Chapter 3The Age of Adaptive Insight

1. Organizational problems land on how a person judges #

The same set of AI tools, handed to different companies, can produce opposite results. One team uses them to shorten delivery. Another generates dozens of conflicting plans in a day. One lets newcomers pick up mature practice faster. Another only copies an unverified preference of the boss into more places. The tool has no stance. It will amplify the judgment it is given, and the error inside that judgment.

Keep taking that organizational problem apart and it lands on particular people. A product lead has to decide whose problem to solve first. A hiring manager has to decide what kind of person is worth bringing in. A founder has to decide where money and time go. AI can retrieve material, list options, play the other side, and produce a first draft. It cannot answer, for the person who owns the consequence, what matters most right now, which result counts as good, or which basis should rule the rest out.

Knowledge stock and execution speed used to be advantages in themselves. Finding existing knowledge is easier now, and AI can take on much of the digital execution. Things that used to sit together in one bag called "ability" are starting to show different prices.

The personal question then appears. Faced with a problem that has no ready answer, how do you find a basis for judgment? And how, through conditions and results, do you tell a reliable standard from an old habit or a lucky win?

2. Knowledge, cleverness, and adaptive insight #

Knowing a lot, reacting fast, and having many ideas often get rolled into one phrase: this person is able. Knowledge, cleverness, and adaptive insight do not solve the same problem.

Knowledge is a stock. It includes facts remembered, concepts understood, cases seen, and also methods of retrieval and checking. AI has lowered the cost of getting and calling much existing knowledge. It has not made knowledge itself useless. When the stock is thin, it is harder to see which field a problem belongs to, and harder to see the conditions under which a standard applies. Knowledge also sets the range of a check. The same model answer can be caught for mixed concepts, stale data, or missing conditions only by someone who holds the relevant concepts and facts.

The fall in cost is still real. Chapter 1 compared call prices on a fixed capability line: the drop is already more than two orders of magnitude. That curve measures model calls, not a claim that knowledge is already free, and it is enough to show that the barrier from asking a question to having a usable first draft is falling fast.

Cleverness, in this distinction, is solving speed. Once the problem and the goal are already given, can you see the relations faster, compare the options, and push through to an answer? That is not a full psychological definition of intelligence. Cleverness still has value. Hard problems still need working memory, reasoning, and speed of learning, and AI has not cancelled those. What has changed is that more and more solving steps, once they are clearly defined, can be done with a machine.

The occupational writing experiment and the customer-support study in Chapter 1 show that, in some digital knowledge work, steps that used to depend on fluency and solving speed are being reassigned. Those results do not prove that all execution will be taken over.

Adaptive insight handles the problem before the problem. It asks whether a judgment standard is missing, whether an existing standard has been verified, whether the conditions of transfer still hold, and how to get feedback soon. Knowledge helps a person recognize what can be borrowed. Cleverness helps a person understand and work it. Adaptive insight decides whether the question and the ruler were chosen right.

The three do not replace one another. Without knowledge, a borrowed standard cannot be read. Without enough cleverness, complex conditions cannot be taken apart. Without adaptive insight, knowledge and cleverness can still serve the wrong question. Material can be memorized and numbers can be computed, and the conclusion can still be useless because the problem or the metric was wrong.

Knowledge and cleverness have not left the field. Once tools can supplement them more easily, more of the gap moves to another layer: whether a person can find where the basis for judgment is missing, and repair it until it can guide action.

3. Adaptive insight is four actions you can practice #

"Adaptive insight is a gift" gives no way in for practice or for a test. Adaptive insight is not yet a psychometrically validated ability, and there is no evidence that training will bring everyone to the same level. The working definition is four actions that can be watched, repeated, and reviewed.

The first action is to discover the gap in standards.

What is often missing is not an answer, but a basis for judging answers. A team puts three growth plans on the table, with data, budget, and timetable complete, and still cannot say whether short-term cash, retention, or long-term brand trust should rank first. Adding a fourth plan will not usually make the choice easier. The gap is not the number of options. It is the standard used to rank them.

Seeing that gap means separating "we still need a plan" from "we still need a scale for ranking plans." The first can keep being solved. The second has to find a ruler first. The test is simple. If two plans both look doable, on what basis do you drop one? If the answer is only a personal feel, an industry custom, or a boss's preference, the gap is still there.

The second action is to borrow a verified standard.

Building a plan from zero the moment the gap appears skips evidence that already exists. Original work is sometimes irreplaceable. Most daily problems are not appearing for the first time. Searching first for results, bases of judgment, and failure records already obtained under nearby conditions is usually safer than paying the full tuition again.

What you borrow is the basis that once helped someone choose, not a winner's whole package and not a saying stripped of its conditions. That basis may come from one person's practice, from a company's public method, or from a study that can be checked. Borrowing needs a knowledge stock: you have to know which fields have solved a similar problem before you can find a standard worth taking. Borrowing is not a stop to thinking. It gives thinking a starting point that has already met reality.

Not every problem has a fitting standard to borrow. When the problem is new enough, the environment has shifted too far, or the available samples cannot be compared, a new hypothesis still has to be made. Mark it as a hypothesis under test, not as a settled answer.

The third action is to judge the conditions of migration.

A standard that held in its original scene does not hold automatically in a new one. The customers, product, resources, time window, and way of getting feedback it depended on may already have changed. The migration check watches two things: why the practice held at origin, and how many of those supporting conditions remain.

A résumé and a reputation cannot replace the check. A record of success only says the standard held in its original scene. It does not carry the conditions of use with it. A standard has no passport. Origin in a famous company or in a celebrated person does not grant it passage. Before you move the practice over, at least three things have to be clear: whether it was solving the same problem, whether the key people still respond in a similar way, and whether an error can be found while the cost is still bearable.

The fourth action is to verify and correct quickly.

However careful the migration analysis, it is still a judgment. What it missed can only show up in results. The first try should use less money, a shorter cycle, and a change in only a few key conditions, so that error appears while it is still cheap. If results match the expectation, widen the range. If they do not, check the standard itself and the conditions of transfer.

Verification is not a search for support of your own judgment. It is an active search for where that judgment will fail. One good result does not automatically produce a standard. Luck, a loose environment, or other factors moving at the same time can produce it too. Only when results repeat, conditions of use become clearer, and counterexamples force a revision does experience begin to have explanatory power that others can reuse.

No study has yet tested these four actions as one package. Adjacent evidence exists for the abilities inside them. Gutierrez de Blume, pooling 56 independent effects and 7,667 participants, found that training in learning strategies improved, to a moderate degree, how accurately people judged how much they actually knew (Gutierrez de Blume, 2021). Alfieri, Nokes-Malach, and Schunn, pooling 57 case-comparison experiments, found that comparing cases side by side and extracting the shared relation helped learning and transfer more than studying a case in isolation, with an overall effect size of 0.50 (Alfieri et al., 2013). The first kind of study sits near discovering a gap and using feedback. The second sits near borrowing and migration. Those learning studies do not guarantee a business result. They do show that the related actions can be practiced, and do not depend on talent alone.

The four actions leave a process that can be checked: where the gap was, where the basis was borrowed, how old and new conditions were compared, and how results revised the judgment. After one round, what you hold is not an answer that will always be right. It is a standard of higher quality. When a new scene arrives, it still has to take the next test.

That is the exact sense of "can be practiced." It does not promise that after practice you will never miss. It means that finding the miss, explaining it, and correcting it no longer depend on accident.

4. With a standard, nearly a line; without one, forks #

Drawn on paper, the four actions produce two different paths.

A standard that has been verified, and that applies to the scene in front of you, rules out many options in advance. Action can still miss, and it still needs feedback, but most decision points do not have to be groped from the start. The path looks more like a line with small corrections: act by the standard, watch the result, then adjust the deviation.

Without a standard, every node keeps forking. A or B first, price or quality, widen the range or redesign—each question has to be decided on the spot. Walk the wrong way, come back, and the new path throws off new branches. What is expensive is not any one choice. It is that none of the choices share a basis, and past feedback has nowhere to land.

Figure: a usable standard narrows options. The four moves find, borrow, migrate, and revise the standard.

A standard does not guarantee success. An expired standard, an unverified one, or one moved from elsewhere without a check on conditions can all produce a straighter path to the wrong place. When the problem is genuinely unknown, the fork itself has value, because different options produce new feedback.

Adaptive insight does not kill exploration. It separates two kinds of fork: the cost that has to be paid to find new knowledge, and the cost paid only because a basis for judgment was never made clear. It does not hand over a one-shot best route. It requires each action to answer two questions: why this way, and what to change if the result is wrong.

5. Tenure records time; verification records growth #

"Two kinds of decade" is, for now, only a portrait awaiting a test, not a long comparison of people of the same age in the same job that has already been found. Two résumés both say ten years. Asked about a failed project, one record holds the basis of the judgment, the contrary feedback, and the later adjustment. The other holds only "the industry has always done it this way."

Existing research supports a more modest claim: years help, and they are not a sufficient account of performance. Quiñones, Ford, and Teachout, pooling 44 studies and 25,911 participants, found a corrected correlation of about 0.27 between work experience and job performance. Experience and performance are not unrelated. The relationship is far from strong enough for "how long" to stand in for "how well." More specific task experience and amount actually completed correlated at about 0.41 and 0.43, higher than the general experience measure (Quiñones, Ford, and Teachout, 1995).

Later work cut "experience" finer. Ng and Feldman, pooling studies of job tenure and performance, found that remaining longer in the same job had little clear relationship with core-task performance (Ng and Feldman, 2013). That does not contradict the earlier finding. How many concrete tasks a person has done, and what kind of experience they accumulated, may help. How many years they spent in the role, by itself, carries much less information.

Years mainly record how long someone stayed on the floor. They do not record which feedback arrived, or whether the next judgment was changed because of it. A decade can hold a large number of different tasks, or the same set of moves repeated. Time alone cannot tell them apart.

Experience has value. It helps a person recognize an anomaly faster, filter the details that matter, and accumulate a feel that never made it into a book. The problem begins when experience is given the last word. One success may come from a right method, or from a market tailwind, a resource edge, a rival's miss, or luck. Without compared conditions, tracked consequences, and revision after a miss, there is no way to tell which of those did the work.

Experience proposes a candidate standard. Verification decides whether it has explanatory power. A candidate can move the starting point of the next search forward. It still has to state the conditions under which it holds and the signal that would force a revision. After that process, tenure is no longer only a sum on a calendar. It leaves more and more bases of judgment that can be reused.

Years of work record time. Verified standards record what a person learned from the time. A reliable standard often needs long contact with real problems before it forms. What has to be ruled out is not experience. It is experience taking the last word before it has been tested.

6. Intuition can hold a standard, or a bias #

A skilled person cannot always list the full basis of a judgment. Some anomalies and risks are recognized first; the reasons can be put into words later. That kind of tacit knowledge does real work.

People can do first and learn to explain after. Long training speeds up many acts of recognition, until the person judging cannot report every cue in time. A practiced doctor, firefighter, designer, or salesperson may see the anomaly before a complete inference appears. Intuition can sit with a standard.

Reliability is a separate question. Kahneman and Klein, comparing the research traditions of "expert intuition" and "heuristic bias," drew a clear boundary. The environment has to have regularities that are stable enough to learn. The person judging has to have had enough practice. Feedback has to come relatively fast and clear. When those conditions hold, people can learn to pick up valid cues. When the environment itself is hard to predict, or results arrive much later and mixed with many other factors, confidence cannot prove accuracy (Kahneman and Klein, 2009).

Reliable intuition can be read as a pattern, trained on a great deal of valid feedback, matching at speed. The person may not be able to name every cue at once, but those cues have been tied to results many times. Unverified experience produces a feeling of familiarity too, without evidence that the practice caused a good result. Familiarity itself cannot tell expert intuition from bias.

The quality of an intuition depends on whether the environment in which it formed was stable, how long results took to return, whether the feedback was clear, and whether failure samples appeared. A stable scene with dense feedback can form reliable intuition. A new scene whose rules change fast will lower that reliability. Intuition, like any other standard, has to be tested again when it migrates.

Tacit knowledge does not have to be written as a full manual before anyone acts. The most important parts can be recorded first: which signals mark an anomaly, which experiences the judgment came from, and what result would prove it wrong. Not being able to say it all does not make the judgment invalid. An intuition that cannot name its source and will not take a counterexample also cannot take the last word on seniority alone.

7. Two migrations: Jobs and Ron Johnson #

The four actions of adaptive insight are easiest to see in a migration. Two people can bring something learned elsewhere into a new scene, and the origin of the standard does not decide the result. What differs is whether the original problem and conditions still exist, and whether a path for feedback was kept open before the full commitment.

In 2005, Steve Jobs, in his Stanford commencement, recalled dropping out of Reed College and staying to sit in on classes that interested him. One of them was calligraphy. He learned serif and sans-serif typefaces, the spacing between letter combinations, and "what makes great typography great." At the time, he said, he could see no practical use for it.

Ten years later, when the team designed the first Macintosh, the material returned. Jobs said it was designed into the Mac, so the computer had multiple typefaces and proportional spacing; without that class, the Mac would not have had those features (Jobs, 2005). The retrospective account records a migration of knowledge. It does not record how the team made each typographic decision, and it cannot attribute the Macintosh's success to one calligraphy class.

Independent design records show that Macintosh typefaces and icons were designed mainly by Susan Kare. The team cut fonts letter by letter, offered repeated options, and settled visual design through discussion and informal tests. Jobs often reviewed progress and talked through visual details; the concrete decisions were made by the team together (Kare, 2000; Computer History Museum, 2017). The division of labor in the record is this: Jobs brought the importance of typography into the product; Kare and the team made it into usable digital type and interface.

The migration kept the problem the original knowledge had handled. The medium moved from paper to screen; typography still had to deal with differences of letterform, spacing, and the feel of reading. A borrowed standard does not have to come from the same industry. If the new scene still holds the problem it solved, it can still produce value. The available material does not record Jobs walking the four actions item by item. It does record calligraphy and typographic knowledge entering a personal computer.

Ron Johnson's story runs the other way. He joined Apple in 2000 and later became senior vice president of retail. In June 2011, JCPenney named him its next CEO; he took the job on November 1. The appointment announcement listed his retail record at Apple and Target as the core of the résumé.

Once in office, Johnson proposed Fair and Square pricing: a more direct price in place of constant promotion. When JCPenney presented the plan, it said the company had run about 590 promotions in 2011. The new plan changed the rhythm to twelve monthly events a year, with everyday prices, monthly value, and clearance prices still in place. The new pricing began on February 1, 2012, across about 1,100 stores (JCPenney, 2012).

The company described the plan as cutting coupons, rebates, and promotional theater so that price would be more direct. The trading conditions at Apple and at JCPenney were very different. Apple sold highly differentiated products of its own in its own stores. JCPenney was a multi-category department store, where customers could compare similar goods across many retailers. Johnson later acknowledged that core customers' dependence on, and preference for, coupons had exceeded what he had understood (TIME, 2013). A retail belief that held at Apple still needed to be verified again in a department-store scene.

Verification is the sharper break. TIME, writing after Johnson left, reported that when someone suggested testing the new pricing on a limited scale first, he was said to have answered, "We didn't test at Apple." The line is a press paraphrase, not a checkable recording. It is consistent with the public method of rollout: the core pricing change was not run in a few stores first, with a decision to widen based on customer response. It was implemented at large scale from the start (TIME, 2013).

Operating results worsened quickly. JCPenney's total sales in fiscal 2012 fell from $17.260 billion to $12.985 billion, down 24.8% year over year. Comparable-store sales fell 25.2%. Full-year net loss was $985 million; fourth-quarter net loss was $552 million (JCPenney, 2013). On April 8, 2013, Johnson left, after about seventeen months in the role, and his predecessor Mike Ullman returned.

In the same period Johnson also changed the merchandise mix, the shop-in-shop format, and store layouts, and JCPenney was already under operating pressure. This was not a single-variable experiment. The whole decline cannot be attributed to ending coupons. What can be confirmed is that a package of changes, not first run as a limited test, was pushed at large scale in a scene whose customers and goods differed, and that a large operating loss followed.

The public record shows a break in the last two actions. The conditions Apple's experience had depended on were not fully rechecked in the department-store scene, and a limited test did not happen before the full commitment. Feedback finally arrived as company-wide operating results, and the cost of correction was borne company-wide.

Jobs and Johnson both brought something obtained elsewhere into a new scene. A successful record can only offer a candidate standard. It cannot vouch for the new scene. When a migration keeps the conditions that matter, old knowledge can open a new possibility. When it skips conditions and verification, a past success record cannot guarantee the result.

None of the four actions is optional. Discovering the gap shows that the problem is not more execution. Borrowing a standard keeps exploration from always starting at zero. Judging migration keeps the look of success from being taken for its cause. Verifying and correcting gives results a veto while the error is still cheap. Miss any step, and existing knowledge and execution speed cannot stop an old standard from being used again.

8. Everyone's head holds unwritten, unaligned rulers #

Personal judgment calls on knowledge, experience, and intuition. Repeated choices and feedback gradually form many rulers: what counts as a good product, whom to trust, which risks cannot be taken, which chances are worth waiting for.

Some of those rulers have been through repeated feedback. Others have only been used for a long time. Some can state their conditions of use. Others can only be spoken as a feel. They may help a person get a string of decisions right. They may also keep pointing to an old answer after the environment has changed. The four actions of adaptive insight are there to let a person find where those rulers are missing, what to add from outside, whether a move-over still fits, and how results keep calibrating them.

As long as a ruler stays inside one person's head, it has a clear boundary. A second person cannot check what it rests on, and cannot tell whether what they have borrowed is a principle, a habit, or a coincidence. When the person who judged leaves, many of the conditions and counterexamples leave too. Even two people with strong adaptive insight may each borrow a standard that works, and still give opposite answers to the same matter.

Tacit knowledge cannot be written down complete in one pass, and action does not have to wait until every part can be said. The most important marks can still be left out: the basis of the trade-off, the conditions under which it holds, and the result that would force a revision. Once that step is taken, a personal judgment has a chance to leave the person, and to be understood and tested by someone else.

Everyone's head holds unwritten, unaligned rulers. When a group of people each act by a different one, a new problem appears: how do those rulers become an asset that can be used in common?

What to Do Monday Morning #

Pick this week's most important decision, the one for which the evidence on hand is thinnest. Do not ask AI for ten more plans first. On one sheet, write four columns:

1. Gap. What would count as good here? Which judgment standard am I missing?
2. Borrow. Who has gotten a result that can be checked under nearby conditions? Which standard am I preparing to take?
3. Migration. In object, resources, stage, and way of getting feedback, what differs from mine?
4. Verification. What is the smallest try I can run this week? What result would force me to change the judgment?

A week later, add one line: what feedback reality gave, and which judgment has to be revised. Walking the four questions once a week is closer to training adaptive insight than writing one review from memory at the end of a year.

Chapter Acceptance Self-Check (against chapter acceptance criteria) #

1. Claim restatable in one sentence ✓: as retrieval and much digital execution get cheaper, more of the gap depends on adaptive insight's four actions.
2. Eight-section spine and whiteboard figure ✓: personal landing → knowledge / cleverness / adaptive insight → four actions → line and fork → tenure versus verification → intuition → Jobs / Johnson → unwritten rulers. Figure footer no longer previews Chapter 6.
3. Evidence caliber ✓: price and workplace studies referred back to Chapter 1; Gutierrez de Blume and Alfieri as adjacent learning evidence, not a business-result guarantee; Quiñones / Ng and Feldman on tenure; Kahneman and Klein on intuition; Jobs from the 2005 speech plus Kare / Computer History Museum; JCPenney from company materials and TIME paraphrase, without single-cause attribution.
4. Case contract ✓: winner is Jobs bringing typographic knowledge into Macintosh; loser is Johnson moving Apple retail judgment into JCPenney without a limited test. "Two kinds of decade" remains an unverified portrait. Author financing sample is out.
5. Chapter boundary ✓: does not define Chapter 4's two unifications, does not issue Chapter 6's stop rule, does not open Chapter 7's Clarity Method.
6. Action ✓: four columns on one judgment this week, plus one line of feedback after seven days.
7. Fluency ✓: rewritten in English voice from the Chinese authority; current prose-standard.