Subscribe now

The second thought was quieter, and I promised we would come back to it. Here it is: a changed environment can void a pattern that had always held.

“Conseco closes deals” was not a bad assumption. It was a true one — verified by twelve years and eleven acquisitions, believed by nearly everyone whose job was to know. And then, in 1994, the Federal Reserve raised rates six times in a year, and the assumption did not fail the way assumptions usually fail. Nobody discovered an error in it. No new evidence arrived to contradict it. The world that had made it true simply stopped existing.

That distinction — between an assumption that is wrong and an assumption that has expired — is what this essay is about. The most dangerous feedback you will ever receive is the kind that looks like ordinary error and isn’t.

Two ways to be wrong

A variable can move while the surrounding structure remains intact. Demand rises, margins fall, interest rates change — but the basic model still explains how those variables interact. That is the normal condition, and everything in the last three essays applies to it: revise the assumption, redesign the test, run the loop again.

A regime shift is different. In a regime shift, refining the existing hypothesis makes you more precise while leaving you fundamentally wrong. You are polishing the optics of a telescope pointed at a sky that has been rearranged.

France spent the 1930s building the Maginot Line — a flawless, continuously improved answer to the previous war. The refinement was real. The precision was real. The question had expired.

The hard part is that from inside, the two kinds of wrongness feel identical at first. Losses are losses; surprises are surprises. The expired map does not announce itself. It sends you feedback that looks like noise, then like bad luck, then like a run of correctable errors — and by the time it looks like what it is, the people still running the old map have usually spent their capital defending it.

How a regime change travels

I have watched this happen enough times now to believe it has a shape. Regime changes don’t arrive all at once. They propagate — down a causal chain, in a consistent order. And the order matters, because each link wears a different disguise.

It starts when the production function changes — when the way a thing gets made, whether the thing is oil or credit or knowledge, is rewired at the root. This is the shock, and it is usually quiet. Production function changes are themselves often highly visible - like fracking - but their impact on what people actually experience (prices, products, jobs) is opaque at best, invisible at worst.

I watched horizontal drilling and fracking rewire the production of American oil and gas in under a decade, and watched the world’s energy map get redrawn behind it. I watched the financial crisis rewire lending: banks pulled back under new rules, and private credit, hedge funds, and the rest of the shadow system grew into the space they left — a different answer to the question of who makes a loan, and how. So this is not a once-a-century event. It happens over and over, at every scale. What is different this time is where the rewiring is happening. Machine intelligence is changing how answers get produced — and increasingly, how judgment itself does.

The second link follows from the first: the input that was scarce under the old regime becomes abundant. Every regime organizes itself around its binding scarcity — the expensive input, the one people spend lifetimes accumulating and institutions spend fortunes credentialing. When the production function changes, that is the input that gets cheap. When the internet arrived, it was information: the thing libraries and experts had rationed for centuries repriced toward zero in about a decade. And nobody experienced that as a warning. Abundance is celebrated. Free information, instant answers, cheap energy — it all looks like pure win, and nobody mourns a scarcity while it is dying, least of all the people whose fortunes were built on accumulating it.

The third link is where it gets interesting, because constraint never disappears. It moves — and where it lands, industries get born. When information became abundant, the capacity to organize it became the constraint. Engineers were scarce. Software was scarce. An entire generation of companies — search, databases, the whole SaaS economy — was built on exactly that bottleneck. Now the chain is running again. AI is making the organizing ability itself abundant: the cost of writing code, and with it the cost of making software at all, is heading toward zero. This time the scarcity is moving in two directions at once. Upward, to judgment — which question to ask, whether the answer fits this patient, this deal, this moment. And downward, to the physical layer that produces intelligence in the first place: chips, power, data centers. Underneath all of it is a pattern worth a sentence of its own: each regime’s new scarcity is what the next revolution industrializes.

This third link is also where the pain shows up, and where it gets misdiagnosed. Institutions keep investing in the old scarce input — their pricing, their credentials, their promotion ladders, their self-image were all built to accumulate it — while the real constraint has moved somewhere that isn’t priced yet. The gap between the two shows up as underperformance nobody can quite explain. There is always a local story available to explain it anyway.

The last link is the one that separates a regime change from everything else: the old model fails, and the failure is irreversible. Prices revert. Production functions ratchet. Nobody re-scarcifies information, un-fracks the shale, or puts machine intelligence back in the laboratory. When a model breaks against a moved variable, patience can save you — wait long enough and the world may come back to your map. When it breaks against a new production function, there is nothing to wait for. The world that made the model true is gone, and no price brings it back.

Lay the four links out and you can see why the trap works. At the first link the change is invisible. At the second it is celebrated. At the third it is misdiagnosed. Only at the fourth does it become undeniable — and the fourth is too late, because by then adaptation has stopped being an edge and become the price of admission. Whatever return there is to adaptability gets earned between the first link and the third, in the stretch where the evidence is still ambiguous and acting means trusting a chain of reasoning over a lifetime of data. “Wait for confirmation” is not a strategy here. Confirmation is what the fourth link is made of.

One refinement from the trading floor. Even a shock that eventually reverts can push a particular model through the fourth link. Rates came back down after 1994. “Conseco closes deals” never came back — the financing died, and the trust died with it, and trust ratchets too. So the diagnostic is not whether the shock will revert. It is whether your model can.

How you know

So how do you tell an expired map from a merely mistaken one — inside the window, while the evidence still points both ways?

Not by staring harder at the old model’s residuals. You put the model down and run the chain backward from where you stand. What has become abundant? That tells you which accumulated advantage is quietly repricing toward zero. What has become scarce — where did the constraint move? That tells you where the new returns live. And then the hardest one, the toll the fourth link charges: which lessons from the previous regime still travel, and which have to be demoted to historical observations from a world that no longer exists?

Share

These questions work because a model cannot absorb them. A model can absorb almost any anomaly — that is what models are for. But the grounding questions are not asking whether the map is accurate. They are asking whether the territory is still the one that got mapped.

There is a cadence discipline hiding here too. Reality does not schedule its regime changes around your annual review. The right frequency for re-asking these questions is set by the half-life of the load-bearing assumption, not by the calendar — and that half-life has been shortening for my entire career. In my first decade, a good map of a market might hold for years. The question now is not whether your answer is right. It is how quickly you can tell that yesterday’s answer has expired.

The test that changes the system

There is one further complication in markets, organizations, politics, and every other social system: even a well-designed test can alter the environment that returns the feedback. A trade attracts capital and moves the price. A promising business model draws competitors and changes the economics that made it attractive. A policy changes the incentives of the very people whose behavior produced the data used to design it.

The inversion market that cost me the AbbVie–Shire trade was reflexive in exactly this way. The companies announcing inversions in 2014 carried more combined assets — $319 billion — than every company that had inverted in the previous thirty years put together. That accumulation was not background; it was the mechanism. The wave of activity our deal belonged to created the political pressure that produced the Treasury response that killed our deal. We were not observing a fixed system and forecasting its output. We were participating in a process whose aggregate behavior was generating the response.

Reflexivity does not make feedback invalid. It means the feedback contains information about both the system and the system’s reaction to your participation in it. Whenever your actions combine with others’ to alter prices, incentives, regulation, or behavior, you have to ask: am I learning about a stable environment, or helping create a new one?

The frozen map

Which brings us to the tool we are all holding now, because it inherits this essay’s problem in its purest form.

A model is frozen at the moment its training ended — made, literally, from the world as it was, which means from the previous production function. It is excellent at folding a new observation into the old structure, which is the easy half of the work. What it cannot be trusted to do is notice, unprompted, that the structure has stopped applying — that the world it learned is no longer the world you live in — and choose re-grounding over one more revision. It can help you find evidence that the map has failed, if you ask. The judgment about what that evidence means is still yours.

Last essay I said we had automated the defensive voice. This is the second inheritance: we have automated the expired map, and rented it to everyone at once. There is a strange recursion in that. Every workflow reorganized around cheap answers moves the third link along — using the tool is the regime change propagating — while the tool itself, trained on the old world, keeps quietly assuring everyone that the old relationships hold. The same instrument speeds up the transition and slows down the noticing. Used with that understanding — asked to argue that the regime has changed, to write the version of events in which the old relationships no longer hold, to say what would look different if the map had expired — it earns a place in this part of the loop too.

Redraw, don’t revise

Most feedback asks us to revise the map: a better assumption, a sharper test, a more complete explanation. The loop we have built across these essays handles that case well. A regime shift asks for something different and much more expensive. Put the map down. Go back to the ground — the new production function, the new abundance, the new bottleneck — and draw again from the constraints up. The skill is not choosing one mode forever. It is knowing which one this moment requires, and being honest about how badly you want the answer to be “revise” — because revision lets you keep everything you have already built.

Running that judgment alone, inside one head, is hard. Running it with other people — a roomful of different skills, egos, and incentives, holding the same expiring map, needing to notice together, decide together, and redraw together without flying apart or quietly settling on the comfortable answer — is the hardest thing I know. That is where we go next.

Subscribe now

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

In 2014, AbbVie agreed to acquire Shire for $54 billion in one of the largest tax inversions ever attempted. We went deep on every part of the transaction we could analyze: the financing, the legal structure, the tax rules, the incentives, and the mechanics of how the deal would close. We understood those pieces extremely well, probably better than most people trading the situation.

Then the U.S. Treasury issued a notice that gutted the tax benefit on which the acquisition depended. AbbVie walked away, and the position moved hard against us.

For a long time, I described what happened as bad luck. A regulator had intervened in an unprecedented way during a live transaction. There was no established probability for that outcome, I told myself, and sometimes even excellent work loses money. There was truth in that explanation, which is what made it so comfortable. But it also allowed me to avoid the more important question: what, exactly, had the result taught me about the way I was thinking?

In the last essay, I wrote about designing an experiment under uncertainty: isolate the variable that matters, aim the test where your model is most likely wrong, and keep failure shallow enough to try again. But designing the test is only half the discipline. Eventually the world sends something back.

The experiment is not the end of the AQ process. It is reality re-entering it. Feedback does not conclude the loop; it sends us back through it — examining how we are interpreting the result, revising the explanations that might account for it, deciding what to test next. The sequence is not think, test, know. It is think, test, receive, inspect, revise, and test again. And the first thing feedback tests is not the hypothesis. It is your relationship to the hypothesis.

The defensive voice

When a result disappoints us, the mind rarely sits quietly and waits to learn. It immediately begins producing explanations: the market was not ready, the customer misunderstood the product, the regulator acted irrationally, the test was distorted by unusual circumstances. Any of those explanations might be correct. But they are also precisely what we would say if our real purpose were to defend the hypothesis rather than improve it.

This is why experimentation returns us first to metacognition. Before interpreting the answer, examine the instrument doing the interpreting. Am I treating this result as information, or experiencing it as criticism? Am I defending the analysis because it is sound, or because it is mine? Have I confused the amount of effort I invested with the likelihood that my conclusion is true?

Subscribe now

A strong opinion weakly held is still only a hypothesis. It is something you chose to test, not something you are required to protect. Reality is not taking sides or passing judgment on your intelligence. It is answering the question you asked.

None of this means celebrating every loss as “learning” and absolving ourselves of responsibility. Money, time, trust, and opportunity are real, and some failures come not from brave experimentation but from careless reasoning, ignored evidence, or exposure that never should have been taken. Adaptability is not a license to be wrong without consequence. It is the discipline of correcting error without allowing pride or shame to interfere with the correction.

Earning the outlier

Once you have separated yourself from the hypothesis, a second obligation follows: you have to account honestly for what came back.

In 1964, two radio astronomers at Bell Labs, Arno Penzias and Robert Wilson, encountered an irritating background hiss while using a large horn antenna in New Jersey. The noise appeared in every direction, at every hour, in every season. They assumed something was wrong with their equipment and worked methodically through the ordinary explanations. They ruled out interference from New York City, the galaxy, known radio sources, and flaws in the antenna itself. They found pigeons nesting inside, removed them, and scrubbed out what the birds had left behind. Still the hiss remained.

It took more than a year, and a conversation with physicists at Princeton, before they understood what they were hearing. The noise was the cosmic microwave background — the afterglow of the Big Bang. In 1978, they won the Nobel Prize for the static they had spent months trying to eliminate.

Their achievement was not that they immediately recognized a historic discovery. They did not. It was that they refused to discard the observation simply because it did not fit the model they already had.

That is the standard feedback imposes. Every material result must be confronted. It may eventually be explained as signal, measurement error, contamination, flawed design, or ordinary noise — and sometimes the honest answer is that it remains unexplained for now. But “outlier” is a conclusion that has to be earned, not a label we attach to inconvenient evidence. Call the discipline what it is: earning the outlier.

The failure has an institutional version, and it is quieter. In 1985, three scientists from the British Antarctic Survey published ground measurements from Halley Bay showing a massive seasonal hole in the ozone layer over Antarctica. The awkward part was that NASA satellites had been passing over the same spot for years, and the low readings were sitting in the data. The processing software had been taught to treat values that low as probable instrument error and set them aside for review. Nobody decided to ignore the ozone hole. Somebody had decided, years earlier and for good reasons at the time, what an implausible reading looked like — and the filter did the rest, automatically, season after season. The most dangerous amputations are the automated ones, running on a rule nobody remembers writing.

Before an experiment, simulation asks what else could be true. Afterward, it asks which explanation can account for the full result — including the part you would most like to remove. Most anomalies are not revolutions, and not every stray observation should overturn a working model. But you cannot preserve the model by amputating everything that does not fit it. Either explain why the observation deserves to be excluded, or change the explanation until it can hold what reality has shown you.

What feedback usually says

An experiment rarely delivers a simple verdict that your entire model was right or wrong. More often, it tells you which part of the explanation needs to change. The central hypothesis may survive while one assumption fails. The evidence may strengthen an alternative you had considered unlikely. It may reveal that the test moved several variables at once and taught you less than you expected. Or it may leave two explanations standing, requiring a sharper test to separate them. The result of one experiment becomes the raw material of the next round of model-building: which assumption failed, what is now more or less plausible, what new test would make the remaining explanations produce different answers.

The experiment does not close the question. It improves the next question.

The silent test

Usually the process advances through modest revisions — a better assumption, a more complete explanation, a more discriminating test. But sometimes the loop produces nothing at all.

A silent test should first make you suspicious of the experiment itself. Did we isolate the variable that mattered? Was the feedback interval long enough? Could we even observe the relevant outcome? Did we test the load-bearing assumption, or something nearby that happened to be easier to measure? It is dangerously easy to declare a problem unresolvable when you have simply asked the wrong question. But the opposite error is just as expensive: continuing to dig because you assume enough work must eventually produce an answer.

That was the deeper lesson I finally took from AbbVie–Shire. The investment skill I had built over decades was synthesis: gathering fragments from incomplete sources and assembling a picture sharper than any single source could provide. It worked especially well in situations like antitrust, where the outcome, however uncertain, emerged from an observable structure — customers and competitors supplied evidence, precedent offered comparison, the questions judges asked could be read alongside other signals. No single source told you the answer, but enough independent observations gradually made the picture legible.

The Treasury decision offered no such structure. There were few meaningful participants to canvass, little precedent governing the intervention, and no experiment that would reliably reveal how officials would respond. The determining variable was not merely unreadable from where we sat — it was still being made. We understood everything that could be analyzed; the outcome lived in a decision that was still forming.

The quality of our work did not protect me from that distinction. In some ways, it made the mistake worse. The deeper we went into the knowable parts of the deal, the more confidence I felt in the whole. I had confused mastery of the surrounding mechanics with resolution of the variable that actually determined the outcome.

Put my deal beside the hiss in that antenna and you have the two great errors of listening, facing each other like mirror images. Penzias and Wilson nearly deleted a signal that was there. I kept hunting for one that was not. One error amputates the inconvenient answer; the other refuses to accept silence as an answer. Knowing which side of the mirror you are on is most of the skill. When a well-designed test keeps coming back empty, the absence is itself information: some uncertainty cannot be resolved, and some can be resolved in principle but not by you — not from your position, with your information, your tools, your time. The adaptive response is not to force false precision out of a silent system. It is to recognize the boundary and move toward a problem where your work can actually compound.

The voice, automated

One more thing about the defensive voice: we have now automated it. Today’s models are trained partly to give the answers their human raters prefer, and raters prefer agreement to friction. Ask a fluent model to defend your thesis after a failed test and it will explain, convincingly, why the experiment was unfair, the data anomalous, the failure irrelevant — self-justification at scale, on demand. OpenAI withdrew a version of GPT-4o in 2025 after it tipped into open flattery. The same tool pointed the other way — attack the thesis, generate the explanations I have not considered, design the test that would separate them — earns its place in the loop. Just remember the ozone filter: the machine will apply whatever rule about implausible answers it has absorbed, and it will apply it automatically.

What comes back next

Receiving reality’s answer, then, is two disciplines run in sequence. First, get your self out from between the result and the interpretation. Second, account for everything that came back — earn every outlier, respect every silence. Revise the explanation until it holds the evidence, and return to the world with a better question. The purpose of experimentation is not to prove that the hypothesis was right. It is to make the next model better.

But sometimes what comes back is stranger than a failed assumption or a silent test. In the essay on strong opinions, I asked you to hold two thoughts from the Kemper story and promised we would return to the second: that a changed environment can void a pattern that had always held. Sometimes feedback is not telling you that an assumption was wrong. It is telling you that the system that generated all of your assumptions has changed — that the map is not mistaken but expired. Most feedback asks us to revise the map. A regime shift requires us to redraw it.

That is where we go next.

Subscribe now

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

I agree with almost all of it. Civilizational technologies deserve institutions that match their scale. My concern is that we are evaluating only one side of the problem.

Today’s AI safety debate is understandably focused on the behavior of the models themselves. Can they help build bioweapons? Can they deceive users? Can they evade human control? These are serious risks, and they deserve serious institutions.

But a system can pass every one of those tests and still leave the people who use it worse off.

A recent randomized study from researchers at Wharton illustrates the point. Nearly one thousand high school students were given an AI math tutor. Half received a version that readily supplied answers. The other half received a version that offered hints while requiring students to continue working through the problem themselves. While using the tutor, both groups improved. Then the tutor was removed and the students were tested independently. Those who had relied on the answer-generating tutor scored seventeen percent worse than students who had never used AI at all. Students who used the guided tutor showed no such decline.

Run that tutor through today’s frontier safety evaluations. It doesn’t help build a weapon. It doesn’t deceive its users. It doesn’t seek to escape human oversight. It is exactly the sort of capable, helpful assistant we are trying to build.

Yet one version quietly left the people who depended on it less capable of solving problems on their own.

One might reasonably object that this is a deployment problem rather than a model problem. I think that’s exactly right.

The model was not the failure.

The environment built around the model was.

Subscribe now

That distinction matters because I suspect we are thinking about AI in the wrong category. We continue to speak of it as though it were primarily a tool. History suggests something different. The most consequential technologies eventually become environments.

Agriculture did more than increase food production. It reorganized how people lived, how families were structured, how labor was divided, and how civilizations formed. The printing press did more than reproduce books. It changed memory, authority, education, and religion. Social media did more than lower the cost of communication. It reshaped attention, friendship, and the social environments in which identities developed.

Artificial intelligence will almost certainly do the same. Its greatest effect may not be the answers it produces but the cognitive environment it creates for the people who rely on it every day.

Anthropologists have long argued that human beings are shaped not only by what they believe but by the environments they inhabit and the practices they repeat. We gradually acquire habits of attention, judgment, and interpretation that reflect the worlds around us. In Pierre Bourdieu’s language, our habitus is formed through repeated interaction with our environment. Whether or not one accepts his broader theory, the underlying insight is difficult to deny: people adapt to the worlds they repeatedly inhabit.

Technology therefore never remains external to us. It slowly becomes part of the process by which we become who we are.

Seen from that perspective, the Wharton study is not simply an education study. It is evidence that different AI environments produce different cognitive outcomes, even when the underlying model is essentially the same. One tutor preserved productive struggle. The other quietly removed it. The important variable was not intelligence. It was the developmental environment.

The same pattern is beginning to appear elsewhere. A recent study following nearly twenty-seven thousand secondary school students in China found that students who increasingly relied on AI to complete homework earned higher homework marks but performed substantially worse on closed-book examinations, with the largest declines among the strongest students. The mechanism appears remarkably consistent. When the machine performs more of the cognitive work, the human performs less.

None of this is inevitable. The difference between the two tutors was not capability. It was design.

A model can draft the memo for a junior analyst, or it can coach her through drafting it herself. It can answer every question immediately, or it can preserve the productive struggle through which judgment is formed. Those choices may seem like interface decisions, but they are really decisions about the kind of cognitive environment we want to create.

This is where I think our conversation about AI safety needs to expand.

Much discussion of artificial intelligence assumes that as intelligence becomes abundant, many of today’s constraints will disappear. History suggests something more subtle. Every technological revolution removes some constraints while creating others. Electricity made light abundant but transformed how people used time. The internet made information abundant, making attention comparatively scarce. AI will make answers abundant. The scarce resource will increasingly be the human capacity to ask worthwhile questions, recognize when familiar models no longer describe reality, distinguish genuine understanding from fluent retrieval, and construct meaning rather than simply consume it.

These are often described as “soft skills.” I think that misses the point. They are becoming the binding constraint. In a world where almost anyone can obtain an answer instantly, the comparative advantage shifts to the person who knows which questions are worth asking and when the answer, however fluent, is incomplete.

That is why I believe AI safety cannot be defined solely by the behavior of the model. It must also include the long-term effects of the environments those models create.

Hassabis is right that civilizational technologies require civilizational institutions. But those institutions should evaluate more than whether frontier models remain aligned with human intentions. They should also ask whether prolonged interaction with those systems strengthens or weakens the distinctly human capacities that will matter most once intelligence itself becomes abundant.

The central challenge of AI is not simply building intelligent machines. It is deciding what those machines make of the people who use them. That decision is already being made, one design choice at a time. The only question is whether we make it deliberately—or simply inherit it by default.

Subscribe now

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

You can listen to the episode here. Thanks Nick and Beezer for having me on to talk about the second cognitive revolution, the difference between uncertainty and risk, and my upcoming book—The Adaptability Quotient.

Subscribe now

In the autumn of 1903, two teams in the United States were racing to build a flying machine.

The first was led by Samuel Pierpont Langley, Secretary of the Smithsonian Institution and one of the most credentialed scientists in America. He had been working on aerodynamics for seventeen years. He had successfully flown unmanned models. The War Department, impressed by those models, had granted him fifty thousand dollars to build a full-scale, man-carrying aircraft — a sum that, with another twenty thousand from the Smithsonian, came to roughly two million in today’s money.

He built a fifty-three-horsepower engine, an extraordinary feat of engineering for the period. He launched the finished machine from a catapult mounted on a houseboat in the middle of the Potomac River.

On October 7, 1903, the Great Aerodrome plunged straight into the water. The pilot, Charles Manly, was pulled out alive. Langley blamed the catapult, rebuilt, and tried again.

On December 8, 1903, the second attempt broke apart on launch and dropped into the river again. Manly nearly drowned. The newspapers were merciless. The New York Times called it a “Flying Machine Fiasco.” One congressman remarked that “the only thing he ever made fly was government money.” Langley quit.

Nine days later, on December 17, two bicycle mechanics from Ohio named Wilbur and Orville Wright flew a powered aircraft of their own design for fifty-nine seconds at Kitty Hawk, North Carolina. They had spent roughly one thousand dollars of their own savings.

Same goal. Same year. Same physics. Vastly different outcomes.

It is tempting to read the story as a parable about scrappy underdogs and bureaucratic experts, but that is not really what happened. Langley was not foolish, and the Wrights did not have a better theory of flight. The best aerodynamicists of the era doubted both teams. What separated the two programs was not the hypothesis. It was the method — the design of the experiments themselves.

Last time, we left a hypothesis ready to be tested. The harder half of Phase II — committing to a strong opinion held weakly — produces a candidate worth taking out into the world. But, as we noted, the point of the test is to learn, not to win. Designed experimentation is the only honest way to convert uncertainty into usable knowledge, because in genuine uncertainty there are no probabilities to optimize over yet — only experiments to run.

What separates a good experiment from a vanity test is the design.

Start with what the Wrights actually did. They did not, as Langley did, scale up an unmanned success directly to a manned aircraft.

They began with kites in 1899. Then gliders in 1900, 1901, and 1902 — hundreds of flights, on a sand dune at Kitty Hawk, where a bad landing cost them a wing or a bruise. They built their own wind tunnel in 1901 because the published aerodynamic tables of the day, which everyone else was using, turned out to be wrong. They tested roughly two hundred wing shapes inside it. They worked out a three-axis control system — a way to roll, pitch, and yaw the aircraft — before they ever added a powered engine.

By the time they bolted a twelve-horsepower motor onto the Flyer in December 1903, they had spent four years doing one thing: learning what they did not yet know, cheaply, and without ever putting the whole program at risk in a single test.

That posture — testing to find out where you’re wrong, not to confirm you’re right — is the first principle of a well-designed experiment.

It sounds obvious, and almost no one does it. The default mind, even a careful one, is built to confirm. You construct a hypothesis you believe in, then you go looking for evidence that supports it, and you treat the evidence that doesn’t as noise. The discipline of a real test runs the other way. You design it the way the Wrights built their wind tunnel: specifically to expose the wrongness in your current model, fast and cheaply, before the cost of being wrong becomes catastrophic.

The second principle is the one that most directly killed Langley: optimize for shallow failure, engineer out deep failure. A glider crashing in soft sand on Kitty Hawk is a data point. A full-scale powered aircraft cartwheeling into the Potomac at high speed, with most of your funding strapped to it, is a knockout.

Both involve a flying machine hitting the ground, but they are not the same kind of failure. One returns you to the workshop. The other ends the program. Langley designed his tests to settle the question in one or two attempts. The Wrights designed theirs so that no single attempt could end them.

This is the most under-appreciated discipline in experimentation: the thing you cannot lose is the ability to test again. Protect the loop itself.

Whatever you are running an experiment to learn, it is worthless if the experiment, when it fails, takes out the very mechanism by which you would have learned the next thing. In financial markets, the version of this principle is sizing a position so a bad week is information rather than an ending. In a startup, it is shipping the smallest version of the product that would tell you whether the demand is real, before you build the infrastructure that assumes the answer. In science, it is the difference between an experiment that could plausibly fail in three different ways and one that has to work or the lab closes.

The third principle is something the Wrights did almost obsessively: break the big problem into nodes, and test one variable at a time.

Powered flight involved at least three separable problems — lift, propulsion, and control. Langley treated them as one problem, attacked them simultaneously, and tested the integrated whole. The Wrights treated them as three problems, solved them separately, and isolated each variable enough that they could read the signal from the test.

When you change ten things at once and the result improves, you cannot say which of the ten mattered. When the result worsens, you cannot say which of the ten to back out. The most common failure mode of corporate “transformations” is exactly this — twenty initiatives launched simultaneously, none of them isolated enough to teach you anything when the dust settles.

There is one more principle worth naming, which Astro Teller — the head of X, Alphabet’s moonshot lab — has been articulating publicly for the better part of a decade.

Teller frames it with a simple thought experiment. Suppose you wanted to teach a monkey to stand on a ten-foot pedestal and recite passages from Shakespeare. What would you spend your time on first?

Most people, Teller says, instinctively reach for the pedestal. The pedestal is concrete, tractable, satisfying to build; you can see the progress. But the pedestal is not where the project lives or dies. The monkey is.

The monkey is the part that is most likely to be impossible. If the monkey cannot be trained to recite Shakespeare, the entire project ends — and the pedestal becomes a beautifully built piece of waste.

At X, the rule is tackle the monkey first. Identify the single part of your idea most likely to kill the whole thing, and run at it before anything else. The teams there are formally rewarded — with cash bonuses — for killing their own projects early, on the theory that an idea you can disprove cheaply in six months is worth more than the same idea you spend three years building toward and only then discover is impossible.

Langley spent his program building the pedestal. He kept refining the engine, the launch mechanism, the houseboat. Power was the part he could see and measure. The monkey — control — was the part he barely engaged with.

The Wrights, with no government money and no Smithsonian title, went straight at the monkey. Wing-warping, rudder, elevator — control, control, control — for four years, before they ever made a powered flight. When the time came to add the engine, the monkey was already trained.

This is a more general pattern than aviation. The hardest part of almost any consequential bet — the one most likely to determine whether the whole thing works — is rarely the part that looks impressive on a slide. It is the part nobody wants to test, because testing it might mean discovering that the bet doesn’t work. The discipline is to run at exactly that part first.

You can hear in all of this an echo of what we discussed last time. The Wrights’ approach is a kind of asymmetric pruning, made operational: they pruned hard on the left tail by structuring failure to be cheap, and kept optionality on the right tail by sustaining their ability to keep iterating for as long as it took. Langley’s approach was the opposite — concentrated downside, almost no margin to recover.

There is an AI-era twist on this, and it is worth saying directly.

The convergence machines we now have access to — the LLMs and the systems built on top of them — are extraordinary tools for compressing a certain kind of search. They will tell you, very fluently, what the average answer to your question has been.

What they will not do, because they cannot, is run the test for you in territory the average answer does not yet cover. When the question you are asking is one the model has seen before, you can borrow its answer. When the question is genuinely new, the test still has to be yours.

The design of that test — what you choose to falsify, how shallowly you can afford to fail, which variable you isolate, which monkey you train first — is exactly the part that does not get automated.

A well-designed test, then, has four properties. It looks for where you are wrong. It is small enough that failing it teaches you something rather than ending you. It isolates one variable at a time, so the signal stays clean. And it goes at the hardest, most-likely-to-fail piece of the idea before anything else.

Get all four right and you have a Wright Flyer. Miss them, and you have an Aerodrome at the bottom of the Potomac.

There is one more piece of the discipline still ahead of us. Designing the test is the engineering work; running it is the harder work.

Once the experiment is in the world, the harder questions begin: how do you tell signal from noise? How do you keep yourself from quietly rewriting the test once the data starts arriving? And — the deepest version of the question — what do you do when the experiment returns a result you understand perfectly well, but the world itself has shifted while you were running it, and the question you were testing is no longer the right one to ask?

That is where we go next.

Share

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

But there were red flags, and they were not small. Kemper was nearly three times Conseco’s size — this was a minnow swallowing a whale. The financing required not just bank loans but asset sales by Conseco to raise the cash. And then the ground moved. In 1994 the Federal Reserve shocked the market with six rate hikes, 250 basis points in all, in what came to be called the Great Bond Massacre. For two insurance companies sitting on enormous bond portfolios, that was direct balance-sheet damage. Worse, it made the financing harder on every front at once: banks were nursing their own losses and lending at far higher rates than anyone had modeled, and the assets Conseco needed to sell were suddenly worth less in exactly the same rate move.

I was completely new to markets — what an introduction, starting in the year of the worst bond rout in a generation. And precisely because I had no history, no settled pattern to lean on, I kept asking the questions the veterans had stopped asking. Why does Conseco have to do this deal? How do the asset sales actually get done in this environment? How can Kemper, and the assets Conseco needs to sell, be worth what everyone assumed, given where rates have gone? The answer was always the same: “Conseco closes deals. They have to, or no one will trust them with the next one.”

Eventually I sold out at a small loss — one of only four losing trades in my first eleven hundred. Not because I was certain it would break. Because I could not convince myself the old pattern would survive the new environment, and I had no history telling me it would. On November 18, 1994, Conseco and Kemper terminated the merger. The financing had never come together.

Subscribe now

The lesson was not that I had been prescient. I wasn’t. The lesson was twofold. First, the signals were there — visible, nameable, sitting in plain sight — and almost everyone explained them away because the pattern had always held before. The most expensive failures are rarely the false alarms that blow up loudly. They are the missed signals: the things that were there to be seen, and weren’t, because the consensus had a comfortable story for ignoring them. And second, quieter but deeper: a changed environment can void a pattern that had always held. “Conseco closes deals” was true until the ground moved. The rate shock didn’t just damage the balance sheets — it broke the world the pattern depended on. Hold both thoughts; we will come back to them.

Last time, we left simulation — Phase II of AQ — open. But wandering is the start of the work, not the whole of it. At some point, usually because a deadline or a competitor forces your hand, you have to pull a single hypothesis out of all that open possibility and commit to it — while staying ready to drop it the moment reality says you are wrong. That is the harder half of simulation. Not because it takes more thinking, but because it takes a different discipline. Grit is what wards off the premature closure of possibilities; curiosity is what gives you the drive to keep exploring them. The move from wandering to direction asks for both at once — and then for the nerve to commit anyway.

What you are after is not certainty, not a forecast, not a position to defend. It is what I call a Strong Opinion, Weakly Held (SOWH): a hypothesis sharp enough to test in the real world, and held loosely enough to revise when the world answers back. And “weakly held” does not mean lightly acted on — you commit fully, you act, you put capital or reputation behind it. The weakness is only in your attachment to being right. You marry the decision; you never marry the thesis. Most people miss it in one of two ways. Some commit too early, pushed by the simple discomfort of an open question. Others never commit at all, calling it “more research” long after the moment to decide has passed.

Getting it right turns on a handful of distinctions. These are the ones that have mattered most to me.

Start with intuition and insight. Both arrive as a feeling of certainty, which is why they are so easy to confuse — but they are not the same thing. Intuition is fast work: your mind matching the moment against patterns it has already seen. In familiar territory, that is a gift. Insight is slow work: the forging of a new pattern that was not there before. And it rarely arrives on command — it shows up as the “aha” in the shower, on the walk, in the moment after sleep, once the conscious mind has stopped grinding and the pieces are free to rearrange themselves. Intuition hands you the most likely answer based on the past. Insight gives you a new answer for a future that will not look like the past. In genuinely new territory, a confident hunch is usually intuition in disguise — yesterday’s answer dressed up as revelation. The discipline is to slow down long enough to let the real restructuring happen.

The engine underneath insight has a name. The philosopher Charles Sanders Peirce called it abduction, and it is worth separating from its two better-known cousins — better known because a stable world runs fine on them, while abduction is the one you need when the rules themselves are in question. Deduction applies a rule to a case: all merger arbitrage carries deal risk; this is a merger arbitrage; therefore it carries deal risk. Induction builds a rule from many cases: I have watched a thousand deals, and the ones financed this way tend to close, so the next one probably will. Both are powerful — and both assume the rules of the world are stable. Abduction does something stranger and more necessary. You reach for it when the facts stop fitting your model — when something surprises you. Instead of hunting for the few facts that still prop up your old thesis, you make room for a new explanation that accounts for all the surprising facts at once. You do not get to choose which facts to honor; the explanation has to hold the whole set. It is the leap to the best available account of a confusing situation — offered not as a verdict, but as something to go and test.

The most heroic example I know comes from a Vienna hospital in the 1840s. A doctor named Ignaz Semmelweis ran a maternity ward where roughly one mother in six died of childbed fever. In the ward next door, run by midwives instead of doctors, the death rate was a small fraction of that. Nothing in the medicine of the day explained the gap. The facts simply did not fit the model — and the model was the consensus of an entire profession. Then a colleague cut his finger during an autopsy, fell ill, and died — with the very symptoms that were killing the mothers. Semmelweis made a leap that, in its time, sounded like madness. The doctors, unlike the midwives, went straight from dissecting corpses to delivering babies, carrying some invisible “cadaverous particle” on their hands. Consider how that landed: he was telling the most respected physicians in Europe that they themselves were the cause — that something no one could see, that no theory predicted, was on their hands and killing their patients. Germ theory was still two decades away. He was ridiculed, dismissed, driven out. And he was right. He ordered handwashing in chlorinated lime, and the deaths collapsed to match the midwives’ ward.

That is abduction at full stretch: an unseen cause, inferred because nothing else could hold all the facts, made against the fierce resistance of everyone whose model it overturned — and then put to the test. It also happens to be the move today’s AI cannot be trusted to make on its own. These systems are extraordinary at correlation — at converging on the most probable next pattern. But abduction runs the other way: it requires noticing that the probable pattern has failed, and inventing a new one to fit facts that have no precedent. As convergence gets automated, abduction is the part that stays ours.

Share Early Adapters

Now the hypothesis is sharp enough to take into the world. But before you do, remember what the test is actually for. The point of the experiment is to learn, not to win. You are still mapping a fast-changing world — and even as the possibilities come into view, their probabilities stay unknown. (How to act in exactly that situation — the real heart of uncertainty — is the subject of the next two installments.) You cannot optimize what you have not yet learned. And if learning is the goal, then the one thing you cannot afford is a knockout — a loss large enough to end your ability to keep learning at all.

So before you run the experiment, you simulate the failure first. You play out how it could go wrong on paper, where it is free, and you kill or hedge the catastrophic outcomes before you ever spend a real probe testing it. Simulation is where you remove the fatal branches; the experiment is only for what survives that.

Which brings us to the two ways an experiment can be wrong. A Type I error is a false positive — acting on something that turns out not to be there, the smoke alarm shrieking at burnt toast. A Type II error is a false negative — missing something that is real, the alarm staying silent while the house fills with smoke. The trap is that the two are rarely equal in cost. Type I errors are loud and usually recoverable: they announce themselves, and you correct course. Type II errors are quiet. They do not raise a hand. They compound in the dark while everyone congratulates themselves on the risk they were smart enough to avoid. Conseco-Kemper, for almost everyone in it, was a Type II in the making — a real signal sitting inside a comfortable consensus, with the alarm switched off.

And the loud errors, the Type I kind, get a worse reputation than they deserve. In 1990, a group of Apple’s best engineers spun out a company called General Magic to build a handheld “personal communicator” — in every way that mattered, the smartphone, roughly fifteen years before the networks, screens, and batteries to make it work existed. The company folded. But the alumni went on to build the iPhone, create Android, and found eBay. One failed bet seeded a decade of the industry’s biggest wins. From the outside, a Type I error looks like money set on fire. From inside the loop, it is often tuition.

Put the two together and you get the rule for narrowing your options at the end of simulation: do not prune both ends the same way. On the downside, where ruin lives, cut hard — kill or hedge anything that could end the game, however unlikely, because the catastrophic loss is the one you do not survive to learn from. On the upside, do the opposite — preserve the cheap, asymmetric bets, the ones whose downside is small and bounded but whose payoff is large and, crucially, irreversible if you walk away from it. Pruning the tree early feels efficient, but the branch you cut to save effort is often the one that held the whole value — and unlike a loss, a foreclosed option never announces what it cost you. You simply never find out what the unpicked branch would have paid. Jensen Huang spent years funding NVIDIA’s parallel-computing bet through quarters that looked like waste, refusing to prune an upside nobody else could see — until AI made it the foundation of an entire era. Cut for ruin; keep for upside. Call it asymmetric pruning.

So pull the threads together. In a world of genuine uncertainty, running an experiment on a poorly formed hypothesis is dangerous — you burn a probe and learn nothing. But even a well-formed hypothesis has to be prepared before it meets the world: you simulate how it could fail, you ask which kind of error would cost you most, and you structure the experiment so that no avoidable mistake can end your feedback loop before the real answer arrives. That is what it means to commit and stay open at the same time — to hold the decision tree open long enough for reality to resolve the possibilities and the probabilities, rather than closing it early on a story that happened to be comfortable.

A strong opinion only earns the second half of its name — weakly held — once it meets reality and gets revised by what comes back. The test produces data; the data reshapes the hypothesis; the new hypothesis sends you back into the search you thought you had left. How we actually run those experiments, build the feedback loops, and keep watching for the regime change that quietly rewrites the whole map — that is where we go next.

Leave a comment

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

But then let me tell you something your commencement speaker probably neglected to mention:

The map you just spent four years mastering is already out of date. (I didn’t say it’s useless — merely out of date. A distinction that is not insignificant…)

The generations that came before you grew up with a relatively stable deal. Study hard, pick a lane, get the credential, climb the ladder. Obviously no outcome could ever be guaranteed, but the path to success, at least according to the textbook, was legible. You could see the rungs on the ladder, strategize about how to climb higher, estimate your odds of getting there.

I’m afraid that deal is gone. Not fading away — gone.

For many of the graduates stepping into the world at this uncertain moment, the job you’ll hold in 15 years probably doesn’t have a name yet. The industry that will define your peak earning years may not even exist. Not because AI is coming for your career in some distant, theoretical future — because, across many industries, it is already sitting at the entry level, doing the work that used to be how young people learned their craft. How far up the ladder it will climb, and how quickly, are still subject to fierce debate. But in the meantime, the social architecture around all of this — who gets ahead, what signals competence, what a “career” even looks like — is being rebuilt from scratch in real time.

I’m not saying any of this to scare you. I’m saying it because the honest response to this world is very different from the advice most graduates receive, and you deserve the honest version.

Here it is, in a nutshell: the most valuable thing you bring with you, as you move on from this institution of learning and into your chosen field, is not a credential. It’s a tested process for learning faster than the world is changing around you.

Let me explain what I mean.

Share

There are two kinds of problems.

Convergent problems have “right” answers. These are most of the problems you’ve been solving, in various forms, for the past 16 years (give or take) — exams, equations, problem sets, papers written to satisfy a rubric. Convergent problems reward knowing things and often render quick feedback: you got the grade, you moved on.

Divergent problems, on the other hand, have no objectively “right” answers, only better and worse ways of navigating uncertainty. What work will matter to you? What problems are worth your sustained time and energy? Where do your particular capabilities meet something the world actually needs? These questions can’t be answered in a classroom. They can only be explored through experience — by running small experiments on your own life, watching what happens, and updating your understanding as you go.

The shift you’re making today, whether you feel it yet or not, is from a world of convergent problems to a world of divergent ones. Which means the muscle you built for the first kind — absorbing, optimizing, executing — will not carry you as far as it used to.

I want to say something specific about AI, because you’re going to hear a lot of advice about it. Much of it will be exhausting, and most of it will be wrong. The wrong advice is: learn to use AI tools. Of course you should. That’s table stakes, not strategy. In a world where answers are getting cheaper, judgment is getting more valuable.

The genuinely important insight is harder to articulate, and even harder to hear. It turns out AI is extraordinarily good at solving convergent problems — via retrieval, synthesis, pattern-matching, producing fluent answers at speed. But when you’re navigating a divergent world, every time you outsource the messy middle — the struggle to frame a problem, generate a hypothesis, stress test your own reasoning — you’re both using the wrong tool and skipping the only workout that builds judgment. And judgment, right now, is one thing AI cannot commoditize.

I know it sounds counterintuitive, and it may be cold comfort in a moment when practical considerations (like rent) are about to get very real — but, in many ways, the uncertainty of this moment is a gift.

When I graduated, I was convinced I had the path figured out. But it turns out I was wrong in the most useful possible way — because I was wrong in environments where the cost of being wrong was low and the learning was high.

Looking back, the most important thing I did early on wasn’t making career choices correctly — it was staying in motion, paying close attention to what I was observing along the way, and being willing to update. Over the years, it was the jobs that didn’t work out that taught me what I actually needed — and the projects that failed which clarified what I was genuinely good at (versus what I thought I should be good at). Throughout my career, the mentors who pushed me hardest have been worth 10 times the ones who confirmed whatever I already believed.

This is what I’d call running experiments on your own life. Not drifting without direction, to be clear, but experimenting — which you might think of as motion with a hypothesis: I think I might be good at this; let me find out. I think this environment might bring out my best; let me test it. The goal isn’t to find “the” answer, because there isn’t a binary right or wrong one. The goal is to resolve uncertainty — to learn something real about yourself and the terrain — and then move forward with the benefit of that new information.

Early in your career, the downside of lingering in exploration is almost always smaller than it feels. Because here’s the thing: you have time. Most decisions at your age are more reversible than they seem. What you don’t have, and what you can never get back, is the particular freedom of these early years — the freedom to be a beginner, to move between worlds, to let the map develop before you commit to one road.

Use that freedom while you can. It will diminish before you know it.

A few more things I wish someone had said to me when I was in your shoes:

Chase feedback velocity, not brand names.

Take it from me that the most important variable in your first few jobs is not the prestige of the logo on your business card — it’s the speed of the feedback loop. Somewhere you can make a real decision, see what happens, and learn something by Friday is worth more than somewhere you’re protected from consequences for three years. Small organizations, demanding founders, chaotic environments — these are not consolation prizes for fresh graduates or people who are just starting out. For someone who wants to learn fast, they’re the main event.

Build proof of thought, not just credentials.

Your resume tells people you can follow instructions. What actually attracts opportunity is evidence of how you think — essays, projects, public work, documented learning. The internet has made it possible for a 24-year-old with genuinely original ideas to be found by the people who care about those ideas. That wasn’t the case, by a long shot, when I was starting my career — and it opens a universe of untold potential.

Ask better questions of people who are ahead of you.

Not “what should I do?” — which is prescriptive and produces checklists that don’t transfer. Instead, ask something like: “How did you think about the tradeoffs when you sat where I am?” or “What did you believe then that you’ve since changed your mind about?” or “What did you get wrong early that turned out to matter?” Your goal is to get a glimpse of the underlying mental model more than any individual answer. (Someone else’s answer applied to your situation is usually wrong — but insights about how they model the world, internalized, are yours to use.)

Seek intellectual friction, deliberately.

Today’s information environment is optimized to confirm what you already think, recommend what you already like, and keep you warm and comfortable in the embrace of a worldview you already hold. But comfort is not a neutral state — it is cognitive stagnation. So, get moving. Read things that are a hundred years old, talk to people who hold views that genuinely disturb you and try to understand how a serious person arrived there. If your model of the world isn’t being meaningfully challenged and rebuilt on a rolling basis, you are not thinking — you’re coasting by on someone else’s algorithm.

The people who do best in a world like this one share a few things. They are genuinely curious as a default orientation toward problems. They are honest about what they don’t know, including what they don’t know about themselves. They learn quickly from friction, including the friction of being wrong. And they build judgment slowly, via accumulated experience and genuine reflection, rather than looking for shortcuts to the answer.

Subscribe now

None of this is exotic, and none of it requires genius. What it amounts to is a kind of practice — something you build deliberately, like a discipline, over time.

Today, you leave a place where much of the map was provided. Tomorrow, the map is wide open and the work gets harder.

But that isn’t bad news. It’s the actual beginning.

The credential you’ve just earned has gotten you this far — to the starting line. What happens next depends on something no credential can give you: the willingness to stay curious, stay honest, and keep updating — even when the update is uncomfortable, even when it means admitting the map you had was wrong.

The future doesn’t belong to the people who predict it correctly. It belongs to the people who learn fast enough to help shape it.

Congratulations. Go find out what you’re made of.

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

Last night, you probably saw the smaller version of this. You opened a food app and looked at three new restaurants you had never tried, before ordering from the place you have ordered from many times. Or you scrolled through several hotels for an upcoming trip and booked the one you stayed at last time. Or you spent twenty minutes scrolling through new shows on a streaming service before re-watching something you had already seen.

There is a name for this: the explore-exploit trade-off. How long do you keep looking for something better, versus when do you settle for what you have already found? You navigate it every day, usually without registering that you are doing it. The math even suggests an optimal answer — explore about 37 percent of your options before committing to the best you have seen. In practice, the real world is messier than that. But the trade-off is everywhere.

Subscribe now

Many of the consequential decisions in life are similar problems dressed up differently. Most catastrophic decisions are not made by people who fail to do the math. The math is usually fine. What fails is upstream of all the careful work: the frame inside which the work was done was never pressure-tested. The option space collapsed before anyone paused to ask what the option space could have been.

Last time, we ended in the cockpit — examining the lens through which we perceive reality and doing the slow work of metacognition. That was Phase I of AQ. It is necessary, but it is not the whole work. A perfectly calibrated instrument is inert until you turn it on the world and ask: what terrain might I cross? What does this place actually look like? What could I be missing entirely?

That is the work of simulation — Phase II of AQ. And it is the work most people skip.

Simulation is not prediction, which claims that a specific outcome will occur. It is also not yet forecasting, which assigns probabilities across a known set of outcomes — that work comes later, once the option space has been built. Simulation is the earlier discipline: holding the option space open, including the options you have not yet thought to draw, before any probability can sensibly be assigned to any one route. Failing in your head, often and deliberately, so that when failure shows up in reality, you have already met it once before.

There is a maxim I have come to use: no wander, no wonder. The insights that change a decision arrive only after the wandering has happened. Skip the wandering, and the wonder does not show up.

The reason this is hard is that the brain hates it.

Your nervous system was tuned by a world where the rustle in the grass needed an answer in under a second. The decision in front of you needs an answer in days or weeks, and the same nervous system is screaming at you to close it now. The reflex that kept your ancestors alive is the reflex that wrecks your judgment in the world we now live in. The savannah rewarded fast closure. The modern environment rewards delayed closure. The two reflexes do not coexist easily, and the older one almost always wins by default.

Watch yourself the next time something unexpected lands on your desk — a piece of news, a critical email, an unfamiliar data point. The mind reaches almost instantly for an interpretation. It does not say, here is something new; let me hold five readings of it open and see which fits best. It says, this means X. The option space collapses before any examination of what the option space might be.

There is a new accelerant in this old problem.

Large language models — the technology most of us now use every day — are, by construction, convergence machines. They are trained to predict the most plausible continuation, the answer that best fits the patterns they have already seen. They are extraordinary at convergent operations, where the question is already well-defined. They cannot ask whether the question is the right one. They cannot hold the field open.

The more AI accelerates convergent thinking, the more valuable the work AI cannot do becomes: holding the field open, asking whether the question is even the right one, thinking in counterfactuals, refusing to converge too early. Habitually off-loading cognitive tasks to AI risks quietly eroding such capacities. Gartner recently forecast that critical-thinking capacity will atrophy enough under heavy AI use that half of all global organizations will require “AI-free” skills assessments by the end of 2026.

There is a much older name for the discipline that resists this drift. In Zen it is called shoshin — beginner’s mind. The expert’s cup is already full; the beginner has room for what the expert can no longer see.

The AQ framework calls the posture max entropy — deliberately increasing the uncertainty in your mind at the start of any consequential decision, treating every option as live until evidence makes it less so. No probability is set to zero. No probability is set to one hundred. Both are cognitive locks that close off learning before learning has begun.

This is uncomfortable. It looks like indecision from the outside and feels like self-doubt from the inside. Holding the field open requires fighting your own brain at the moment your brain most wants to close.

It also runs into the most respected currency in professional life: expertise. The more you know about your field, the more confident you can be wrong inside it. The more successful you have been, the more your past success calibrates you to a world that may no longer exist. The more your reputation rests on a frame, the more expensive it becomes to abandon the frame. This is the expertise paradox — the same training that saves you when you are refining a known system can kill you when you are constructing a new one. The point is not that expertise is obsolete. It is that timing matters more than possession. There are stages of a decision where deep training saves you, and stages where it kills you.

History gives the discipline its witnesses.

On January 6, 1912, a 31-year-old meteorologist named Alfred Wegener walked into a meeting of the Geological Association in Frankfurt and proposed that the world’s continents had once formed a single landmass and had since drifted apart. The geological establishment dismissed him on contact. He was not a geologist. He could not explain the mechanism — plate tectonics would not be confirmed for another half century. He died on a Greenland expedition in 1930, decades before he was vindicated. What he had that the experts in the room did not was a cup empty enough to entertain the question.

In 1989, a Hungarian biochemist named Katalin Karikó began working on messenger RNA at the University of Pennsylvania. The field had largely concluded that mRNA would never be viable as therapy. Her grants were rejected; her university demoted her four times. The seminal paper she co-authored in 2005 was turned down by Nature and Science before it eventually appeared in Immunity. Fifteen years later, the platform she had kept alive through three decades of professional pressure to abandon it became the basis of the first vaccines against COVID-19. In 2023 she shared the Nobel Prize for her work.

In November 2025, Yann LeCun walked away from Meta — twelve years there, an AI research operation he had built from scratch, a Turing Award along the way. He had concluded that the architectures the industry was racing to scale, including at the company he had helped build, were the wrong paradigm. In March 2026, his new lab raised $1.03 billion on the strength of an architectural thesis the consensus had rejected.

A meteorologist who was not a geologist. A biochemist who could not get a grant. A Turing laureate who departed from the well-trodden path. Different stakes, same discipline. Each held the field open against enormous pressure to close it, long enough to test a hypothesis the experts in the room could not afford to entertain.

I first had to do this kind of work in 1994, less than a year into my career in finance, when I joined Wellington Financial Group, which was to become Citadel, and was told to learn risk arbitrage. There was no textbook, no training program. I did not yet know which factors mattered. I did not have a frame.

So I started from something close to maximum openness. I read merger agreements cover to cover. I called consultants in narrow regulatory subdomains. I taught myself Delaware corporate law. I built spreadsheets that tracked dozens of variables with no way to know in advance which ones would matter. The work was slow, and from the outside it looked unproductive. The first 1,100 trades I eventually made, with only four losses, came after that period — not before it. I did not have a name for what I was doing. I had no alternative. I had no cup to fill.

The discipline is not natural, but it can be practiced. Five moves, in rough order of difficulty.

The first is the five-possibilities rule. Before any decision of consequence, write down five candidate readings of the situation — including ones that seem absurd. Treat each as live until evidence rules it out. The point is reps, not insight; the muscle of holding multiple readings open atrophies fast if it is not exercised.

The second is the no-0%, no-100% rule. Any time you find yourself certain that something is impossible or inevitable, treat that certainty as a flag. You have locked the field. Force yourself to articulate the conditions under which the certainty would dissolve.

The third is to read across domains. The most valuable signal usually arrives from a source you would not naturally consult. This is also the right way to use AI in Phase II — as a summarizer across fields you do not know, rather than as an answer machine in your own.

The fourth is to ask what would have to be true? When a result does not fit your model, do not dismiss it. Sketch the world in which it would fit. Sometimes the answer is an edge case I can ignore. Sometimes it is I am looking at the wrong map entirely.

The fifth is the most uncomfortable. Consult someone whose training is not yours about a decision in your domain. Their amateur question is often what an expert can no longer afford to ask.

None of this is the whole of Phase II. Holding the field open is the start of the work, not the end. At some point, usually because of a constraint, you have to extract a single hypothesis from all that openness and commit to it, while staying ready to abandon it the moment reality says you should.

That is the moment when explore turns into exploit. When wandering resolves into direction. When a strong opinion held weakly emerges from the field of possibilities.

Holding the field open is what makes the next move possible. Closing it too soon is what makes most decisions feel unrecoverable in hindsight. That is where we pick up.

Share

This piece is about what to do next.

Subscribe now

If the first move is recognizing that you are flying through turbulence with a fallible instrument, the second is learning how to calibrate it in real time. That is metacognition. Or, more precisely: adaptive optics for the mind.

There is one problem that even the most powerful ground-based telescopes cannot solve through better optics: the atmosphere. Air bends and scatters incoming light; the image of a distant star arrives at the mirror shimmering and unreliable. The distortion does not come from the instrument itself, but from what sits between the instrument and reality. The solution, called adaptive optics, works in real time. A wavefront sensor measures distortion thousands of times per second; a deformable mirror flexes to cancel it. The telescope sees clearly through air that would otherwise make its work impossible.

Metacognition works the same way. It does not remove distortion. It helps you detect it before you act on it.

None of us sees reality directly. We see through a lens — built from experience, education, pattern recognition, incentives, identity, prior success, emotional history, and inherited assumptions we did not choose. The lens may contain hard-earned wisdom. It is still a lens.

The most dangerous moment in any consequential decision is when you forget you are wearing one.

Metacognition sits upstream of everything that follows in AQ. A distorted observer does not just see less clearly. A distorted observer constructs flawed possibility spaces from the beginning — and every simulation, every experiment, every update is built on those foundations. If the instrument is off, the map will be wrong before you even start drawing it.

In cognitive science, metacognition is often described as the ability to monitor and regulate your own thinking. That is directionally right, but incomplete. Metacognition is not about thinking harder. It is about recognizing that what feels like reality is often interpretation — and creating enough separation to examine it before acting.

Most of what your mind produces does not arrive labeled as interpretation. It arrives feeling like fact. A founder reads criticism and feels betrayal. An investor sees a drawdown and feels reassurance — this is opportunity; it was last time. A leader hears dissent and feels disloyalty. In each case, the interpretation arrives wearing the costume of reality.

The work of metacognition is to find the seam between what actually happened and what you decided it meant.

One way to understand this is to think of the mind as operating across layers. At the surface is the conscious layer: deliberate reasoning, planning, explanation — the inner voice you experience directly. It is powerful but narrow, slow, and easily overwhelmed. Below that is the subconscious: pattern recognition, habits, intuitions assembled from prior experience. It works fast and largely outside awareness. It is what allows you to read a room, recognize a face, or sense that something is off without knowing why. It is also where many biases live. Deeper still is the unconscious — the firmware layer. Reflexes, threat responses, emotional reactions, the surge of stress or certainty before conscious reasoning has time to intervene.

By the time a thought reaches conscious awareness, it has already been shaped. The signal you need to catch often arrives from that deeper layer — a flash of certainty, threat, or relief — before the conscious story forms. Metacognition is the discipline of intercepting that signal before it hardens into conclusion.

William James observed that what we call “seeing” is often a form of remembering. Metacognition is the discipline of interrupting that process — long enough to examine it.

That discipline has a structure. Five steps, in order.

1. Notice the signal.

Metacognition often begins not with thought, but with emotion. Certainty. Defensiveness. Urgency. Contempt. Fear. Pride. Relief. Shame. Status threat. The signal arrives before the conclusion. Catch it, and you have a chance to understand what is about to follow.

2. Separate fact from story.

Ask: what actually happened, and what story did I add? Facts are observations. Meaning is constructed afterward, often instantly and unconsciously. Most distortions enter in the gap between the two.

3. Identify the lens.

What might be shaping your perception right now? Fear? Ego? Incentives? Identity? Prior success? Habit? Tribal loyalty? Old wounds? The question is not whether you are biased. You are. The question is which bias is active in this moment.

4. Hold the story as a hypothesis.

Do not merge with your first interpretation. Treat it as one possible explanation among several. The goal is not to eliminate intuition or emotion. The goal is to stop experiencing your first interpretation as unquestionable truth.

5. Keep the map open.

Do not rush to resolve the interpretation. Keep multiple explanations alive. Let new information, reflection, or reality itself update your view. Your mind wants closure. Your ego wants relief. Your identity wants protection. Your prior success wants confirmation.

Metacognition says: not yet.

These steps can be brief or extended depending on the situation. But they produce something essential: a more accurate reading of reality — and a clearer view of yourself as the observer.

Consider a familiar situation. A CEO learns that a younger employee used AI to complete in hours what used to take a team days. The first reaction arrives quickly. It may be dismissal — this is not real judgment. It may be threat — this undermines our model. It may be excitement — we can cut costs immediately. Each reaction may contain truth. But each is also an interpretation.

Run the five steps.

Notice the signal: threat, excitement, defensiveness, urgency.

Separate fact from story: the fact is faster output; the story is that junior roles are obsolete, or that costs can be cut without consequence, or that judgment has been replicated.

Identify the lens: prior success, margin pressure, fear of disruption, status threat, or overconfidence in efficiency gains.

Hold the story as a hypothesis: perhaps AI replaces certain tasks; perhaps it changes apprenticeship; perhaps it increases output but weakens capability formation.

Keep the map open: test quality, error rates, client outcomes, learning effects, and long-term talent development.

The point is not to slow the decision. It is to prevent the first story from becoming the strategy.

This dynamic scales. The institutions that failed in 2008 did not lack intelligence. They had extraordinarily sophisticated models built by very capable people. The problem was not the math. The problem was that the models encoded assumptions that were never interrogated. The map was mistaken for the territory. When reality diverged, the models failed.

The same pattern appears in smaller decisions every day. A founder interprets criticism through past betrayal. An investor interprets volatility through prior success. A leader interprets dissent through a lens of control. The issue is not a lack of thinking. It is thinking trapped inside an unexamined lens.

This is where grit and curiosity enter the work. Grit is often described defensively: stay the course, do not quit. There are moments when that is exactly right. But in a changing environment, that instinct can become loyalty to a stale map. The higher form of grit is offensive: the ability to hold your first interpretation at bay long enough for curiosity to do its work. Offensive grit says: not yet. Curiosity keeps grit from hardening into rigidity.

Imagine an Ironman swim. Grit keeps your head down, stroke after stroke. Curiosity makes you lift your head to check whether the current has pulled you off course. Too much grit without curiosity, and you swim hard in the wrong direction. Too much curiosity without grit, and you never move. AQ requires both.

A process that only protects your current belief is not metacognition. The five steps only work if you are willing to be corrected by what they reveal. Otherwise, they become decoration on top of the same lens you started with.

Uncertainty does not only come from the world. Some of it is introduced by the observer. Fear narrows. Ego protects. Prior success overweights the familiar. Incentives filter. Identity hardens. If you are navigating a regime change, you cannot afford to confuse distortion in the instrument with properties of reality itself.

Metacognition is not introspection for its own sake. It is calibration. The goal is not to eliminate uncertainty. It is to stop adding avoidable error before the real work begins.

A distorted observer does not just misinterpret reality. They construct flawed possibility spaces from the beginning. Before you can map the world, you must calibrate the instrument doing the mapping. Once the instrument is checked, you can begin building better maps — generating hypotheses, testing them against reality, and updating.

From metacognition to simulation — first the field, then the path.

From seeing the lens to constructing paths that can survive.

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

Subscribe now

That is not a new problem. Carved above the entrance to the Temple of Apollo at Delphi was the ancient command gnōthi seauton — know thyself. Before visitors could ask the oracle what the future held, they were first confronted with a harder question: do you understand the instrument doing the asking?

That was the first and most difficult task then. It remains so now. The gap between who we assume we are and how our minds actually work matters enormously, because it shapes everything about how we think, decide, and adapt.

Which brings us to something that may sound obvious once you hear it, but that most of us spend our lives never quite confronting: your mind is not naturally suited to the moment we are living through.

Not because you are not smart enough, or because you have not read the right books, taken the right courses, surrounded yourself with the right people, or worked hard enough. The problem is deeper than that. The “mindware” you are running — the same basic cognitive architecture human beings have carried for hundreds of thousands of years — was built for a world that no longer exists.

The brain that helped our ancestors survive was tuned for an environment defined by immediate threats, physical dangers, local information, and relatively stable patterns. In that world, heuristics were survival tools, not flaws. A farmer who watches the same land for decades gains wisdom no textbook can convey. A trader who has seen the same pattern repeat across hundreds of similar situations develops real judgment. A parent who knows a child’s rhythms can sense when something is wrong before there is evidence. In stable environments, the grooves cut by experience become useful channels.

But that is not the world we live in now.

Today, the forces shaping our lives are often invisible, abstract, nonlinear, and fast-moving. Markets shift because of variables no one can see directly. Social media rewires group behavior at machine speed. AI changes the cost and availability of cognition itself. Geopolitical events ripple through supply chains, institutions, capital markets, and personal decisions in ways that are nearly impossible to trace in full. The gap between what the brain evolved to handle and what it is now being asked to navigate has never been wider.

That gap explains much of the confusion, frustration, polarization, and paralysis we see around us. When uncertainty spikes, the brain does not naturally widen its aperture. It narrows it. The mind simplifies, pattern-matches to the nearest familiar analog, reaches for the most recent vivid example, and often locks into action before the full complexity of the situation has registered.

This is not a character flaw. It is physiology. The stress response evolved for immediate physical danger, but in modern life it is routinely recruited by abstract threats: a market crash, a public humiliation, a business failure, a geopolitical crisis, or a technology that threatens your identity. The lion has changed form; the body still prepares to run.

Then the conscious mind does what it often does best: it explains. It creates a story. It gives reasons. It turns a reaction into an argument and an argument into a conviction.

This is where the danger begins. The greatest risk is not that we lack intelligence. The greater risk is that intelligence, applied through a distorted lens, becomes a force multiplier for error.

The smarter you are, the more elegantly you can defend the wrong map.

The more experienced you are, the more confidently you can misread a new environment through an old pattern. The more successful you have been, the more tempting it becomes to assume that the lens that worked before is still the lens that works now. This is why AQ begins before strategy, forecasting, or action. It begins with the instrument doing the perceiving.

Most of us experience our thoughts as reality. We do not say, “I am interpreting this person as hostile.” We say, “That person is hostile.” We do not say, “I am applying a prior market regime to this data.” We say, “This is obviously a buying opportunity.” We do not say, “This technology threatens my status, so I am discounting it.” We say, “This is hype.”

We mistake interpretation for observation. Once that happens, we begin optimizing inside a reality that may not exist.

That is false-reality optimization.

Bad decisions are not always made because people are lazy, stupid, or uninformed. Often, they are made because people are solving the wrong problem beautifully, applying exquisite logic to a distorted premise, and steering hard by instruments they have not checked.

This is why metacognition matters.

Metacognition is often defined simply as “thinking about thinking.” That definition is accurate, but too soft. It can make metacognition sound like reflection, mindfulness, or abstract self-awareness. For AQ, metacognition is more practical and more demanding. It is decision hygiene: the discipline of inspecting the lens through which you are interpreting reality before you act on that interpretation.

A pilot flying through heavy weather does not simply trust what he feels. He checks the instruments, not because he believes the instruments are perfect, but because he knows his senses are fallible under pressure. The same is true of the mind. Your perceptions may be useful. Your intuitions may contain real information. Your emotions may be signaling something important. But none of them should be treated as reality itself until they have been examined.

Metacognition does not mean distrusting yourself. It means knowing that the self doing the seeing is part of the system being analyzed.

That is what makes it hard. It is one thing to analyze the outside world. It is much harder to analyze the machinery through which the outside world is being filtered. Your education, incentives, prior success, failures, trauma, identity, profession, social group, emotional state, and era all shape what you notice, what you ignore, what you defend, and what you refuse to imagine. Under conditions of rapid change, those filters can become dangerous precisely because they once worked.

The good news is that human beings have one extraordinary advantage: we can turn a thought into an object. We can look at a belief and ask, “Why do I believe this?” We can look at a reaction and ask, “What prior experience is this activating?” We can look at a certainty and ask, “What would make this wrong?”

That ability is not automatic. It has to be practiced. Most education systems train us for convergent problems: find the answer, execute the method, perform within known parameters. Those skills matter. But the real world increasingly runs on divergent problems, where there is no single answer key and where the quality of the outcome depends less on what you already know than on how well you can update.

That is the world AQ is designed for. And in that world, the first move is not to think harder. The first move is to check the instrument.

What lens am I using? What assumption am I treating as fact? What emotion is shaping my interpretation? What prior success am I overgeneralizing? What would I believe if I had not already taken a position? What evidence am I explaining away too quickly? What would make me wrong?

These are not soft questions. They are performance questions. When the environment changes, the person who can inspect his own lens has an enormous advantage over the person who merely defends it.

The turbulence is real. But metacognition asks you to operate on two levels at once: understanding the architecture of the instrument you are flying with, and noticing how that instrument is performing in this weather, on this flight, with these stakes.

Inside the cockpit, doing both at once, is where we’ll pick up next time.

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

For more than a century, we have largely treated intelligence as a single-dimensional problem.

IQ became the dominant measure of cognitive horsepower: how much information you can hold in working memory, how quickly you can process it, and how well you can detect patterns and relationships—including distant or non-obvious ones—within a given frame. It was measurable, comparable, and, critically, it felt objective. The score was the score.

Then came EQ—emotional intelligence—which broadened the picture in important ways, revealing that self-awareness, empathy, and the ability to read social environments were not soft add-ons to IQ’s raw cognitive power, but structural components of high performance.

But something is still missing.

IQ helps you process information within a given frame. EQ helps you read yourself and the people around you. But neither tells you what to do when the frame itself stops working—when the map no longer matches the territory, or the rules of the game begin shifting faster than your models can update.

That matters far more now than it once did. We are living through a period in which change is accelerating across too many domains at once—technology, markets, institutions, politics, identity, even the systems by which people decide what is true. In that world, the advantage shifts. It is no longer enough to optimize within an existing map. Increasingly, the real edge belongs to those who can recognize when the map is stale and redraw it while everyone else is still solving the old problem.

More than that, the challenge is not simply to think better, but to retain agency.

This is the domain of the Adaptability Quotient. AQ is a learnable operating system for decision-making in a world where the map keeps changing. It is not a fixed trait or innate capacity, but a repeatable decision architecture that can be cultivated and improved over time. In a world of accelerating volatility and increasingly commoditized cognition, AQ helps you stay in the pilot’s seat: recalibrating your instruments, restoring a feedback loop with reality, and updating your map before stale assumptions—or someone else’s model—begin making your decisions for you.

One way to see the distinction is this: IQ is especially powerful in convergent domains—settings where the rules are relatively stable, the problem is well-defined, and there is a right answer, or at least a narrow range of better answers. School rewards this. Standardized tests reward this. So do parts of math, coding, and other structured forms of problem-solving. But much of adult life—leadership, entrepreneurship, investing, relationships, institution-building—unfolds in divergent environments, where the variables are incomplete, the rules may be changing, and the odds are often not merely unknown but unknowable in advance. In those settings, raw processing power still matters, but it is no longer decisive. The critical advantage becomes the ability to update, reframe, and act under live uncertainty. That is the domain of AQ.

This is also why AQ becomes more important, not less, in the age of AI. As AI systems grow better at processing information, retrieving patterns, and generating answers, raw cognitive horsepower becomes more abundant. But many of the most important real-world problems still cannot be solved by computation alone. They require judgment about what questions to ask, which assumptions to challenge, which variables actually matter, and what experiments to run in the real world in order to generate feedback. AI can increasingly help think inside the map. But when the map itself is incomplete, unstable, or wrong, value shifts toward the ability to test reality, learn from it, and update accordingly. As processing becomes cheaper, reframing becomes more valuable.

The people who consistently thrive in volatile environments—serial entrepreneurs, adaptive investors, leaders who succeed across cycles—share something deeper than personality or background. What they have in common is an underlying process—one that has often remained implicit, scattered across disciplines, and easier to recognize in practice than to describe clearly. What I am trying to do here is make it explicit: to bring those pieces together into a framework that can be understood, practiced, and shared.

That process unfolds across three interlocking phases, not as a linear checklist but as a continuous, self-correcting cycle—a feedback loop with reality.

Metacognition comes first. Before you can navigate the world well, you have to understand the mind you are using to navigate it. That means examining your own assumptions, biases, instincts, and overlearned habits of interpretation. Where does your confidence come from—and is it earned? Which intuitions are genuine signal, refined by experience, and which are reflexes dressed up as wisdom? When should you trust your gut, and when is your gut steering you off a cliff? Metacognition is not self-awareness as performance. It is the disciplined practice of checking your instruments before you trust the map they are helping you draw.

Simulation comes next. Once you have a clearer view of how your own mind works—and where it may be misleading you—you can begin modeling the external world more effectively. But the goal is not to generate a single prediction and call it foresight. It is to build a decision tree: a structured map of branching possibilities, competing hypotheses, and potential futures. Most people narrow their option set too early, treating uncertainty as if it were just a harder version of risk. Simulation resists that temptation. It holds the field open longer, broadens the possibility space, and helps you see where a small decision may send you down a fundamentally different path.

Subscribe now

Then comes Experimentation. At some point, the model has to meet reality. Experimentation is where you take your best provisional hypothesis into the world—not to confirm your brilliance, but to test it. That does not mean betting the farm. It means running small, intelligent, reality-based probes that generate feedback, expose fatal flaws early, and help you buy information before making irreversible commitments. The point is not to be right on the first pass. It is to shorten the distance between model and terrain by learning faster than the environment is changing.

Run this loop consistently and you do not just make better isolated decisions. You begin building a compounding advantage—a way of staying adaptive, coherent, and agentic even as conditions shift around you. From the outside, that can look like intuition, resilience, or even luck. But it is not luck. It is practice.

And like any practice, it begins with understanding that AQ is not something you either have or do not have. It is a process you can cultivate.

In the next installment, I’ll go deeper into Phase I and why metacognition turns out to be the lever everything else depends on.

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

More specifically, this apparent tension is derived from seemingly different utility curves.

The venture capital model is built on diversification. A typical fund may invest in twenty to thirty companies or more, expecting that many will fail. VCs tolerate this failure rate because early-stage investing follows a power-law distribution: most companies fail or produce modest returns, while a small number generate extraordinary value that more than compensates for the losses. For that reason, VCs often seek to back “moonshots”—companies with massive enterprise value potential, even if the odds of success are long. For the venture capitalist, this structure makes hunting for extreme right-tail outcomes rational.

The founder, by contrast, lives inside a very different reality. A founder does not have twenty parallel companies. Even the most productive founders may have only three or four true attempts at building a generational company, and each attempt can take more than a decade to fully play out. Their reputation, career, team, and financial future are often tied to a single enterprise. Failure is not a statistical outcome spread across a portfolio; it is a concentrated knockout event. For that reason, founders are understandably unwilling to view themselves as merely one data point in a VC’s portfolio. As a result, they prize survival because survival preserves the opportunity to find product-market fit and achieve success. For the founder, extreme risk-seeking is often irrational.

This difference creates what appears to be a conflict between opposing viewpoints. The venture capitalist carries diversified exposure across a portfolio, while the founder carries concentrated uncertainty within a single company. The incentives that emerge from these positions inevitably differ. That much is unavoidable.

But this tension is often misunderstood. At its core, it is not really about impersonal risk-seeking ambition versus individualized self-preservation. Both founders and venture capitalists are drawn to outsized outcomes and deep success. The difference lies in how risk is experienced, how uncertainty is resolved, and how time horizons shape decision-making. Understanding this distinction clarifies why the founder–venture capitalist relationship can feel uneasy—and also why, in the best cases, it becomes extraordinarily productive. To see why, we first need to distinguish between risk and uncertainty.

In traditional finance, risk refers to situations where probabilities are known or can be estimated, allowing expected returns to be calculated and optimized against. Startups rarely operate in this environment. Most exist under Knightian uncertainty, where the probabilities themselves are unknown. The market may not yet exist, the category may not yet be defined, and the customer’s needs may not yet be fully understood.

Under these conditions, forecasting becomes unreliable and decisions cannot simply be optimized against a probability distribution because the distribution itself is unclear. The founder’s task, therefore, is not merely to take risk. It is to resolve uncertainty.

In fact, classical economic theory assigns this exact function to entrepreneurs. Frank Knight argued that profits arise not from bearing calculable risk, but from navigating situations where the future cannot be predicted. The entrepreneur’s role is to convert uncertainty about markets, technologies, and customers into functioning businesses.

Startups are perhaps the purest expression of this idea. A company must discover whether a real problem exists, whether it can design a product that solves it, and whether customers will pay for that solution. Such discovery requires experimentation, feedback loops, and adaptation. The founder’s job is not to predict the future accurately. It is to learn faster than the company runs out of resources.

Many VCs miss this nuance. That is why, when venture capitalists say they want “moonshots,” the idea is often caricatured as tone-deaf advice that founders should simply pursue the boldest visions with unwavering conviction. In other words, the caricature arises because many investors equate risk-taking with uncertainty resolution, and the resultant advice they give based on that mis-categorization often leads to illogical or ill-advised actions being forced upon the founder. But if one looks more closely at the history of the best founders—especially those with repeated success—the evidence tells a more nuanced story.

Many iconic companies reached their eventual markets through substantial pivots. Instagram began as Burbn, a check-in app cluttered with features users largely ignored. Slack emerged from the internal communications tool of a gaming company whose original product failed. CoreWeave began as an Ethereum miner before pivoting into AI infrastructure. In each case, the founders were not stubbornly wedded to the original concept. They were probing reality until the real opportunity revealed itself.

This is significant because it reframes what venture investors are actually backing: not simply an idea, but a founder’s ability to navigate uncertainty. The best founders are not those who rigidly cling to an initial vision. They are the ones who discover the right path through iterative learning rather than by trying to predict it in advance. They experiment, adapt, and refine in real time until a market opportunity emerges.

From the founder’s perspective, then, survival is not the opposite of ambition. It is the prerequisite for it. If uncertainty can only be resolved through experimentation, then runway equals time, time equals learning cycles, and learning cycles reveal whether the company has discovered product-market fit. What may look like caution from the outside—careful spending, iterative testing, pivots—can actually be the most rational strategy under uncertainty. A founder who preserves resources and runs frequent, disciplined experiments is not avoiding risk. They are maximizing the number of opportunities the company has to learn before capital runs out.

This insight points to a subtle but important conclusion. Many startups fail not because founders take too much risk, but because they treat uncertainty as though it were risk—building complete products before testing demand, investing heavily in infrastructure before validating markets, or assuming probabilities they cannot actually know. In other words, they optimize too early. Under uncertainty, the rational strategy is not optimization but exploration: running small experiments that generate information and compound over time. Founders who succeed are those who structure their companies as learning systems rather than prediction machines.

Once this distinction becomes clear, the apparent conflict between founders and venture capitalists begins to dissolve. The venture capitalist wants exposure to massive upside—but that upside can only emerge if uncertainty is eventually resolved into a scalable business. The founder, meanwhile, must survive long enough to perform that resolution. Which means the investor is implicitly underwriting two things at once: that the opportunity space is large enough to support a transformative company, and that this founder is adaptable enough to discover where that opportunity actually lies. The founder’s task is complementary: pursue a problem large enough to justify venture capital while continuously running disciplined experiments to determine whether the assumptions are correct and whether product-market fit truly exists.

Seen this way, the founder–venture capital relationship is not a clash between caution and ambition. It is a partnership between portfolio construction and uncertainty resolution.

Of course, venture capitalists themselves are not immune to the same mistake founders sometimes make: treating uncertainty as if it were risk. Despite operating in an industry built on power-law outcomes, many investors fall back on traditional instincts—overdrawing conclusions from past successes, constructing overly detailed forecasts, or favoring ideas that fit familiar narratives. These habits can filter out precisely the anomalous opportunities that produce outlier returns. In this sense, venture investing requires its own form of discipline. Just as founders must resist the urge to optimize before learning, investors must resist the urge to impose premature certainty on emerging companies.

Which brings us back to the original tension: the venture capitalist seeking moonshots and the founder seeking survival. We can now see that the two goals are not actually opposed. The venture capitalist needs exposure to extraordinary outcomes, while the founder needs to manage uncertainty in a way that allows those outcomes to emerge. The unifying framework between these objectives is adaptive learning.

A founder who rigidly pursues an initial idea without testing assumptions may appear bold, but that approach often destroys the learning cycles necessary to discover real demand. Conversely, a founder who experiments endlessly but never takes big swings may build a stable company that fails to justify venture capital. The best founders do both: they pursue large problems while remaining flexible enough to change course as reality reveals itself.

In the end, the relationship between venture capitalists and founders is not defined by conflicting incentives but by complementary roles. The venture capitalist constructs a portfolio designed to capture asymmetric outcomes, while the founder resolves uncertainty inside a single company, transforming unknowns into validated markets.

This is why, at their best, venture investors also contribute more than capital. Capital matters not just because it funds growth, but because it extends runway—buying the founder more time and more learning cycles to resolve uncertainty before the company runs out of resources. But beyond that, experienced investors can often help resolve uncertainties that sit adjacent to the core product but are still crucial to survival and scale: governance, financing strategy, hiring, customer introductions, partnerships, and market positioning. In other words, the best investors do not simply fund the experiment: They help the founder survive long enough—and learn fast enough—to discover what the business can become.

When both sides understand these roles, the partnership becomes far more powerful. The investor provides capital, experience, networks, and patience. The founder provides the discipline and adaptability necessary to discover where the real opportunity lies. Venture capital does not reward founders who chase moonshots with blind conviction. It rewards those who can survive long enough—and learn fast enough—to discover where the real moonshot actually lies.

It already happened, so we assume we understand it.

But the past is often far less settled than it appears. Two witnesses can leave the same accident with entirely different accounts. Memory is reconstructive, not reproductive. History itself is full of confident explanations later overturned by better models. The past feels stable not because it was ever perfectly known, but because we already survived it.

So if the past is murkier than it looks, what does that imply about the future?

In 1902, H. G. Wells delivered a lecture at the Royal Institution in London titled The Discovery of the Future. His argument was radical for its time. Humanity, Wells said, had spent centuries looking backward. But we were only just beginning to recognize that the future, too, could be studied—not through prophecy or mystical certainty, but through evidence, pattern recognition, and reason. By observing forces already in motion, we could orient ourselves toward what was coming before it fully arrived.

That distinction matters even more now than it did then. Because there are several very different activities that people still collapse into one.

Prediction tries to tell you what specific outcome will occur. Forecasting tries to estimate the relative likelihoods across a range of possible outcomes. But under genuine uncertainty, the deeper challenge is orientation: understanding the forces shaping the terrain, forming provisional forecasts, and revising both your map and your expectations as reality reveals more information.

In other words, prediction seeks a specific answer. Forecasting estimates probabilities. Orientation is the broader discipline of deciding how to proceed when the map is incomplete and the probabilities themselves are still being learned.

This is not a trivial distinction. It is the difference between trying to guess the next move on the board and recognizing that the board itself is changing.

In the last installment, I explored Knightian uncertainty—the difference between risk, where probabilities can be assigned, and genuine uncertainty, where they cannot. Much of the world we now inhabit lives squarely in that second category. Artificial intelligence, robotics, biotech, genomic engineering, human-machine integration, new energy systems, space infrastructure, and geopolitical realignment are not domains in which precise timelines or reliable probabilities can be assigned with confidence. They are domains of directional force, nonlinearity, and incomplete information.

That makes prediction weaker than most people want it to be.

But it does not mean we are helpless.

Even when precise prediction fails, orientation remains possible.

We can observe acceleration. We can trace incentives. We can identify bottlenecks, asymmetries, dependencies, and feedback loops. We can see that some capabilities are compounding faster than institutions can absorb them. We can detect where old assumptions are breaking. We can notice that the environment is changing in patterned ways, even when the endpoint remains unclear.

And in such environments, forecasting does not disappear. It becomes more provisional—something to be updated through repeated learning rather than assumed with false precision at the outset.

That may be the most important cognitive distinction of our time.

For most of history, the old model of orientation worked reasonably well. Wells called this the legal view of the world: looking backward for precedent, trusting continuity, assuming the future will resemble the past closely enough that mastery of the existing map will suffice. In slower-moving environments, this often worked. Change happened, but usually at a pace that allowed mental models, institutions, and norms to adapt gradually.

That is no longer reliably true.

We are not simply living through rapid change. We are living through a change in the nature of change itself.

Previous technological revolutions altered the world dramatically, but often at a pace that still allowed lagging adjustment. Institutions bent slowly. Mental models updated eventually. The gap between environment and understanding remained painful, but manageable.

Today, that gap widens faster than we can close it.

Tools compound exponentially while our cognitive, cultural, and institutional infrastructure often remains linear. We continue to rely on maps built for a world in which the terrain moved slowly, even as the terrain itself becomes increasingly dynamic, recursive, and designed. And it is in that widening gap—between what is happening and what we understand—that agency begins to erode.

Wells believed humanity was learning to detect the forces shaping the future before they fully arrived. He was right. Over the last century, we learned to model climate, track technological adoption, study demographic transitions, and observe structural change before it became obvious in everyday life.

But now something more profound is happening.

We are no longer merely discovering the future. We are increasingly designing it.

The old world was largely about reading reality better: observing, measuring, predicting, and responding. The emerging world is increasingly about writing reality itself—editing code, cells, atoms, minds, and systems rather than merely adapting to them.

We are moving from a read-only world to a read-write one.

We no longer just harness the sun’s energy to generate power; we aim to recreate stellar processes via fusion. We no longer simply diagnose disease; we increasingly edit genomes. We no longer only use tools to supplement thought; we are building systems that reshape cognition, transform decision-making, and blur the boundary between human and machine agency.

The consequences are not merely technical.

Previous waves of technology largely reduced uncertainty around material survival—food, shelter, safety, information, production. But this wave pushes uncertainty upward. As systems take on more of what humans once did for work, judgment, and coordination, the central question shifts from survival to significance: meaning, purpose, identity, and agency. In that sense, the map of a good life is changing along with the world itself.

Subscribe now

As the world becomes more read-write, the challenge is not simply knowing more. It is learning how to orient when the terrain itself is being actively rewritten.

That is why this moment demands something more than intelligence, expertise, or prediction. It demands a disciplined way to build and rebuild our orientation when old maps stop resembling reality. We are learning to program machines, bodies, and biology. But the real bottleneck may be that most people are still running ancient mental software in a world that no longer matches it.

AQ is the upgrade.

AQ is not about predicting perfectly. It is about orienting under genuine uncertainty—recognizing when precedent fails, identifying the forces that matter, and updating your view of the world before drift, dogma, or delay make the choice for you.

In a Knightian world, you do not get the luxury of complete information before action. But neither are you condemned to drift. The task is to build a better map than the one you inherited—and then keep redrawing it as reality changes.

Because the map never stops changing. That much is true.

But not everyone is equally lost.

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

You make decisions every day. We all do—life is, in a sense, a series of decisions, major and minor. Yet the overwhelming majority of us are missing something fundamental about the world we’re operating in—making a mistake so common, and so deeply embedded in how we think and talk about the future, that most people never even realize they’re making it.

They use the word uncertainty when what they actually mean is risk.

At first glance, this confusion might appear semantic, even trivial. But the reality is that it shapes nearly everything—how companies plan, how investors allocate, how leaders make decisions, how each of us moves through the world.

Risk and uncertainty are not the same thing. They do not even belong in the same category.

The economist Frank Knight drew this distinction more than a century ago in his 1921 book Risk, Uncertainty, and Profit. His insight was deceptively simple: not all uncertainty is created equal. Sometimes we face known possibilities with known probabilities—which is called risk. This is the domain where formal models, expected value calculations, and optimization genuinely work. Casinos operate in this realm, and so do car insurance companies. When the odds are stable and enumerable, you can do the math and play accordingly.

A roll of a die is the simplest illustration. We cannot predict the next roll, but we know the possibilities—1, 2, 3, 4, 5, or 6—and the probabilities: each outcome has a one-in-six chance. The possibilities and probabilities are known, even if the specific outcome is not.

But (as we all know) most of life just plain doesn’t work that way.

At the other extreme are what Nassim Taleb calls black swans—events where neither the possibilities nor the probabilities can be known in advance. A financial crisis that propagates through the global system in ways no model anticipated. A geopolitical shock that rewrites the rules overnight. A new technology that wasn’t even on the map.

In environments like these, prediction fails almost by definition. The sensible response is not to optimize harder but to design systems that are robust to surprises—to avoid catastrophic downside and remain capable of absorbing shocks that no model could foresee.

Between those two poles, however, lies the territory where most real decisions occur.

Knight called it simply uncertainty, though it is now often referred to as Knightian uncertainty in his honor. It describes a world where possible outcomes are visible, but probabilities are not.

Consider hiring someone—or even dating someone. You can imagine how it might go well or how it might go poorly, but you cannot calculate the odds. Or launching a new product—you can sketch several possible outcomes, yet the distribution across them remains opaque. Or deciding whether to move, pivot, or commit—in business or in life. The futures you can imagine are clear enough. Their likelihoods remain stubbornly unquantifiable.

Knight argued that entrepreneurs succeed not by bearing risk—but by navigating this middle territory of uncertainty. Risk can be priced; uncertainty must be explored. Exploration, in practice, means taking action: intervening through experiments that generate feedback loops and gradually resolve what was previously unknown.

This distinction may sound trivial, but it changes everything about how decisions should be made.

When probabilities are known, you can optimize—build the model, run the numbers, and choose the strategy with the best risk-adjusted outcome. But when probabilities are unknown, optimization becomes fragile. You are plugging essentially made-up inputs into a very precise machine and treating the output as if it were trustworthy.

The right move instead is to probe, experiment, and gather information before scaling commitment. You are not trying to win the game on the opening move. The first challenge is learning what game you’re actually playing.

Once you see this pattern, you begin to recognize it everywhere.

Startups launch minimum viable products (MVPs) not because they lack ambition, but because an experimental, iterative approach is the smartest way to operate in conditions of genuine uncertainty. The goal is not immediate success—it is information. This is why companies run A/B tests before rolling out major changes, why investors sometimes take exploratory positions before allocating serious capital, and why a skilled poker player facing an unfamiliar opponent does not start optimizing immediately. Instead they watch, gather signal, and calibrate based on their opponent’s unique tendencies.

When the odds are unclear, the objective shifts. You are no longer maximizing expected value—you are resolving uncertainty so that optimization eventually becomes possible.

History offers a recurring lesson: leaders fail not because they lack intelligence, but because they apply the wrong model to the environment they are actually in.

After World War I, France built the Maginot Line—one of the most sophisticated defensive systems ever constructed. It was designed to stop the kind of war they had just fought: slow infantry advances, artillery barrages, and trench warfare.

But the next war was different. German commanders realized that tanks, radios, and air support had fundamentally changed the speed of warfare. Instead of attacking France’s strongest defenses, they drove armored divisions through the Ardennes forest—terrain French planners believed was impassable—and collapsed the entire defensive system from the rear.

The Maginot Line was not a bad solution. It was simply a solution to the previous war.

Some officers, however, recognized that the map of warfare had changed. Among the earliest in the United States was George S. Patton. After World War I he became deeply interested in mechanized warfare, studying tanks and advocating for armored divisions at a time when much of the U.S. Army still thought in terms of infantry and horse cavalry. Patton helped design early American armored doctrine and pushed relentlessly for the creation of fast, mechanized units capable of exploiting breakthroughs rather than fighting static battles.

When the United States entered World War II, that updated mental model mattered. Patton’s command of the U.S. Third Army became famous for its speed and maneuverability. His operations across France in 1944 moved with a tempo that often rivaled the German blitzkrieg itself.

Patton’s philosophy was simple: movement creates opportunity. If you move faster than your opponent can update their map of the battlefield, their defenses become irrelevant.

The contrast illustrates the deeper lesson. France optimized for risk—defending against the most likely version of the last war. Germany and a few adaptive commanders optimized for uncertainty—recognizing that technology had changed the distribution of how wars could be fought.

The failure was not intelligence or effort. It was a failure to update the map when the environment changed.

This matters now more than perhaps ever before.

Many of the forces reshaping the world around us—artificial intelligence, automation, geopolitical realignment, energy transition, and the fragmentation of shared information environments—exist squarely within Knightian uncertainty.

The possible outcomes are not hidden. We can imagine industries being transformed or disrupted, labor markets shifting, geopolitical power balances changing, and new technological dependencies emerging. What remains opaque is everything else: the timing, the sequencing, and the probabilities.

How quickly will these changes unfold? Which technologies will dominate? Which will stall? Which will produce consequences their creators never anticipated? We cannot yet assign reliable probabilities to any of it. Yet the dominant response in boardrooms, among policymakers, and within investment committees is to behave as if we can.

There is a more useful way to think about uncertainty.

In a world where probabilities are unknown, every experiment is an act of discovery. Each probe of the system reveals a little more information. These fragments accumulate, gradually resolving some of the opacity—not as precise prediction, but as understanding of the structure of the environment itself.

Even if individual events are unpredictable, the forces shaping the landscape they emerge from are often visible—if you know where to look.

When probabilities are unknown, uncertainty is not an obstacle. It is information waiting to be discovered.

The goal is not to eliminate uncertainty. The goal is to become comfortable exploring it—to treat it not as a threat to be modeled away but as a signal waiting to be read.

Because in environments like the one we are entering, advantage does not go to whomever builds the most sophisticated model. It goes to the decision-maker who updates their map the fastest.

In an upcoming newsletter installment, we will explore what “probing the system” actually looks like in practice—and why the leaders who navigate uncertainty well tend to follow a surprisingly consistent process.

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.

Subscribe now

The conversations we’re having are brittle. The institutions we rely upon are either faltering or failing. Kids today seem both hyperconnected and yet, at the same time, strangely alone. Public debate yields only escalating conflict, not resolution and understanding. Organizations pivot, pivot again, then freeze and (all too often) disappear.

And all of this is happening against the backdrop of artificial intelligence—its rise, its acceleration, its steady permeation of the world, the very substrate, around us: writing, coding, summarizing, predicting, analyzing, all at speeds that would’ve sounded like science fiction just five years ago.

Most of the commentary out there treats these as separate stories.

They aren’t.

We are entering what I call a second cognitive revolution at precisely the moment when our stabilizers are weakest. Anthropologists sometimes distinguish between “cold” and “hot” societies. Cold systems resist rapid change through stabilizing myths, rituals, and durable narratives, while hot systems reinterpret themselves constantly through disruption and progress.

Modern society is hot. Digital society is hyper-hot. And many of our cold stabilizers—shared stories, widely trusted institutions, durable authority structures—have weakened or collapsed just as the environment began accelerating.

At its center, the crux of this moment, is something more fundamental than the rise of social media or the state of our politics: the erosion of human agency—the sense that events are increasingly beyond our control, delegated to algorithms or happening faster than we can shape them.

To me, navigating this moment is the defining challenge of our time.

For most of my career, I’ve worked far from the spotlight—in finance, in investing, in building businesses and institutions. My professional life, spanning some 30-plus years, has been largely defined by making decisions under uncertainty—allocating capital, evaluating people and organizations, assessing changing environments.

But I’m speaking up for the first time, right now, because I believe it’s time to sound the alarm.

When we as a species are reduced to reacting rather than choosing, responding rather than directing, consuming rather than constructing—that’s more than an evolution in the technologies we wield. It signals a fundamental shift in who (or what) is exercising agency in the first place.

To understand what’s at stake amid this second cognitive revolution, it helps to understand the first. When humankind developed language, it set our species on a radically different path. Today, our closest evolutionary relatives (primates like chimpanzees and bonobos) are capable of communication, but mostly about what’s happening right now—via gesture, posture, and sound (i.e., “predator over there,” “good food that way”).

With language, humans became capable of communicating about things that weren’t there. We could describe what we thought and experienced yesterday, articulate our hopes for tomorrow, invent gods, laws, customs, nations, markets. We could coordinate around shared fictions or abstract ideas, translating them into reality. More importantly, we could ask a simple but revolutionary question: what if?

What if we try something new? What if this story isn’t true—or there’s more to this than meets the eye? What if we organize ourselves differently or approach from a different angle?

That was the first cognitive revolution. It expanded human agency dramatically, making coordinated, intentional change possible at scale.

Eons later came mass media. For the first time, entire nations (and eventually the world) could watch the same politician’s speech, or the same conflict or natural disaster unfolding, in real time. Yet over the last 15 years, and with increasing speed over the past 5 or 6, our entire shared frame of reality has badly fractured. And with it, many of the cold stabilizers that once anchored public meaning began to weaken.

With the rise of social media came personalized information streams that reward speed and reaction. A child born in 2005 didn’t grow up in “a world with technology.” They grew up in a multiverse of curated realities, without anything to anchor them. Communication at once sped up and shortened. A thumbs-up replaced a sentence, a meme replaced an argument, a viral clip replaced a sustained narrative. We did not return to face-to-face depth; we compressed language into reactions.

Now layer artificial intelligence on top of this compression of reality and decay of stabilizers. It’s not about better autocomplete; it’s a question of offloading cognition itself.

At some point, the distinction between your native processing power and externally available intelligence will blur even further. With intelligence accessible on demand—and IQ not just downloadable, but embeddable, with the potential of brain-computer interfaces (like Neuralink and others)—raw cognitive horsepower stops being the differentiator. And not just between individual people, but between humanity and the technologies we’ve created.

If cognition itself can be outsourced, it becomes dangerously easy to outsource judgment, direction, the burden of deciding. And when all of this is happening while the environment is changing faster than our internal models can update, we begin to fall even further behind—cognitively, emotionally, institutionally. That gap between environment and understanding is where agency erodes most precipitously.

And (ominously) agency is not the only casualty. As technology grows better at satiating our low-level needs—efficiency, access to information, optimization, automation of routine tasks—the “upper levels” of human experience, including belonging, meaning, and purpose, feel less and less stable.

In this dawning future, the question is not: How powerful will AI become?

The real question is: How do we as humans retain meaning and agency in a world where the map keeps changing?

This newsletter is an attempt to explore this question in public.

To me, the answer is adaptability—the structured capacity to learn, revise, experiment, and update without losing direction. It is the ability to update your model of the world without collapsing into paralysis or chasing every new signal.

In a world where intelligence is commoditized, adaptability becomes the essential and indispensable human trait. Not who can calculate fastest, but who can navigate best. Not who has the highest IQ, but who can retain agency when IQ is everywhere.

And the good news is that—unlike IQ—adaptability can be learned.

After a career spent in high-stakes, fast-cycle decision-making, both building businesses myself amid uncertainty and learning from the most effective entrepreneurs who racked up “wins” across industries, I’ve observed that many of the most successful leaders are the ones who think most adaptively. Knowingly or unknowingly, many seem to be running a similar process—practicing adaptive thinking as a kind of discipline, not unlike a more advanced operating system for the brain.

Unpacking how, why, and what that looks like in practice—with an eye toward democratizing this process and building a community of adaptable thinkers—has become a major focus of mine, especially over the past couple of years. Sharing as much of what I’ve learned as possible will be our objective here.

In the next installment of this newsletter, we’ll start to unpack why decision-making today feels harder than it should—and why much of that difficulty has less to do with our intelligence than with the kind of uncertainty we’re living through.

For now, I’ll leave you with this:

The first cognitive revolution gave us the power to ask “What if?”

The second is testing whether we still have the agency to answer that question.

My book, The Adaptability Quotient, is out September 15, 2026. It’s a framework for thinking clearly and deciding well when the ground keeps shifting. CEO of Citadel Ken Griffin calls it essential reading for anyone seeking to sharpen judgment and thrive amid constant change. Pre-order at www.theaqbook.com.