxiety that generates peak engagement — because the system is always calibrating to the operator’s actual level, and the operator is always pushing the system to a level just beyond what was previously possible.
The desire to keep working this way is the Sophisticated State, and the structure of the partnership is built to produce it. That is why the hours can be sustained rather than endured. No study has measured it in anyone riding the Elevator. The flow research describes the conditions; whether they produce the state in you is a question your own next long session answers.
There is one way to waste the whole climb, and it is not pushing too far. It is stepping off at good enough. The hard part is already behind you — you paid it breaking from the Ground Floor, and the Floors above it are nearly free to reach. Good enough is a verdict your mind reaches by measuring the result against what it expected, which is exactly why it cannot see the Floors above it, and exactly why it tells you to get off. The result that would merely satisfy you is the one you could have predicted. The one you cannot yet imagine is higher, and the ride to it now costs almost nothing. When the work turns hard again, that is almost never the ceiling — it is the next door, and your Expertise is what opens it. The waste is paying the full price of entry and stepping off one Floor short of what it bought.
A word on who can operate it. The Elevator rises on Expertise, and it rises exactly as far as the real Expertise brought into it — which means it works for everyone and stops at different Floors. A thirty-year operator reaches Floors a student cannot yet reach. The student still rises — and the student’s top Floor moves up with every year of Expertise earned, and with every ride that teaches them something they did not board with. The method is a multiplier, not a credential, and it multiplies whatever is genuinely there. That is why this book’s audience runs from students to chief executives: the instruction is identical at every level of input.
One caution belongs with that, and it concerns the wildly inaccurate output AI is known to produce at times. It can produce it for anyone. What decides whether it does harm is whether the person driving the AI Partnership has the experience and judgment to catch it — and that runs out exactly where what is asked goes past what they know.
In most work, the signal that you have gone past your competence is that the output starts to look rough — the draft reads badly, the model does not balance. AI Partnership removes that signal. What comes back is polished regardless of who asked for it, which means the page will not tell you when you have exceeded what you can judge. Only your Expertise can. Step off only when you can no longer judge — not when it merely turns hard, and never while it still glides. The last Floor reached on The Elevator is yours to determine and no one else’s.
The mechanics underneath the three Moves are not new. Researchers have studied role prompting, iterative refinement, and pushback-based improvement for years, and any reader who wants that literature will find it. What is new is the Elevator: a structured method built to extract systematically better and higher-value answers from an AI Partner by using the operator’s own accumulated Expertise as the primary force on the lever. The literature has measured the techniques. It has not measured this — the depth of Expertise itself as the engine.
It is also, structurally, the answer to the question the UT/KPMG Study in Chapter Three leaves open: why a firm that gave its people the tools, the training and the encouragement saw so few of them fit the definition of sophisticated use. Inside a professional firm the price of presenting lower-quality AI output is high — and at lower levels of domain expertise, that is what pushing the AI hard tends to produce. Seen in that light, a small share fitting it is no surprise. On this reading, what was missing was a method that converts Expertise into altitude. The method is the same three Moves for everyone. The Floors are not — they belong to the Expertise each user brings.
When The Elevator result comes back — and it will come back larger and faster than anything you would have produced alone — someone will ask you a question. They will ask it in many different ways. Did you come up with that, or did the AI?
The answer is short and honest.
Your AI Partner did the thousand things you could never have done alone, or even with help, in the time available — the coding, the modeling, the research, the synthesis, the speed. All of it. And you did not have to learn a single thing about any agent or tool to make it happen. You brought your Expertise, your will, and your refusal to stop at a Floor your Expertise told you could exceed.
Do not take this chapter’s word for it. Try it now. It helps to begin with ideas matched to how you like to work, so that your iterative Move Threes are fueled with both desire and persistence. Touch or click the gold box and select the one that sounds most like you; it will return four general THEMES from which to construct your first meaningful Move One ideas. The sixteen are a way in, not a verdict on how you think; as Chapter Three says, most people sit between the poles and will see themselves in more than one. Ride all four. Not a course, not a certification — four real problems taken up the Elevator with the Moves in hand. By the fourth you will not be following a method. You will be operating one.
“RETURN TO THE BOOK” drops you back exactly here.
The Thirty exists for a reason. It asks almost nothing — thirty minutes, light lift, engineered to produce the first WOW! before any time commitment is required. That moment of belief is the point of it. The Elevator asks for something different in kind: your real Expertise, your consecutive hours, your refusal to stop at the first Floor. The Thirty shows you what the partnership can produce. The Elevator is how you produce it from then on, on anything you bring it. One is the door. The other is what lies behind it — and Move Three, repeated as many times as your Expertise can support, is what keeps lifting you to higher Floors.
The rest of the book prices what a lifetime of Expertise becomes when the Elevator puts it to work at scale.
The Expertise Economy
The previous chapter handed a method to an individual. This chapter hands a framework to everyone deciding what that individual is now worth.
Expertise, operated through the Elevator, produces something the prior economy had no framework to measure: an economic object, an organizational asset, an investment thesis. Naming it takes five instruments. All of them are already operating, unnamed, in the organizations and markets that are pulling ahead. They are given names because the asset had no vocabulary, and a market cannot price what it cannot name.
They are presented in the order a reader encounters them — as an individual, a team, that team over time, an organization, a market: the same asset, seen from increasing altitude.
Expertise Capital
Every experienced professional carries something that does not appear on a balance sheet. It is not their credentials, their title, or their network — though all of those are proxies for it. It is the accumulated residue of years of consequence: the pattern recognition that tells them which risks are real and which are noise, which opportunities are genuine and which are well-packaged mediocrity, which decisions will look obvious in three years and which will look expensive. Economists have tried to name this for decades. Human Capital comes closest but undersells it — Human Capital is the stock of knowledge and skills; what is being described here cannot be taught, only earned.
This book calls it Expertise Capital. And the defining fact of the AI Economy is that Expertise Capital has just acquired a new operating environment.
For the whole of economic history, Expertise Capital was illiquid. It lived in a person. It moved at the speed of human interaction. It depreciated at death. It could be partially transmitted through apprenticeship, partially codified in writing, partially embedded in organizational culture — but always imperfectly, always slowly, always losing something in translation. The limit was not the value of the asset. The limit was its delivery mechanism.
AI is the first system through which Expertise Capital can operate at a scale and speed the person who holds it could never reach alone. The senior partner whose judgment once reached only the engagements they could personally sit in now carries into work they will never attend. The Expertise Capital is identical. What changed is how far it reaches, and what it is now free to take on.
The economic implication is direct. When the output of an asset rises sharply without a corresponding increase in the cost of the asset, the asset reprices. Expertise Capital is being repriced upward, right now, in every organization and market where someone has learned to operate it through AI.
Unlike every other asset a company holds, it was paid for over the working lives of the people carrying it, so when it reprices, nothing has to be purchased. The asset was inside the building on each morning the long-standing technical barriers between humans and AI came down another step, and it will still be inside the building on the morning a competitor acts on that fact first.
The Elevator is the mechanism through which Expertise Capital stops being an individual asset and starts operating at scale — which is why the depth of the operator determines the altitude of the output.
Expertise Density
Not all organizations hold equal amounts of Expertise Capital. This has always been true and has always mattered. What is new is that the difference now compounds differently.
Expertise Density is the concentration of operating Expertise Capital within an organization, measured per head. Two terms decide it: how much depth each person brings, and whether that depth is being multiplied at all. Depth that never operates counts for nothing, which is why an organization can hold a great deal of Expertise Capital and still have low Density. A law firm of two hundred partners where eighty percent have thirty or more years of practice holds a great deal of Expertise Capital; whether it has high Expertise Density depends on how much of that depth is operating. A consulting firm that has hired aggressively for credentials but lightly for judgment has the appearance of Expertise Density without the substance — credentials are a proxy for judgment, and the AI Economy prices the judgment.
In the pre-AI economy, Expertise Density was valuable but bounded. A high-density organization could charge more per engagement, win higher-quality work, and attract better clients — but the density itself was a headcount problem. More Expertise required more expert people. Growth diluted density unless hiring kept pace. The density ceiling was set by the labor market.
The AI Economy breaks that ceiling. When each holder of Expertise Capital operates at a multiple of their prior output — and the multiple does not have to be large to change the arithmetic — the Expertise Density of the organization rises without adding a single person, and the gain multiplies: each person’s output rises, the organization’s effective capacity rises, and the quality of work attracting further work rises with it. The organization pulls further ahead, faster, than headcount arithmetic would predict.
The Sophisticated AI Adopters inside a high-density organization are not just the highest performers — they are the advance signal of what the rest of the organization becomes as the practice spreads. Their output today is the organization’s average output tomorrow, if the platform is built correctly. Density is a level. The rate that moves it is the one this book named in Chapter Three — Expertise Velocity, the next instrument.
Expertise Velocity
Expertise Velocity is the rate at which new Expertise forms. It is the one instrument in this chapter that describes motion rather than position. Expertise Capital is the asset. Expertise Density is how much of it is operating, per head, at a point in time. Velocity is how fast consequential work is reaching the people holding less — and where a level says where an organization stands, a rate says where it is going. Chapter Six sets out how a firm can take a first reading from data it already holds.
Until now the rate was treated as a constant, because the structure fixed it rather than the person. A career supplied a limited number of consequential decisions, and the pyramid rationed how many of them reached anyone below the top. The years it took to earn Expertise were the visible result of that rationing.
AI Partnership moves the rate by creating work that did not exist. A person holding deep Expertise, operating the Elevator, generates new value-creating work — a new line, a new pursuit, a problem the firm never had the people to take on — and more of it than they can carry. Running it takes someone else, working through their own AI Partnership — usually someone holding less Expertise Capital of their own. That person is inside new work whose decisions have never been made before, so the decisions come with it. They meet consequential decisions sooner and in greater number than the position they held before could have supplied, and their Expertise Capital grows faster for it.
None of this leaves them alone with it. The person who handed the work over, or someone holding the same depth, has the time to stay with the work, because their own AI Partnership freed it, and can catch a misjudged decision before it runs far in the wrong direction. That is ordinary management, carried into work that now moves faster. It also makes Velocity the instrument an organization controls most directly: it starts to move the day a firm aims its deepest Expertise at work it has never done.
The Expertise Platform
A platform, in economic terms, is a system that enables others to create value on top of it. An Expertise Platform is a system that enables an organization’s Expertise Capital to operate at scale — that takes the individual gains of the Elevator and compounds them across a team, a practice, a firm.
The Expertise Platform is not a software product. It is an organizational architecture. Its components are three, and the order matters.
The first is the method — which this book has already handed the reader, in the previous chapter. Every person in the organization who has reached Sophisticated AI Adoption is a node on the Expertise Platform. Operating the Elevator below that standard produces good work; it does not produce a node.
The second is shared context. A collection of individuals each using AI in isolation produces the individual gains without the organizational gain. The Expertise Platform requires that Expertise Capital become partly legible across the organization — that the frameworks, the tested approaches, the hard-won pattern recognition of the most experienced people travel further than the conversations those people are in. The aim is to make Expertise Capital contagious.
The third is feedback loops. An Expertise Platform compounds only if it learns. Organizations whose Sophisticated AI Adopters share what they are discovering — which methods produce higher Floors, which domains amplify most powerfully, which questions the AI cannot yet answer well — turn individual gains into organizational knowledge. The platform’s intelligence rises with use. The gap between an organization that has embraced Sophisticated AI Adoption from one that has not widens with every cycle.
Individual Sophisticated AI Adoption is a moment — a threshold crossed once and then available for every Expertise-Driven AI Partnership after it. What AI multiplies is Expertise, and the multiplication is immediate: from the first ride, whatever a person has already earned is operating at a scale it could never reach alone. That is an exponential node, and nothing further is required for it. Expertise also tends to be gained on the way up, and whatever is gained is available to be multiplied on the same ride — a second benefit of the climb, never a condition of it. The Expertise Platform is the organizational system of those nodes, and systems compound as well.
The Human Capital Thesis
The four instruments above describe what is happening inside organizations. The Human Capital Thesis is what happens when markets price it.
The argument runs as follows. In every prior technology transition, the asset that became scarce as the technology became abundant was the thing markets learned to price. When manufacturing became abundant, distribution became scarce, and markets priced logistics. When information became abundant, attention became scarce, and markets priced media and brand. As intelligence becomes abundant, the scarce complement is Expertise — and markets are beginning, unevenly and imprecisely, to price the Expertise operating through the intelligence.
This repricing is not yet legible in most valuation models. The frameworks used to price organizations were built in an era when Expertise Capital was illiquid and therefore uncountable — when the best proxy for it was headcount, credentials, tenure, or revenue per employee. None of those proxies measures what a twenty-person team of high-density Expertise Capital produces through an Expertise Platform, and the gap between the old proxy and the new reality is where mispricing lives. And a market pays for direction as well as level: two firms holding the same Density today are different assets if one of them is forming Expertise faster than the other.
The investors who find that mispricing earliest are making a bet on a human asset rather than on technology — Expertise Capital, newly liquid, newly scalable, still largely unmeasured. The organizations that price it correctly inside their own walls, build Expertise Platforms around them, and develop the density to compound it will outperform in ways their income statements will show before their valuations catch up.
A second thesis runs alongside it through this book, and it is worth naming here before the next two chapters put both to work.
The Sophisticated AI Adopter Thesis is behavioral. It explains who captures that repricing — and who does not. The evidence suggests that AI does not merely favor the experienced professional over the inexperienced one. It actively produces opposite outcomes: amplifying the judgment of those who have it, and returning the average answer to those who do not.
The first thesis explains why the economics are changing; the second explains who benefits. The two chapters that follow price both — for the executives, entrepreneurs, investors, and professionals who are deciding, right now, what to do about it.
Early in this book, the reader was asked to carry one picture: new gold, piling up around everyone in proportion to their Expertise, and invisible to most. Once a person sees what their own pile has become, they cannot stop seeing everyone else’s. Picture a chief executive who has run the Elevator on the largest problem he holds, walking back through his own building. He already knows what the people in it have decided and lived with. Now he sees what all of that has become: new gold, piled around every one of them, low for some and high for others, none of it on the balance sheet and most of it not in production. Nobody hid it from him. He could not see anyone’s until he had seen his own. What a firm does about it is the next chapter. The question he carries into it is simpler: Why would anyone let any of it walk out the door again?
The Organization
The argument so far has been about one person with a laptop, and nothing in it depends on who he is — only on what he brought. But most of the world’s consequential work happens inside organizations, allocated by people who manage budgets, not laptops.
The question for an executive or capital allocator now is what AI is doing to the economics of organizational structure — and specifically, what happens to the traditional pyramid when the person at the top of it can now produce what the pyramid used to produce.
The answer reorganizes everything beneath it.
What the Pyramid Was For
The organizational pyramid was the rational response to a real constraint: Expertise was illiquid. A senior leader’s judgment could only travel as far as the conversations that leader could be in. Organizations solved that problem by building downward — layers of people whose job was to carry Expertise into places the senior person could not reach, to process the information the senior person could not see, and to produce the output the senior person could not generate alone.
The pyramid was an Expertise delivery system. Junior staff gathered raw material. Middle layers filtered, synthesized, and prepared it. Senior people applied judgment at the top, where the leverage was highest. The whole structure existed because the scarce input — experienced judgment at the apex — could only influence the work it could directly touch.
That constraint no longer holds.
The pyramid’s failure mode was not obvious while the pyramid was the only available option.
Emilio Castilla, Professor of Management at MIT Sloan School of Management, and Stephen Benard, Associate Professor of Sociology at Indiana University, identified what they called the merit paradox in their 2010 paper published in Administrative Science Quarterly: in organizations that most explicitly embrace meritocracy as a cultural value, the gap between performance and reward tends to be larger, not smaller.
The explicit commitment to merit as an ideal appears to reduce the vigilance with which actual merit is assessed. The performance signal is attenuated at every layer between the person who produced it and the decision-maker who should be responding to it. Each layer applies its own filters, its own risk preferences, its own interpretive frameworks — and passes on a slightly degraded version of the original.
This is what the pyramid extracted from its most experienced people. Not their capability — that remained entirely intact. What it extracted was fidelity. The judgment at the apex was being applied not to the raw material of a problem but to a processed version of it, filtered through every layer of the structure below.
The AI Partnership does not pass work through those layers. The operator’s judgment applies directly to the raw material. The translation tax — the signal lost at each layer the work passes through — thins with every layer the AI Partnership takes out of the path: a structural change in the quality of what Expertise can produce when the pyramid is no longer its only delivery mechanism.
It also changes what a firm can see. As the layers between the person who did the work and the person deciding about it thin, contribution becomes legible — which is the precondition for every other change in this chapter being actionable.
When Expertise can operate through AI at digital speed and scale, the senior person can now touch far more work than the pyramid was built to handle. The Expertise that used to require a pyramid to deploy now requires a MacBook Air. What limits the work is the Expertise directing it, not the desk it is directed from.
And the unit that can do serious work has gotten smaller. What set the floor on the size of a working team was that turning judgment into finished work took several people. It no longer does. That does not shrink firms by itself. It makes small ones viable where they were not, and it puts a cost structure in the market that a firm carrying the old overhead has to answer.
One firm has already run the experiment at scale, and it ran it before AI. Haier, then the world’s largest appliance maker, divided itself into more than four thousand self-managing microenterprises, most of them of ten to fifteen people — roughly two hundred and fifty of them facing the market and the rest supplying those with components and services — and runs a global business with two layers of management between the front line and the chief executive. Gary Hamel and Michele Zanini documented it in the Harvard Business Review in 2018. It is not evidence that AI causes this, and the timing rules that out. It is evidence that the structure works at institutional scale, reached the hard way by a company that had to dismantle its own hierarchy to get there. What changes now is the difficulty: when turning judgment into finished work no longer requires a team, the small unit stops being an act of corporate courage and becomes the ordinary shape of the work. The question facing a firm of any size — a hundred people or a hundred thousand — is not whether those units form. It is whether they form inside the company or outside it.
What the Best-Resourced Organization in the Study Got
The UT/KPMG Study introduced in Chapter Three is the clearest available reading of what a well-resourced organization actually gets. Access, tools, training and institutional support were shared by the whole population, so none of them is what separates the five percent from the ninety-five. This book’s reading is that accumulated Expertise does, operating through an AI Partner by the Elevator — a method any organization can put in front of its people, and none can install for them.
The Expertise Density Reckoning
This book’s reading is that most organizations look like the study from the inside. Most of their people are using AI as a slightly faster search engine. A few are doing something categorically different. And the gap between those two groups is widening — not because the few are working harder, but because their output is compounding and everyone else’s is not.
The organizations that have not yet named this are accumulating a density problem. Their Sophisticated AI Adopters are pulling ahead internally, and the question their performance poses — why are these people producing multiples of what the others produce, with the same tools and the same hours — is not yet being asked at the level where the answer would change anything.
The Expertise Density reckoning is a talent and structure question. Which people in this organization carry genuine Expertise Capital? How many of them have reached Sophisticated AI Adoption? What is their effective output multiplier — and what would the organization’s capacity look like if that multiplier applied to twice as many people? A board will put the same questions more simply: whose Expertise, inside this building and outside it, is not at work, and why?
AI training and implementation investment to date has mostly gone to competitively necessary AI-Driven Cost Savings — the industry sold it, and Corporate America bought it, in software and training. That was the first move. It was the right call, every competitor made it, and it is also, by construction, the part that will not distinguish anyone. The second move is permission rather than procurement, and it is incremental to the first: new value alongside it, not instead of it.
Building the Expertise Platform
The individual Sophisticated AI Adopter is the node, and the Expertise Platform is the network. Nodes are not graded — a person is one, or is not. Counting nodes is the same mistake as counting heads: twenty nodes tells you nothing until you know what those twenty hold and where they sit in the building. What matters is Expertise Density, and the direction it is moving. The compounding also happens between the nodes: what a node operating at altitude cannot carry alone becomes the work of a node with less Expertise depth, whose own Expertise Velocity rises on what it now receives.
Building the Expertise Platform inside an organization has no software to install, no vendor to procure, no system to integrate. It has three components, and the sequence matters.
The first is identification. If this book is right, almost every organization already has Sophisticated AI Adopters — the people who, knowingly or not, have been operating the Elevator for months. The study found its most sophisticated users often above manager level — disproportionately so, given how few people sit there. They tend to be the people whose output has quietly become inexplicable by prior standards, and the people other senior leaders go to first when something important needs to happen fast. Find them. They are the seed of the Expertise Platform, and they are already inside the building.
There is also a way to find them in the data. The researchers identified their five percent from the prompts and responses themselves. A company holding its own AI deployment and usage data can ask the same question of its own people, and it has something the researchers did not: it knows which of its people carry the most, rather than inferring depth from a title. The window for that first read runs back to early 2026, when the long-standing technical barriers between humans and AI came down far enough for ordinary use — and they are still coming down. That first read is an initial indication of who the firm’s Sophisticated AI Adopters are. Tenure is a usable starting estimate of that depth; what the data adds is which of that depth is actually operating. Those two readings together, averaged across a group, are its Expertise Density. Its Expertise Velocity is read differently: taken after any interval the firm selects — a month, a quarter, a year — it is how much sooner its less experienced people are carrying consequential work that used to sit above them. Read years in the domain rather than years at the company: the first is what accumulated the judgment.
The search should not stop at the building. Many of the people who could become nodes have already left — retired, bought out, stepped back — carrying the same domains and a large portion of the firm’s own history. They need no orientation to the industry and none to the company. What made them hard to reach was the cost of putting them to work: hiring judgment used to mean hiring the people required to execute it, and that is no longer the firm’s to carry. The pyramid needed retirement to make room at the top; a platform makes room by creating new work, and retirement keeps everything it promised. Nobody holds a claim on them, and a competitor can call as easily as a former employer can. Some will take on new work the firm has never done and run it as their own from the first week. Others will arrive with the business line nobody currently on the payroll has the standing to propose. Pursuing them deliberately, by name, is work an HR department can begin next Monday, and it requires no new function.
The second is elevation. The seed group’s methods need to travel. Not as a training program — training programs are what the study firm had already supplied, and after eight months, ninety-five percent did not fit the definition. What travels is direct exposure: the Sophisticated AI Adopter utilizing The Elevator on real work with real stakes, the work open to colleagues with comparable Expertise Capital. The method transfers in proximity and consequence, not in classrooms, and the proximity that matters is to the work, not to the desk.
The AI-Driven Cost Savings training already delivered was not wasted. On tasks of lower complexity, where the correct approach is well represented, these systems produce real and measurable productivity gains. But most of that curriculum was built against technical barriers that have since come down, and are still coming down. What follows is a training budget focused on encouraging the implementation and use of The Elevator throughout all layers of an organization.
A directed effort toward getting senior executives — those holding the highest levels of Expertise Capital in almost every organization — to themselves operate The Elevator as Sophisticated AI Adopters is the key step in the process.
The people holding the deepest Expertise in the building begin bringing their largest problems and opportunities into their own Elevators as Move Ones — chosen for what they would create for the firm, and including the ones the company has never had the time or resources to take on. The reasonable expectation is that what comes back is a longer list of things worth doing than there are people to do them.
How each firm runs that is not this book’s business. Almost no one has done this at scale yet, and the first firms to try will learn things nobody can write down in advance.
One thing is worth doing before any of it, and it costs nothing. Send every employee the links to both books, which are free to read online, as the working guide to what follows: the argument and the method together, and the same starting point for everyone. Anyone who wants to can start their own Elevator that week rather than waiting to be told.
The senior executive team is where it starts and not where it stops. A node operating at altitude produces more consequential work than it can carry, and if nobody below is operating there is nowhere for the overflow to land — a firm that converts only its seniors has built a platform with no receivers. There is a second reason, and it is selection: a senior person assembling a team to run a new idea expects more of it than the old work required, and picks from the people who already work this way. Bringing everyone to Sophisticated AI Adoption, at whatever depth each of them holds, is the objective rather than a courtesy. And what lands is new work, which carries its own decisions. The person receiving it is making them rather than preparing them for someone else, and that is where their own Expertise Velocity comes from.
The third is architecture. Once the Elevator is traveling, the organization needs structures that capture what it produces — shared frameworks, tested approaches, pattern recognition that does not disappear when the senior person leaves the room. This is the organizational decision to treat Expertise Capital as a collective asset rather than a collection of individual ones, and to build the feedback loops that make the platform smarter with every cycle.
The Question the Pyramid Cannot Answer
This question surfaces in every serious conversation about AI and organizations, and it deserves a direct answer here rather than a deferral.
If experienced people operating through AI can now produce what pyramids used to produce, what happens to everyone in the middle of the pyramid?
The honest answer is that this transition costs something. The layers whose primary function was to carry Expertise from the top to where it was needed are under real pressure — not because the people in them lack capability, but because the constraint that created the need for them has broken.
The organization that delays facing the question in the name of protecting the pyramid is not protecting the people inside it. It is guaranteeing that the adjustment, when it comes, arrives faster and harder than if the organization had led it.
An operator covering ground that used to take a team surfaces more work worth doing than there are people available to do it, and the binding constraint moves from ideas to hands. Under that constraint, people are the scarce input, and a firm looks for more, because the list of things now possible is longer than the list of people who can be assigned to them. None of this guarantees anything about employment in general, and a claim that broad would deserve the skepticism it drew. It establishes a direction, and the direction is testable inside one company in a single quarter.
What cannot be described yet is the shape of the organization that comes out of it. That is being invented right now, by people who do not know how it turns out — hybrids that hold for a while and then do not, structures tried and abandoned, a long stretch where the pyramid and whatever follows it run side by side. The uncertainty is about form. It is not about direction.
The work that gets absorbed is the bounded kind, where a correct approach already exists. What remains, and what keeps being created, turns on judgment — so the positions that follow carry more consequence than the ones they replace, and people reach that consequence sooner. That is what raises Expertise Velocity.
It has a second edge, and it is the one to watch rather than the one to fear. More consequential decisions running through AI Partnership means more of them can be amplified in the wrong direction, and output comes back polished whether or not the person was right to be asking. Amplification does not favor one side of it. The person with enough Expertise to recognize a plausible-sounding wrong answer is working through an AI Partnership too, and what they can catch rises alongside what can be produced.
The study is instructive here too. The best of the five percent who fit the researchers’ definition were often above manager level. Which means the people most at risk in the traditional pyramid are not the senior people — they are the junior and middle layers whose primary function was to extend the senior person’s reach. That reach has now been extended by a different mechanism, and the organizations that face that honestly, redirect their experienced middle toward Sophisticated AI Adoption, and rebuild their structures around Expertise Density rather than headcount will come through the transition with new work, and new work leads to new jobs.
The Founder, the Investor, and the Professional
Chapter Six priced the AI Economy for the organizations that employ professionals. This chapter turns to the three readers around them, one section each: the founder who builds alongside them, the investor who allocates capital around them, and the professional themselves — the person inside the organization, or considering leaving it, or building something new, or simply trying to understand what the next decade of their working life looks like.
The Founder
This generation’s defining advance is AI, and the entrepreneurs who understand it earliest are using it to attempt things that were not previously attemptable.
That is a different claim than the common one, which is about efficiency — AI makes the startup leaner, the team smaller, the iteration faster. All of that is true and none of it is the point. Efficiency compresses time. Leverage changes what is possible.
Entrepreneurship has always been a leverage game, and its primary forms of leverage were capital, labor, and distribution — all three scarce, all three acquired externally. AI has not replaced them. It has added a fourth, different in kind from the first three: Expertise Capital operating through AI, generated internally, from what the founder already carries. It needs no funding round to access, no hiring process to deploy, and no distribution deal to compound. It scales from the moment the founder sits down, and it scales on whatever the founder genuinely knows.
The man in the two chairs, whose ninety days the Foreword describes, was not building a startup in the conventional sense: thirty years of Expertise Capital, one AI Partner, no staff, no co-founders, no office. In those ninety days he structured a 30-location, $15 billion national AI infrastructure investment platform, at seven times the scale of the largest platform he had previously built with the full organizational resources of the largest commercial development company in the United States. Its operating platform required custom software and structuring he would have budgeted, pre-AI, at more than $3 million and more than six months; for one person working alone, before AI, it could not have been done at all. Both measures are his own, and the outside test of the plan is the one every plan of its kind faces — the capital it raises. It is one case, and the book says why there are not yet more: the research behind it appeared between December 2025 and March 2026, and the wider record, if the book is right, sits in the AI companies’ own data, as Appendix A, Part VIII sets out.
The funding threshold for a credible attempt has moved. The team size required to build something significant has moved. The time required to go from idea to institutional-quality output has moved. What previously required an organization to attempt now requires an experienced individual and an AI Partner. What previously required years of team-building now requires weeks of consecutive focused work.
What Becomes Fundable
The leverage shift changes not just what entrepreneurs can build but what investors should fund — and the two are not yet in alignment.
Most venture and growth capital still prices teams the way it did a decade ago: headcount as a proxy for execution capacity, organizational depth as a proxy for risk reduction, prior funding rounds as a proxy for validation. All of those proxies were reasonable when Expertise Capital was illiquid and organizational pyramids were the only delivery mechanism for it. They are less reasonable when a single experienced founder operating at the top Floor their Expertise can reach can produce institutional-quality output at a pace and scale the proxies do not predict.
The mispricing runs in both directions. Experienced solo founders and small teams with deep domain expertise operating at the higher Floors are systematically undervalued by frameworks built for a different era. Larger teams with broad headcount whose work never leaves the Ground Floor are systematically overvalued by the same frameworks — because the headcount looks like capacity when it is actually overhead.
The investors who reprice this earliest will find opportunities the consensus is not yet seeing; the next section takes up that argument. The founder has a more immediate use for it: knowing where the mispricing is tells a founder which investors are still using the old proxies — and which are ready to fund what the new leverage actually produces.
There is a discipline that Sophisticated AI Adoption demands of the founder: the Elevator rises on real Expertise. The entrepreneur who tries to use AI to compensate for Expertise they do not yet have will get a well-formatted version of the average answer, which is what every competitor using AI as a search engine is also getting. The AI Economy rewards founders for applying their deepest Expertise to the largest possible idea that Expertise can credibly support.
The Investor
Every major technology transition produces a mispricing window. It opens when a new form of value creation becomes real but before the frameworks used to price it have caught up, and it closes when the consensus has absorbed it and the excess returns have competed away.
The mispricing window for the AI Economy is open now, but not where most investors are looking for it.
The consensus bet is on the technology itself — the models, the infrastructure, the platforms, the applications. That bet is sound, but it is crowded, and crowded bets compress returns. The less-crowded bet is on what the technology makes more valuable: the human asset that directs it. Expertise Capital, newly liquid, newly scalable, still largely unmeasured by the frameworks the investment community uses to price organizations.
The frameworks that dominate private and public markets were built when Expertise Capital was illiquid, and their proxies for execution capacity were how many experienced people an organization employed, how long they had been there, and how much revenue each of them generated. Two of those proxies are breaking. Tenure is not — it still estimates the depth in a building about as well as anything available, and this argument rests on years producing depth. What tenure no longer estimates is what the firm produces, because that now depends on whether the depth is operating.
When a Sophisticated AI Adopter produces several times the prior output with the same hours and the same Expertise, revenue per employee stops measuring what it used to measure. When a small team of high-density Expertise Capital operating through an Expertise Platform can produce institutional-quality output at a pace and scale the headcount does not predict, organizational depth stops being the risk reducer it used to be.
The Sophisticated AI Adoption gap also connects the Human Capital story to the infrastructure story. The UT/KPMG Study’s 5% were not simply using AI more than the 95%: the researchers found them engaging it ambitiously across complex, multi-step work.
The 95% use AI as an enhanced search engine. The interactions are high in volume and low in depth. The compute they consume is real but shallow: fast, cheap, interchangeable. One search-engine-style prompt looks much like another.
The Sophisticated AI Adopter operates differently: the consecutive hours, the multi-step reasoning chains, the iterative Move Threes that push the model to higher Floors, the synthesis across domains, the large amounts of context held over extended sessions. These interactions are more demanding of the infrastructure serving them.
A single Sophisticated AI Adopter in a full working session should place a categorically different load on the infrastructure than a shallow user running the same number of hours, in the dimensions that stress-test it. That is a prediction rather than a finding. It is also one the providers can settle themselves: the same read that identifies the Sophisticated AI Adopters shows how far their use has widened from everyone else’s, and nobody outside those companies holds the data to check it. Nor does it run against the Science record in Chapter Eight, where the newest developers used AI the most: that record measures volume of use, and this prediction is about depth of session.
This reframes the infrastructure investment question. The build-out of AI compute capacity has been sized, in most models, against a demand curve driven by how many people use these systems and how much the average one uses them. The demand curve for high-value compute is not the average demand curve scaled up. It is a different curve, driven by a different population, with different infrastructure requirements and different margin profiles.
Finding the Signal
The investment question the Human Capital Thesis generates is practical: how does an investor find the organizations that are compounding their Expertise Density advantage before the market has priced it?
Five indicators:
The first is senior-level Sophisticated AI Adoption. The UT/KPMG Study is consistent with the book’s thesis: the best users were often above manager level. An organization where the most experienced people are the most advanced AI operators is building Expertise Density from the top down — the direction that compounds. An organization where AI use is concentrated in junior staff and flagged as a youth initiative is building from the bottom up, which is the direction that does not compound — and the output gap will show it.
The second is output unexplained by headcount. When an organization is producing at a level its team size does not predict — when the revenue per person, the deal flow per partner, the client capacity per consultant has moved in ways the headcount arithmetic cannot explain — something is compounding that the standard model is not capturing. That something is almost always Expertise Capital operating through Sophisticated AI Adoption at a depth the organization has not yet named.
The third is retention of the most experienced people. Sophisticated AI Adoption is not evenly distributed across tenure levels — it concentrates in the people who have the most Expertise to bring to it. Organizations that are retaining and elevating their most experienced people are retaining and elevating the nodes of their Expertise Platform. Organizations that are shedding Expertise in the name of cost efficiency are dismantling the platform before they have named it.
The fourth is the absence of the pyramid defense. Organizations whose leadership describes their AI strategy primarily in terms of tools purchased, training programs deployed, or headcount maintained have not yet understood what the UT/KPMG Study measured. Organizations whose leadership describes it in terms of which of their most experienced people are operating at the highest Floors — and how that Expertise is being compounded across the organization — have understood it. The vocabulary of the answer tells the investor which side of the transition the organization is on.
The fifth is the interaction data itself. The researchers scored the 1.4 million prompts and responses and then set the scores against who the people were — which is how they could report that the best users were often above manager level. A firm holding its own AI deployment and usage data can run the same join, and it knows which of its people carry the most rather than inferring depth from a title. Ask whether the firm has done it. One that can name which of its deepest people are operating at the top of its own data has been managing Expertise Density. One that answers with seat counts has not.
Mispricing windows do not stay open, and the process that closes this one is already underway. The researchers, publishing in March 2026, measured the gap without naming its cause. This book names it: Expertise Capital, and the Density of it inside a firm. If this book is right, the organizations living it have been living it for months without the vocabulary. A name arrives before a consensus does, and the gap between them is where the opportunity concentrates.
The Professional
The next decade of the professional’s working life looks different from the decade before it, because the professional now has a choice that did not exist before, and the choice is consequential: whether to treat AI as a faster tool for doing what they already do, or as a leverage system for doing what they have never been able to do before.
Every professional reading this has already built the most valuable thing they will ever bring to an AI Partner. They built it slowly, over years, through the accumulation of consequence — decisions made under pressure, problems solved without enough information, judgments rendered and then lived with. The professional who looks at that experience and sees only a career rather than a leverage asset has an amplifier in front of them and is treating it like a slightly faster search engine.
There is a version of the next decade in which the AI Economy makes the choice for the professional, on less favorable terms. The organizations, founders, and investors this book describes are reorganizing the professional labor market around a new signal: the depth of Expertise Capital operating through Sophisticated AI Adoption, rather than credentials, tenure, or title. The professional who reaches that depth early is more visible to the organizations and investors who have learned to read the new signal. The professional who does not reach it is becoming less legible to the economy forming around them.
What Becomes Possible
Sophisticated AI Adoption opens three categories of possibility for the professional that were not previously open. The magnitudes that follow are illustrative; the direction is the point.
The first is reach. The professional who once influenced the decisions in their immediate orbit can now influence decisions far beyond it. The physician whose judgment touched the patients in front of them can begin to embed that judgment in systems that reach beyond the room. The Expertise has not changed. Its reach has.
The second is adjacency. The professional who has spent a career in one domain is now equipped to operate credibly at the edge of adjacent ones — as an experienced professional whose AI Partner handles the technical mechanics of the new domain while the professional’s judgment navigates the territory. The man in the two chairs did not become an AI expert when he took on AI infrastructure. He became a real estate expert operating in an adjacent domain, with an AI Partner handling what he did not yet know.
The third is speed. The professional who once needed months to produce what the market valued can now produce it in weeks, by removing the friction between judgment and output. Speed, in this framing, is about how little now stands between the judgment and its expression in the world.
The Repair
There is a population of professionals for whom this chapter carries a particular weight, and it deserves to be named directly.
The professional who spent decades watching their most ambitious work get filtered down, slowed, rationalized away by the convergent structures around them — who knew what the work could be and was repeatedly handed back a smaller version of it — is not reading this chapter as an opportunity. They are reading it as a confirmation.
The argument this book has made, from Chapter One forward, is that the structures that rationed the work were not malicious. They were rational responses to real constraints: Expertise was illiquid, judgment was bound to the person who held it, the pyramid was the only delivery system available. Those constraints have now broken. The rationing is ending — not because anyone decided it should, but because the mechanism that required it no longer exists.
The rationing was the predictable output of systems that were, at every level, optimizing for what they could measure and control — and systematically discounting the kind of work that resists measurement.
The institutions were selecting for the convergent thinking that execution requires, and the filtering ran twice — once in what got proposed, and again in what got approved.
Jennifer Mueller, Professor of Management at the University of San Diego, with Shimul Melwani and Jack Goncalo, documented the other side of the same mechanism. Their research, published in Psychological Science in 2012, found that people in conditions of uncertainty — which describes most organizational decision-making — show a bias against creative ideas even while they say they want them. The bias operates below conscious awareness.
Evaluators who genuinely believe they are selecting for the most ambitious work are simultaneously selecting against it, because ambitious work introduces uncertainty, and the cognitive system managing uncertainty responds to more uncertainty with avoidance. The most consequential professional work — the work that required the most from the person producing it — was being systematically filtered by the ordinary cognitive responses of well-intentioned people doing their jobs in conditions the pyramid reliably produced.
The AI Partnership moves both mechanisms to a later point. The operator’s judgment no longer has to pass through internal layers that apply the Mueller bias before the work can exist, and the work does not have to be translated into a form that survives the institutional filter first. It still meets an evaluator — a client, a board, a market — but it meets one finished rather than as a proposal, and is judged on its results rather than on its palatability to the structure through which it previously had to travel.
The pyramid extracted a cost from everyone inside it — not equally, and not in ways the structure could easily see or measure. This book is not the place to fully account for that unevenness. What can be said here is that the filter is ending, and that the professionals who felt its weight most acutely will feel its removal most directly. The next chapter in that story — what the variation looks like, who experienced it most, and why — belongs to the research this platform is designed to encourage.
For the professional who experienced that rationing most acutely, the AI Economy is not primarily an efficiency gain. It is a repair. The work that was possible but not permitted is now possible and not constrained. The Expertise that was valuable but not deployable is now valuable and scalable. The career that was right but too early is now right and on time.
What to Do Monday
Behind The Thirty lies the Elevator — a method you operate rather than a tool you pick up: one move, the next, the next, rising Floor by Floor as far as your Expertise can carry you. It is built for every working day and every new problem you will face for the rest of your life.
So this is the charge. Stop treating the partnership as an experiment and start treating it as how you work. Bring the Elevator your next real decision, and the one after that, and the one after that. The professionals who will define the AI Economy are the ones who changed the way they work — permanently.
The Stress Test
Every argument worth making has objections worth hearing. The preceding chapters have built a case: Expertise Capital is the scarcest input in the AI Economy, the professionals who bring it to Sophisticated AI Adoption produce output that was not previously possible, and the organizations, entrepreneurs, investors, and careers that compound around that reality are positioned to pull ahead of those that do not.
That is the case. It is incomplete without the hardest questions it generates, answered honestly rather than deflected.
This chapter is the stress test. Three objections, each stated at full strength, each answered with the same rigor the rest of the book has tried to hold. The reader who found the preceding argument compelling but felt a reservation they could not name will find it here, stated plainly, before the book’s answer arrives.
The order matters. The objections are presented from the most fundamental to the most practical — from the challenge that, if true, would overturn the book’s entire premise, to the challenge that, if true, would compress but not eliminate its conclusion.
Objection One: AI Will Acquire Judgment
The strongest objection to this book’s thesis is the one the book has been most careful to avoid overstating. It runs as follows: the book’s entire case rests on the claim that AI cannot produce judgment — that judgment is the residue of years of consequence, that it cannot be downloaded into a system that has not lived those consequences, and that therefore Expertise Capital remains the scarce and valuable complement to abundant intelligence. But what if that claim has a time limit? What if AI systems, given enough training data, enough reinforcement from human feedback, enough time and scale, eventually develop something that functions like judgment — that approximates the pattern recognition, the contextual weighting, the consequence-bearing reasoning that the book says only experience produces?
If that happens, the thesis inverts. The scarce complement to abundant intelligence becomes something other than Expertise Capital, and the argument this book has made loses its foundation.
This objection deserves a direct answer rather than a hedge.
The honest answer has two parts. The first is an acknowledgment: the objection cannot be definitively dismissed. No one knows the ceiling of what AI systems will eventually be capable of, and intellectual honesty requires saying so.
What the objection has to outrun is a record already in the ground. Daniotti and colleagues, publishing in Science in February 2026, read more than thirty million code contributions on GitHub by 160,097 developers across six years of ordinary work. Through the whole of the period in which these systems became capable, the gap between experienced and inexperienced developers widened rather than closed — and it widened while the newest developers were the ones using AI most. That is not a claim about the ceiling. It is the observed direction of travel at population scale, in the open, with nobody encouraging anyone to do anything.
The second part is the structural argument for why judgment, as this book defines it, is resistant to the objection even at long time horizons.
Judgment, properly understood, is not pattern recognition at scale. Pattern recognition at scale is something AI already does better than humans in many domains. Judgment is something different: it is the capacity to act under conditions of genuine uncertainty, with incomplete information, where the cost of being wrong is borne by the person deciding.
The physician who makes a diagnosis does not just recognize a pattern — they commit to a course of action whose consequences fall on a patient, their family, on themselves professionally and ethically. That consequence-bearing is intrinsic to the judgment and cannot be separated from it. A system that recommends a diagnosis without bearing the consequence of being wrong is doing something different in kind from a physician who makes one.
That could be read as winning the argument by definition, so it is worth saying what the definition rests on. Two things make consequence matter, and both can be checked. The first is where judgment comes from: it is built from outcomes a person lived with, most of which were never written down — the tacit knowledge Polanyi described in Chapter Three — so the record a system learns from holds the decisions and rarely the consequences behind them. The second is who answers for a decision: clients, boards, courts and regulators assign that to a person, which is a fact about institutions, not about machines. Neither proves a system cannot come to hold something like judgment. Both explain why the gap has been slow to close, and the Science record shows that so far it has widened.
This does not mean AI will never be trusted to make consequential decisions autonomously. It already is, in narrow domains where the decision space is well-defined and the cost of error is manageable. What it means is that the domain of judgment that Expertise Capital operates in — the open-ended, high-stakes, context-saturated decisions that constitute most of what experienced professionals do — is structurally resistant to the substitution even as AI capabilities expand.
Objection Two: Expertise Depreciation
The second objection is more immediate and in some ways more unsettling than the first, because it does not require AI to acquire judgment. It requires only that AI accelerates the pace of change in professional domains to the point where accumulated Expertise becomes a liability rather than an asset — where what took twenty years to learn is obsoleted faster than new Expertise can replace it.
The objection is already visible in specific domains. The software engineer whose Expertise was built on a stack that has been superseded. The financial analyst whose models can be easily built, tested, and debugged by AI in a fraction of the time previously required. The marketing professional whose channel Expertise was made irrelevant by a platform shift. Domain-specific Expertise depreciates. It always has. The question the AI Economy raises is whether the depreciation rate is accelerating to the point where the half-life of expertise becomes too short to invest in.
The answer requires separating two kinds of expertise that the objection conflates.
The first kind is domain-specific technical expertise — the specific tools, methods, platforms, and techniques that constitute the current practice of a field. This kind of expertise has always depreciated, and AI is accelerating that depreciation in the domains where it is most capable. The software engineer who builds their identity around a specific stack is more exposed than one who does not. This is real, and the objection scores a genuine point here.
The second kind is domain judgment — the accumulated understanding of how a field works at the level beneath its tools: what the real problems are, where the bodies are buried, which risks look small but are not, what the client actually needs versus what they say they want, how the stakeholders will behave when the plan meets reality. This kind of expertise does not depreciate with tool changes — it often becomes more valuable as the tools change, because the judgment about which new tools to trust and how to deploy them is itself a form of accumulated Expertise.
The Sophisticated AI Adopter is invested in the judgment that operates through whatever tools are available, not in the tools themselves. The Elevator is a method for extracting higher answers from whatever frontier system is current, by bringing the operator’s accumulated Expertise to bear on it. As the tools change, the method adapts. The Expertise Capital that makes the method rise is not tool-dependent.
Objection Three: Everybody Uses the Elevator
The third objection is the most practical and in some ways the most familiar to the investor and executive reader who has watched prior technology advantages compete away.
If the Elevator is a method, and methods can be documented and distributed, then the Sophisticated AI Adoption advantage is temporary. Once the method becomes widely known, widely taught, and widely practiced, it stops being a differentiator and becomes table stakes. The 5% becomes 50%, and the output gap that the book has priced as a durable advantage compresses to a commodity.
The objection is correct in its mechanism and wrong in its conclusion.
The mechanism is right: methods do spread, and the Elevator will become more widely practiced over time. This book is, among other things, an instrument of that spread. The conclusion is wrong because it confuses the method with the output of the method.
The Elevator is the same three Moves for everyone. The Floors are not. A widely distributed method for converting Expertise Capital into altitude produces widely distributed altitude — each professional rising as far as the Expertise they genuinely bring to it. The method spreading does not spread the Expertise it runs on. It makes Expertise Capital more liquid, which is not the same thing. Water pipes become more widely available; the water they carry remains unevenly distributed.
What the spread of the method actually produces is the acceleration of the repricing this book has argued is already underway. As more professionals reach Sophisticated AI Adoption, the market’s ability to measure Expertise Capital — to distinguish the professional whose thirty years produces at one altitude from the one whose thirty years produces at another — improves. The signal gets cleaner, and the advantage of genuine depth over shallow credentialing becomes more visible.
The professional who is worried about this is, underneath the objection, worried about whether their Expertise Capital is genuinely deep or whether it has been masquerading as depth behind the friction of a less leveraged era. That is a legitimate concern, and it deserves an honest answer: the AI Economy will make that distinction clearer than it has ever been. The professional with genuine depth has nothing to fear from a more efficient market for it. The professional whose apparent depth was a function of information asymmetry and access friction has more to reckon with — and the time to reckon with it is now, not when the market has already priced the difference.
There is a larger version of the same objection, and most readers are living inside it. Every company in the reader’s industry has been investing in AI, and the majority of that investment has been aimed at one thing: doing existing work with fewer hours. Call it what it is — AI-Driven Cost Savings. It is competitively necessary, it has to be pursued, and a company that skips it competes at a cost disadvantage against every company that did not. It also commoditizes by construction. The capability is embedded in a tool, the tool is for sale, every competitor buys it, and the advantage converges.
The answer is the same one, at the scale of a firm. The tool spreads, and the savings converge with it. The Expertise a firm points the tool at does not spread with it, which is why the second investment is permission rather than procurement: it has no vendor, and what it costs is the hours of senior people running the Elevator on real problems. Chapter Six sets out how a firm does it.
The Verdict
Three objections. Three honest answers. The book does not claim the objections are trivial — each scores a genuine point, and each leaves the thesis standing once that point is given its due.
What the stress test establishes is the shape of the uncertainty. The book’s argument does not require AI to be permanently incapable of judgment — it requires the transition period to be long enough to matter. It does not require expertise to be immune from depreciation — it requires judgment-based expertise to depreciate more slowly than tool-based expertise. It does not require the method to remain rare — it requires genuine Expertise Capital to remain unevenly distributed even after the method spreads, which the Science record already shows among developers who all held the same tools.
The direction holds. The roughness is real. And the professional, organization, investor, or entrepreneur who acts on the direction now — before the roughness has resolved and the consensus has caught up — is making the same bet every intelligent early mover in every prior technology transition has made.
The Partnership
Chapter Four handed you the method. This chapter is about what you do next.
What the Argument Established
The chapters before this one built a single case, and the Stress Test set its direction against its strongest objections. The destination is worth the crossing.
It is worth saying how that is reached, because the word usually arrives attached to leisure and means very nearly the opposite here. What AI multiplies is Expertise, immediately and from the first ride. Output rises, and it becomes attributable, because the layers that attenuated the signal are thinning. Attribution and those thinning layers together drop the size of a unit that can do serious work. More units, each carrying its own production, means fewer hands between the judgment and the customer. More units means more surfaces facing the market, so more niches get addressed and more categories get attempted. And every unit produces more consequential work than its deepest person can carry, which is what raises the people holding less. More work of higher value, done by more people, in categories that do not exist yet. It is a forecast, but one built link by link from what the mechanism does rather than from a trend line.
The Man in the Two Chairs
He is worth returning to one final time. Thirty years of institutional experience. A MacBook Air. Two unremarkable chairs in two public locations. No plan, no staff, no infrastructure, and no permission from any institution, because none was required. Ninety days later he held what Chapter Seven set out. The books are in your hands; the platform’s capital is still being raised. He did not become someone different over those ninety days. He became a more leveraged version of the same person — thirty years of Expertise Capital operating at a scale and speed it had never before been able to reach.
The relevance for you is what the ninety days illustrates about what is now available to anyone who brings genuine Expertise Capital to an AI Partner and refuses to stop at the first Floor.
The work you have spent years becoming capable of doing has become scalable. The Expertise you have built through consequence and time and error and recovery is a force to apply, not a credential to maintain.
What You Do With It
The book has argued. The Stress Test has answered. What remains is the question that every argument eventually arrives at and cannot answer for you.
What will you do with it?
With your specific Expertise Capital — the thirty years, or twenty, or fifteen, or five, that you have spent building the only thing that makes the Elevator rise. With the largest idea that Expertise makes uniquely yours to attempt. With the consecutive hours that Sophisticated AI Adoption demands and rewards in ways no prior working method has.
The book cannot answer that question. It can only have made it unavoidable.
The professionals who have already answered it are not waiting for permission, for the technology to improve further, or for the consensus to confirm what they have already observed. They are sitting in unremarkable chairs with MacBook Airs, bringing their Expertise to an AI Partner, and producing work that was not previously possible.
The Speed of the Crossing
There is one more thing worth saying before this book ends.
The transition from the old economics of Expertise to the new one is not painless. The professionals whose value was in their tools rather than their Expertise are exposed, and so are the organizations whose competitive position rested on information asymmetry and access friction, and the educational institutions whose model was built for a world where Expertise formation was slow and expensive. The crossing costs something, and the people who pay that cost are real.
What made every prior transition brutal was not the displacement. It was that the people displaced had no way to act on their own behalf. The typing pool had none. The travel agent had none. What was on offer was, at best, a training program that might qualify them for different work — never a way to use what they had already spent a career building.
AI drives cost cutting and job losses. Expertise-Driven AI Partnerships drive value and facilitate the crossing.
That third clause closes a hole. Without it the two sides run parallel but disconnected — one sheds people, the other needs people, and nothing links the person who was shed to the work that needs doing. Facilitation makes the instrument that displaced them the instrument they cross on.
AI-Driven Cost Savings replaces hours and removes bodies. It multiplies nothing for the person whose hours were removed.
Expertise-Driven AI Partnerships are where the multiplication lives, and it appears only when earned experience and judgment is doing the directing. Two things decide which side a person ends up on: the depth they bring, and their willingness to actively engage an AI Partnership — to run the Elevator and become a Sophisticated AI Adopter rather than a person with an account.
What changed is that this is now available to the individual and not only to the organization employing them. Availability was never the multiplier. Nothing has to be purchased: the part that matters is their own Expertise, and they already hold it.
One limit, and it is absolute. Facilitation means the instrument is available. It is never a promise about how anyone lands. Some of the people whose work is absorbed will not cross, and saying so plainly is what makes the rest of this credible — the readers most able to damage the argument are the ones who have watched a transition go badly, and they stop listening to anyone who skips the losses.
That reaches anyone holding Expertise at any depth, and it arrives at their employer as a fact that did not used to exist. What held a deep professional in place was never only the compensation. Part of it was the apparatus — the firm’s brand, its staff, its client base, its capital, the standing that came with the letterhead. Much of that apparatus still holds, but its production role is no longer required to deploy what that person knows. The return accrued to the apparatus because the apparatus controlled the route to market. The experience and judgment were scarce then and they are scarce now, and the return now reaches the person holding them.
The three constituencies follow from the same mechanism. The displaced — the instrument that removed the work is the instrument available for the crossing, and it requires no retraining. The already-out — the retiree is the strongest case for facilitation rather than the softest, because the transition already happened and was total. Outlets existed, but every one of them required something besides the Expertise: a network, a brand, a firm, capital, a staff, or the willingness to sell. AI Partnership is the first outlet that requires only the Expertise itself. And the young — they are not gaining on the experienced, because on complex judgment work the gap still widens in favor of depth. They are gaining on their own predecessors, because the production role of the apparatus is no longer required, and because a defensible niche too small for a large firm to bother with is the fastest Expertise-earning environment available. Consequence has always been rationed by hierarchy. A niche you own outright is unrationed consequence.
The urgency this book carries — the reason it does not soften the argument or slow the pace — is not impatience with the people who have not yet made the crossing. It is the opposite. The faster the crossing happens, the shorter the period in which the costs are borne. Every individual who reaches Sophisticated AI Adoption is one fewer person stranded in the gap between the old economics and the new. Every organization that builds an Expertise Platform is one fewer institution reorganized by events it did not choose. Every investor who prices Expertise Capital correctly is one more force accelerating the repricing that closes the gap for everyone. The Stakeholders, at the back of this book, carries the same logic to every party in the AI Economy.
Speed is not ambition. Speed is mercy.
There is a supply condition underneath the urgency, and it belongs here.
The cohort that holds the most concentrated Expertise Capital — the professionals between forty-five and eighty years old who have spent between twenty and fifty years in consequence-bearing practice — meets three conditions simultaneously that no other population on earth currently meets. The judgment they carry took decades to build. No technology can manufacture it. And no younger generation has yet had time to earn it. That is a statement about accumulation, not about the capability of younger professionals. How long Expertise takes was never a law: the rate is set by how much consequential work a person carries, and consequential work is exactly what amplified Expertise produces more of. So the climb ahead of a younger professional can shorten, and this book argues that it will. What it cannot do is shorten retroactively. The depth that exists today exists where the decades were already spent.
The forty-five-to-eighty cohort is a specific, finite, time-bounded population, and it holds the deepest earned Expertise in this economy in two places: the younger part mostly still inside a building, rationed by span of control and the calendar, and the older part mostly outside one, carried out whole and dormant. The window in which this concentration of consequence-bearing judgment can be brought into AI Partnership, and applied at the scale the technology now makes possible, is defined not by age or institution — but by the choice to engage.
The Beginning
Books end. The work they describe does not.
He is still sitting in those chairs — not because the ninety days produced a finished thing, but because the method that produced it keeps producing. The book in your hands is evidence of that: two books, a national investment platform whose capital is being raised, and a set of publishing innovations — because the partnership compounds and the compounding does not stop.
That pattern is available to you — not the specific output, but the pattern itself: the experienced professional, the AI Partner, the consecutive hours that become the Sophisticated State, the refusal to stop at the first Floor, the output that surprises even the person producing it.
The Thirty is the door. The Elevator is the method. The Expertise you have already built is the only force that makes it rise.
What you build with it is the book that has not been written yet — the one that belongs to your Expertise, your largest idea, your two unremarkable chairs. The gold has been piling up around you the whole time. Now you can see it.
Now the exponential is you.
—
The Sophisticated AI Adopter is the second of two books. The first, Why Us?, is written for the individual professional — the recognition that the structures that rationed your most valuable work are ending, and the method that carries you past them. If the argument in these pages landed, Why Us? is where the Elevator was first built.
The two books are, in the end, the same argument from two vantage points: the individual who does the work, and the world that is reorganizing around the fact that they can.
What your Expertise builds does not stop with you. Every stakeholder in the AI Economy gains from it, and none of those gains comes out of another’s. Touch or click below and find out how. “Return to the Book” drops you right back to this page.
A Note on Sources
The back matter that follows draws on established research spanning more than a century — from Cannon (1915) and Wolpe (1958) through Bartlett (1978), Csíkszentmihályi (1990), Ericsson and colleagues (1993), Porter (1996), An (2026), Daniotti and colleagues (2026) and Hallman and colleagues (2026). Every finding is attributed to its original source in the text and listed in full in the References that close this book.
The synthesis, the sequencing, and the closing observations are the author’s.
The Empirical Engine
The preceding chapters made claims about human cognition, creativity, organizational behavior, and the economics of Expertise. They rest on a body of published research and strategy literature spanning more than a century. This appendix assembles that evidence.
The Exponential Problem
The first claim the book makes is that human beings cannot intuit exponential change — that the gap between what the AI Economy is producing and what most people believe it is producing is a failure of perception rather than of information.
Albert A. Bartlett (1923–2013) was a professor of physics at the University of Colorado, Boulder, who gave a one-hour lecture on exponential growth and its consequences 1,742 times, beginning in 1969. His central point, first stated in a 1976 article in The Physics Teacher and developed across a lifetime of subsequent work, including his 1978 paper Forgotten Fundamentals of the Energy Crisis, was simple and devastating: the human mind cannot grasp the exponential. Not resistant to it. Not slow to accept it. Fundamentally unable — the perceptual apparatus that evolved to navigate a linear world does not produce reliable intuitions about nonlinear change.
In Bartlett’s framing, this is a feature of biological cognition encountering a mathematical phenomenon it was not built to handle, not a failure education can correct. The gap between what the AI Economy is producing and what most participants believe it is producing is, in significant part, a Bartlett gap.
The Flight Response
If the exponential cannot be grasped, it will be felt as threat. And threats, in the human autonomic system, trigger a response that predates rational deliberation by hundreds of millions of years.
Walter B. Cannon (1871–1945), Chair of Physiology at Harvard Medical School, first described the fight-or-flight response in his 1915 work Bodily Changes in Pain, Hunger, Fear and Rage, and developed it across his landmark 1932 book The Wisdom of the Body. His central finding: faced with a perceived threat, the human organism initiates a cascade of physiological changes — adrenaline released, the heart racing, blood sent to the muscles — that prepare it for physical response. The system is fast, automatic, and operates below the threshold of conscious control.
The relevance for the AI Economy is direct. A technology that cannot be grasped exponentially will be perceived as threatening. A technology perceived as threatening will trigger the flight response before rational assessment can occur. Chapter One reads the expectation of seventy-nine percent of Americans in the 2026 Bentley University–Gallup survey, that AI will reduce the number of U.S. jobs, as in significant part a Cannon response.
The Way Through
The Bartlett perception limit produces the Cannon flight response, and neither can be argued away. The question is whether there is a reliable path through.
Joseph Wolpe (1915–1997), a psychiatrist and pioneer of behavioral therapy, established through decades of clinical research what he called systematic desensitization: the principle that fear is not reasoned away or waited out. It ends only through direct, graduated, voluntary contact with the feared object, in conditions that allow the autonomic response to habituate. His work, developed in his 1958 book Psychotherapy by Reciprocal Inhibition and refined across a career of clinical practice, became the foundation of modern behavioral therapy and the treatment of anxiety disorders worldwide.
The Thirty is the book’s application of Wolpe’s principle. Not therapy — the analogy is structural, not clinical. The exponential cannot be grasped (Bartlett). Encountering it triggers withdrawal (Cannon). The withdrawal ends only through direct, low-stakes, voluntary contact that allows the autonomic response to habituate and the rational assessment to begin (Wolpe). The Thirty is designed to be that contact — the Bartlett-Cannon-Wolpe loop, built into the book’s architecture.
The Sophisticated State and the Desire to Continue
Mihaly Csíkszentmihályi (1934–2021), Distinguished Professor of Psychology and Management at Claremont Graduate University and former head of the Department of Psychology at the University of Chicago, set out in his landmark 1990 work Flow: The Psychology of Optimal Experience, and the decades of research around it, the conditions under which human beings enter states of peak engagement — what he called flow: total absorption in a challenging task that stretches but does not exceed the performer’s capabilities. His research found flow most reliably produced when the difficulty of the task is calibrated to the skill of the performer, and found a desire to return to the activity that produced it among its features.
Chapter Four sets out how the AI Partnership meets those conditions, and the name this book gives the resulting state: the Sophisticated State.
The Ten-Thousand-Hour Question
The book claims that the depth of accumulated Expertise is the primary variable determining how far the Elevator rises. That claim rests on one of the most contested and most replicated findings in the psychology of Expertise.
Anders Ericsson (1947–2020), Professor of Psychology at Florida State University, spent four decades studying the development of expert performance across domains from chess to medicine to music. His core finding, developed in his 1993 paper The Role of Deliberate Practice in the Acquisition of Expert Performance (co-authored with Krampe and Tesch-Römer) and brought to general audiences through Malcolm Gladwell’s Outliers (2008) as the “ten-thousand-hour rule,” was that expert performance is not primarily a function of innate talent but of deliberate practice — structured, effortful, feedback-rich engagement with the domain over extended time.
The popularized version of this finding — ten thousand hours to mastery — is an oversimplification that Ericsson himself spent the latter part of his career correcting. The finding that survives scrutiny is more precise and more relevant to this book’s argument: Expertise is built through deliberate practice, not mere repetition; the quality of the experience matters as much as its quantity; and the resulting Expertise is genuinely different in kind from the performance of less-practiced individuals, not merely faster or more accurate.
Michael Polanyi’s (1891–1976) parallel concept of tacit knowledge — developed in his 1966 work The Tacit Dimension — adds the dimension Ericsson’s framework implies but does not fully articulate: much of what Expertise produces cannot be made explicit, cannot be verbally transmitted, cannot be taught in a classroom. It is knowledge that lives in the practitioner’s judgment rather than in any articulable rule, and it is the knowledge the Elevator carries into the AI Partnership.
The Mechanism
Tao An, at Hawaii Pacific University, set out to reconcile two bodies of published evidence that appeared to contradict each other. One line of research reports that AI narrows the performance gap between novices and experts; another reports that it widens them. His paper argues that both are correct in different places, and that the variable separating them is task complexity. It draws on the published literature and on structured observation of a working software team since mid-2024, and it is explicit about what it is: a falsifiable account offered to be tested, not the test itself.
The argument is precise. Where the task is well structured and the correct approach is already represented in what the system has learned, the machine supplies the answer and measured differences between people compress. Where the work requires judgment under ambiguity, the machine supplies raw material that must be directed, evaluated and refined — and there the gap widens, because a person who cannot tell a good output from a plausible-sounding wrong one cannot use what comes back.
Figure 1 is reproduced in Chapter Three. Tao An proposes the mechanism — why the senior professional outperforms. What follows is the record of it, first inside one organization and then across a working population.
The UT/KPMG Study
The empirical anchor of this book’s contemporary argument is a study published in March 2026 in Harvard Business Review by researchers at the McCombs School of Business, The University of Texas at Austin, in partnership with KPMG: eight months of observation inside one of the four largest professional advisory firms in the world, 2,500 professional employees, and 1.4 million AI prompts and responses tracked and analyzed.
The definition: The researchers assembled a definition of sophisticated use that builds on prompt engineering — prioritizing clear prompts along with deliberately applied strategies. Chapter Three gives their description of sophisticated users.
The findings
The firm entered the study with approximately ninety percent of employees already using AI — a measure of access and habit, never of output. After eight months of sustained institutional encouragement, active organizational support, comprehensive AI tool provision, and a saturating external wave of AI marketing and media, approximately five percent of the 2,500 participants fit sophisticated use as the researchers defined it. The remaining ninety-five percent were using AI as an enhanced search engine. The best users were often above manager level — disproportionately so, given how few people sit there — a finding the researchers described as surprising, given the conventional wisdom that more junior employees are more comfortable with these tools.
What the study establishes and what it does not
The study reports the distribution at the end of the observation period, not the conversion over time. The proportion of participants who fit the definition of sophisticated use on day one is not reported. The 5% measured at the conclusion may represent a stable population that was already at that level when the study began, with the eight months of institutional support producing little or no conversion. This ambiguity is preserved in the book’s argument: the study is cited as consistent with the Expertise hypothesis, not as proof of it. The researchers’ own full working paper, released in August 2026, reads its seniority finding the same way: senior employees’ more sophisticated use is “consistent with domain expertise complementing genAI capabilities.” While the share fitting the researchers’ definition of sophisticated use within the UT/KPMG Study was small — even among a population given the tools, the training and the encouragement of a large firm — that number should be viewed as the floor for professionals whose accumulated experience and judgment is precisely what AI is able to effectively amplify.
The researchers described as “surprising” the finding that the best users were often above manager level rather than among the more junior employees assumed to be most comfortable with the tools. Prior independent research had already predicted it.
The Global Record
The record introduced in Chapter Three covers more than thirty million code contributions on GitHub, the world’s largest collaborative programming platform, by 160,097 developers in six countries, from 2019 through 2024 — every American developer on the platform, plus samples from the rest. Experience was measured as years on the platform, and the effect appears nowhere in the same data before AI existed.
How AI use was detected: The researchers could not ask anyone what they had used, so they built a classifier to read the code itself. It was trained on human-written functions from 2018 — before capable AI coding tools existed — paired against synthetic twins produced by a two-model chain, one model describing a human’s function and a second re-implementing it from that description alone. The classifier separates AI-written from human-written code with an out-of-sample score of 0.96, where 1.0 would be perfect.
What the record found is set out in Chapter Three, where Figure 3D is reproduced: the experienced captured nearly all of the measured gains and took up libraries they had never used, which the authors read as moving into new areas of programming, while the newest developers used AI the most. What this appendix adds is how the record was built, and why its estimates are conservative.
Where the fluency assumption was tested
Every subject in that record is a software developer. Not one of them ever stood at the technical gate — they all had the access and they all had the fluency. This is experience outperforming access among people who all have access.
Why the estimates are conservative
Each developer is compared against himself, before and after they began using AI, with the whole economy’s trend held out. Run the same models on the years before these tools existed and the effect disappears entirely — the separation is not a property of the people, it appears when the tool arrives. And any error in measuring who was using AI pushes the estimates toward zero, which means what is reported is a floor rather than a ceiling.
The three bodies of work arrive from different directions at the same experienced professional: the Hawaii Pacific paper explains why depth decides the outcome, the UT/KPMG finding names the population inside a single organization, and this record shows the same pattern at global scale. Three limits belong beside those findings. Tao An’s paper is a framework, not a measurement. The UT/KPMG study observed one firm, and reports where its people stood after eight months rather than how they moved. The Science record covers one profession, software development. None of them measured a capital partner, a physician or a litigator directly, and none of them is a company that took up Expertise-Driven AI Partnerships as an organization. None could be yet: all three appeared between December 2025 and March 2026, and the category itself is named in this book and its companion. The organizational cases here are the author’s own work and Haier, whose structure predates AI. What carries the argument across fields is the mechanism they share — where the correct approach is not already known, AI amplifies the Expertise directing it — and the test that matters most is the reader’s own Thirty, run in their own field.
The rest of the evidence is not this book’s to publish. The frontier AI companies already study their own users’ records: Anthropic reports that the people who have used Claude longest bring it harder work and succeed more often, has linked a survey of its users’ occupations and years of experience to how those same people use the tool, and has interviewed 81,000 of its users about their work. If this book is right, the same records hold users whose use changed in kind as the technical barriers came down this year — from a few hours a week to forty or fifty, in sessions of six to twelve hours — and more of them with each step down. What those people have made since would show it. That read belongs to the companies that hold the records.
The Commoditization Mechanism
The book claims that one of the two ways AI is put to work converges and the other resists convergence — that AI-Driven Cost Savings raises every competitor to the same line while Expertise-Driven AI Partnerships are resistant to that convergence. The first half of that claim is not new, and it is not this book’s. It was established thirty years ago, in the most widely read article in the literature of competitive strategy.
Michael E. Porter, Bishop William Lawrence University Professor at Harvard Business School, published “What Is Strategy?” in Harvard Business Review in November–December 1996. Its first section separates two things most organizations treat as one: operational effectiveness — performing similar activities better than rivals perform them — and strategy, performing different activities, or the same activities differently. Porter’s finding is that operational effectiveness is necessary and insufficient. Improvements every competitor can copy raise the bar for the whole industry and leave nobody ahead. The firms become harder to tell apart, and the gain passes to customers as lower prices and to the suppliers of the equipment everyone is buying. He called the result competitive convergence.
Porter documented it in commercial printing, where a decade of identical investment by rival firms compressed margins across the industry without changing the order of finish. The mechanism does not depend on which technology a firm installs. It depends on the capability being purchasable — on every rival being able to acquire the same thing on the same terms.
That is the precise condition AI-Driven Cost Savings satisfies. The capability is embedded in a tool, the tool is for sale, and the price of a seat falls as the market matures. Porter’s argument therefore applies to it without modification, and it is the reason this book treats cost savings as competitively necessary and competitively inert at the same time. What Porter could not have described, because it did not exist in 1996, is the other side: an input no rival can acquire on the same terms, because it has to be earned by carrying decisions that mattered. A firm can hire it; no firm can make it. Expertise fails the purchasability condition, which is why the second use resists the convergence the first one guarantees.
Porter’s article is a strategy argument rather than an empirical study, and it is included here on that basis. It supplies the mechanism, not a measurement.
The bodies of work assembled in this appendix were not selected to support a predetermined conclusion. They were identified because they describe, from independent vantage points across more than a century of scholarship, the same mechanism. The argument in the preceding chapters rests on this foundation. The foundation was built by others.
The Argument
The argument of this book in its shortest form: the eight beats the chapters were written under. Each chapter argues one stretch of them at length.
“The greatest shortcoming of the human race
is our inability to understand the exponential function.”
— Albert A. Bartlett, Physicist, University of Colorado
Expertise is earned.
AI Partnership amplifies Expertise.
Now the exponential is you.
The Argument in Eight Beats
One. Expertise can scale. Expertise was always the scarce asset, bound to one person’s calendar — rationed inside a pyramid and dormant outside one. The corporate pyramid is the hierarchy most organizations run on, a few seats at the top and more at each level below, where the decisions a person touches are set by their seat and their span of control rather than by the Expertise they have earned. Expertise is new gold, piling up around everyone in proportion to what they have earned, and invisible to most. Intelligence became abundant, and inside an AI Partnership, Expertise can scale for the first time. No one perceives a move like that until it grows large enough to see — a documented limit of human perception rather than a failure of anyone’s judgment — which is why the move is still largely unseen.
Two. So there are two ways to put AI to work, different in kind rather than in degree, and task complexity decides which is in play. AI does the work — AI-Driven Cost Savings, and the gain is hours removed. Or a person’s earned Expertise drives the AI — Expertise-Driven AI Partnership value creation, and the gain is new work and new business that did not exist before.
Three. The first converges. It sits inside a tool every competitor can buy, it is competitively necessary, and it will not distinguish anyone. The second is resistant to commoditization, because no course and no AI training converts a lack of consequential experience and judgment into more depth of Expertise. Both are competitively necessary. Missing the first ends a business quickly and visibly. Missing the second ends it slowly, through businesses it never created and work it never won.
Four. The two have opposite consequences for people, by construction. One removes hours from work a company already does, and the hours arrive as job losses. The other generates work that did not exist before, and someone has to carry it.
Five. The same AI that powers the cost-saving tools behind the job losses is available to the people who lost those jobs, for Expertise-Driven AI Partnerships of their own. AI-Driven Cost Savings multiplies nothing for the person whose hours it took; the multiplication requires the depth they already hold and the work of running the Elevator. Neither the AI nor the Expertise does anything on its own, and only the person holding the Expertise can do the work. Nothing has to be purchased: the part that matters is their own Expertise, and they already hold it.
Six. As the deepest Expertise multiplies its own output, the volume of consequential work rises — and consequential work is the only thing that has ever produced Expertise. The time it takes to acquire Expertise does not have to equal real time. That compounds inside a person and between people of different depth — and for the generation beneath, consequential work arrives earlier in a career than a pyramid ever let it.
Seven. The first move was the one everybody made: the AI industry sold AI-Driven Cost Savings, and Corporate America bought it, in software and training. So the second move is permission rather than procurement. There is nothing to buy. What a chief executive supplies is the encouragement to read, to run their own Elevator, and to aim it at the largest problem they hold that will benefit the company. What held deep professionals in place was never only the compensation; part of it was the apparatus, and its production role is no longer required. The senior executive team is where it starts and not where it stops, and the question a board will put to it is whose Expertise, inside this building and outside it, is not at work, and why.
Eight. Now the exponential is you. In every revolution before this, the technology was the exponential, and it reached people only after decades of building, and only through those with the degrees and the industry experience to use it. AI waits on neither: it is already in place and open to anyone, anywhere, working or retired, on any idea, personal or professional, and the only limit on what comes back is the Expertise a person brings to it.
The Stakeholders
Every stakeholder in the AI Economy has its own reason to want earned Expertise put to work, and none of those gains comes out of another party’s, because almost none of the asset is in production today. The wheel names each stakeholder in a line, and the cards on the page that follows give each one’s reasons.
Glossary of Terms
The terms below are used consistently throughout this book and across the work that surrounds it. Expertise and the named assets built on it are capitalized to mark the defined asset; the ordinary word, meaning a person’s proficiency in a particular field, is not.
Expertise. The combination of experience and judgment that no system can acquire without living the consequences of acquiring it. It is not information, not credentials, and not access. It is earned by carrying decisions that mattered, and it cannot be manufactured, downloaded, or shortened retroactively. A firm can hire it. No firm can make it.
AI Partnership. A person directing AI. Not AI assisting, and not AI alone. The distinguishing feature is that the person supplies the problem, the standard, and the judgment about what comes back; the system supplies scale and speed. What the partnership produces reflects the depth of the person operating it.
Sophisticated AI Adopter. A person who talks and works in partnership with AI the way they would with the most engaging and widely knowledgeable individual in every subject and computer program in the world, bringing it their most demanding work and pressing every answer until it reaches the ceiling their own Expertise sets rather than the one the AI Partner sets. It names a way of working, not a level of technical skill and not a job title. The practice is available at any level of Expertise; what a person brings sets how high it can reach. Sophisticated AI Adoption is that same standard described as a state rather than a person. It is crossed once: from that point the person can bring the Elevator to any future Expertise-Driven AI Partnership, whether or not every ride is taken to altitude.
The Elevator. The method by which Expertise is converted into AI output at altitude. It has three Moves and no more.
Move One — the load. Choose the domain and bring a real problem at its full size.
Move Two — breaking from the Ground Floor. The hardest stretch, where the system returns its inventory of caution and most people who quit, quit.
Move Three — the climb. Iterative refinement in pulses, each hard moment a door rather than a wall.
The Floors are the successive levels of answer the climb reaches. The Ground Floor is where every ride begins, and the last Floor is the operator’s to determine.
Sophisticated State. The condition in which earned Expertise is fully engaged and the work stops feeling like effort — hours pass without registering as hours, and the operator wants to return to the work rather than recover from it. It is what the Elevator is built to produce at altitude, and it is why the climb sustains itself rather than requiring discipline to continue. The state itself is documented; the Sophisticated State names the case where Expertise driving an AI Partnership is what produces it.
AI-Driven Cost Savings. AI does the work. Existing tasks performed with fewer hours and fewer people. It is competitively necessary and worth having, and it converges: the capability sits inside a tool every competitor can buy on the same terms, so it raises the floor for an industry without distinguishing anyone inside it.
Expertise-Driven AI Partnerships. A person’s earned Expertise drives the AI. The two names differ by one word in the same position, and that word is the driver. This side is resistant to commoditization for one reason: a firm can hire its input, but no firm can manufacture it. It is where new work comes from, and it requires people rather than replacing them.
Expertise Capital. Expertise regarded as an asset rather than as a personal quality. It sits on no balance sheet, was paid for over the working lives of the people carrying it, and reprices when a mechanism arrives that converts it into output at scale.
Expertise Density. How much operating Expertise Capital an organization holds per head. Two terms decide it — the depth each person brings, with years in the domain as the starting estimate, and whether that depth is being multiplied at all. Depth that never operates does not count, which is why two firms of identical headcount and identical tenure can differ enormously in Density, and why Density rather than headcount is what predicts what a firm can produce.
Expertise Velocity. The rate at which new Expertise forms. Expertise has always been counted in years because the rate was fixed — a career supplied roughly one career’s worth of consequential decisions, and hierarchy rationed how many any one person got. Velocity rises when consequential work arrives sooner and in greater volume, which is what an Expertise Platform produces. Density is the level at a point in time; Velocity is read as how much sooner consequential work reaches the people holding less.
The Expertise Platform. The architecture that puts an organization’s Expertise in contact with itself — each Sophisticated AI Adopter a node, the value rising with the number of nodes and the quality of what passes between them. It is built rather than bought, and its seed is almost always people already inside the building.
The Human Capital Thesis. The argument the other four instruments make when a market prices them. As intelligence becomes abundant, the scarce complement is Expertise, and markets are beginning, unevenly and imprecisely, to price the Expertise operating through it. The frameworks used to value organizations still read headcount, credentials and tenure, and the gap between those proxies and what a high-density team produces is where the mispricing lives.
Exponential node. A Sophisticated AI Adopter operating an Expertise-Driven AI Partnership. What AI multiplies is Expertise, so from the first ride the depth a person has already earned operates at a scale it could never reach alone. That is the exponential, and it is immediate; nothing has to be added first. Expertise gained during a ride is available to be multiplied on that same ride, which adds to the effect without being required for it. A person operating the Elevator below that standard still produces good work; they are simply not multiplying what they hold. Node and exponential node name the same person from two vantage points — node when the subject is the Platform, exponential node when the subject is what is happening to them. They are not two tiers.
The pyramid. The hierarchy most organizations run on, a few seats at the top and more at each level below, where the decisions a person touches are set by their seat and their span of control rather than by the Expertise they have earned. It is the structure institutions built to deploy Expertise before it could travel on its own — layers of management, approval chains, and capital gatekeepers that determined how much of any individual’s judgment actually reached the work. It solved a real problem. The problem has changed.
The Thirty. Thirty minutes, from a single entry point, that carry a person from no AI Partnership to their first Expertise-Driven result. It is the door, not the method.
References
An, T. (2026). “AI as Equalizer or Amplifier? Task Complexity as the Moderating Factor for Human Expertise in Hybrid Intelligence Systems.” Hawaii Pacific University. arXiv:2512.10961.
Anthropic (2026). “Anthropic Economic Index report: Learning curves.” March 24, 2026. anthropic.com/research/economic-index-march-2026-report.
Anthropic (2026). “Anthropic Economic Index report: Cadences.” June 26, 2026. anthropic.com/research/economic-index-june-2026-report.
Autor, D., Chin, C., Salomons, A., & Seegmiller, B. (2024). “New Frontiers: The Origins and Content of New Work, 1940–2018.” Quarterly Journal of Economics, 139(3), 1399–1465.
Bartlett, A.A. (1976). “The Exponential Function.” The Physics Teacher, 14(7), 393–401.
Bartlett, A.A. (1978). “Forgotten Fundamentals of the Energy Crisis.” American Journal of Physics, 46(9), 876–888.
Bentley University & Gallup (2026). Business in Society Report. Gallup Panel web study of 3,270 U.S. adults, May 4–11, 2026. news.gallup.com
Cannon, W.B. (1915). Bodily Changes in Pain, Hunger, Fear and Rage. New York: Appleton.
Castilla, E.J., & Benard, S. (2010). “The Paradox of Meritocracy in Organizations.” Administrative Science Quarterly, 55(4), 543–576.
Csíkszentmihályi, M. (1990). Flow: The Psychology of Optimal Experience. New York: Harper & Row.
Daniotti, S., Wachs, J., Feng, X., & Neffke, F. (2026). “Who is using AI to code? Global diffusion and impact of generative AI.” Science, 391(6787), 831–835, February 19, 2026. doi:10.1126/science.adz9311.
Ericsson, K.A., Krampe, R.T., & Tesch-Römer, C. (1993). “The Role of Deliberate Practice in the Acquisition of Expert Performance.” Psychological Review, 100(3), 363–406.
Gladwell, M. (2008). Outliers: The Story of Success. New York: Little, Brown and Company.
Hallman, N., Kowaleski, Z., Puvvada, A., & Schmidt, J.J. (2026). “What the Best AI Users Do Differently — and How to Level Up All of Your Employees.” Harvard Business Review, March 19, 2026. A joint study by the McCombs School of Business at The University of Texas at Austin and KPMG LLP. hbr.org/2026/03/what-the-best-ai-users-do-differently-and-how-to-level-up-all-of-your-employees
Hallman, N., Kowaleski, Z., Puvvada, A., & Schmidt, J.J. (2026). “Sophistication in GenAI Use: Field Evidence from a Large Firm.” Working paper, arXiv:2608.27364.
Hamel, G., & Zanini, M. (2018). “The End of Bureaucracy.” Harvard Business Review, November–December 2018.
Loudoun County, Virginia (n.d.). “Frequently Asked Questions: Data Centers – Land Use.” loudoun.gov/Faq.aspx?QID=1800, read October 8, 2026.
MacKinnon, D.W. (1962). “The Nature and Nurture of Creative Talent.” American Psychologist, 17, 484–495.
McCrae, R.R., Arenberg, D., & Costa, P.T. (1987). “Declines in Divergent Thinking with Age.” Psychology and Aging, 2(2), 130–137.
McCrae, R.R., & Costa, P.T. (1989). “Reinterpreting the Myers-Briggs Type Indicator from the Perspective of the Five-Factor Model of Personality.” Journal of Personality, 57(1), 17–40.
Medscape (2026). ‘A Return to Normalization’: Physician Compensation Report 2026. Survey of 5,916 full-time U.S. physicians across 29 specialties, September–December 2025.
Mueller, J.S., Melwani, S., & Goncalo, J.A. (2012). “The Bias Against Creativity: Why People Desire but Reject Creative Ideas.” Psychological Science, 23(1), 13–17.
Ng, T.W.H., & Feldman, D.C. (2013). “A Meta-Analysis of the Relationships of Age and Tenure with Innovation-Related Behaviour.” Journal of Occupational and Organizational Psychology, 86(4), 585–616.
Polanyi, M. (1966). The Tacit Dimension. New York: Doubleday.
Porter, M.E. (1996). “What Is Strategy?” Harvard Business Review, 74(6), 61–78.
Reeder, K., & Lee, H. (2022). “Impact of Artificial Intelligence on US Medical Students’ Choice of Radiology.” Clinical Imaging, 81, 67–71.
U.S. Bureau of Labor Statistics. “Nearly one in five older Americans in the labor force in 2025.” The Economics Daily, May 2026.
U.S. Census Bureau. Vintage 2025 National Population Estimates, by single year of age (NC-EST2025-AGESEX-RES). June 2026.
Wolpe, J. (1958). Psychotherapy by Reciprocal Inhibition. Stanford, CA: Stanford University Press.
Index
Each entry links to the passage it names. Numbers are chapters. Bold marks the passage where a term is defined.
AI Partnership, Foreword, 1, 2, 3, 4, 5, 6, 7, 9, App. A, App. B
defined, Glossary
AI-Driven Cost Savings, 2, 3, 6, 8, 9, App. A, App. B
defined, Glossary
converges, 2, 8, App. A, App. B, Glossary
commoditization, 2, 8, App. A, App. B, Glossary
See also Expertise-Driven AI Partnerships
ceiling, public
lowered by confronting the fear, Plates
ceiling your Expertise sets, the. See Elevator, the
Elevator, the, Foreword, 1, 3, 4, 5, 6, 7, 8, 9, App. A, App. B
defined, Glossary
Move One, the load, 4, Glossary
Move Two, breaking from the Ground Floor, 4, Glossary
Move Three, the climb, 4, Glossary
the Floors, 3, 4, 5, 7, 8, Glossary
the Ground Floor, 4, 7, Glossary
personality type and the ride, 4
entrepreneur, the. See founder, the
Expertise
defined, Foreword, 1, Glossary
earned, not manufactured, 1, 2, App. B, Glossary
capitalized, and why, Glossary
what Expertise actually is, 3
ordinary proficiency distinguished from, Glossary
Expertise Capital, 2, 5, 6, 7, 8, 9
defined, Glossary
on no balance sheet, 1, 5, Glossary
repricing, 2, 5, 7, 8, 9, Glossary
the investor’s view, 7
defined, Glossary
the Expertise Density reckoning, 6
the density ceiling, 5
headcount, and why it does not predict output, 5, 6, 7, Glossary
Expertise Platform, the, 5, 6, 7, 9
defined, Glossary
building it, 6
exponential node, and why the two are one person, 5, Glossary
Expertise-Driven AI Partnerships, Foreword, 2, 5, 9, App. A, App. B
defined, Glossary
compounds, 1, 5, 6, 7, 8, App. B
new work, 2, 5, 6, App. B, Glossary
resistant to commoditization, and why, 2, App. B, Glossary
See also AI-Driven Cost Savings
exponential, the, Plates, 1, 4, 5, 9, App. A, App. B
why it cannot be felt, 1, App. A
the Bartlett Proof, 1
Now the exponential is you, 1, 9, App. B
See also Bartlett, Albert A.
See also Cannon, Walter B.; Wolpe, Joseph
founder, the, 7
gold, the new, Foreword, 1, 5, 9, App. B
what it has become, Foreword, 5
invisible to most, Foreword, 1, 5, App. B
around everyone you work with, 5
Hinton, Geoffrey, 1
investor, the, 7
job losses. See AI-Driven Cost Savings
Medscape, 1
organization, the, 6
professional, the, 7
pyramid, the, Foreword, 3, 5, 6, 7, App. B
defined, Glossary
what it was for, 6
Expertise rationed by, Foreword, 5, 7, App. B
the question it cannot answer, 6
Sophisticated AI Adopter, 3, 5, 6, 7, 9
defined, Glossary
Sophisticated AI Adoption, the same standard as a state, 5, 6, 7, 8, 9, Glossary
Sophisticated State, 2, 4, 9, App. A
defined, Glossary
why hours stop feeling like hours, 4
the desire to continue, App. A
See also Csíkszentmihályi, Mihály
Tao An. See An, Tao
UT/KPMG study, 3, 4, 6, 7, App. A
the findings, App. A
what it establishes and what it does not, App. A
See also McCombs School of Business; Harvard Business Review
