AI-Proof: What Machines Can Never Be — Volume I: Awakening, Part III: The First Awakening, Chapter 13: The Meta-Skills That Compound Forever

Part III is the first awakening. After the daily practice of Chapter 12, this last chapter of Part III, and of Volume I, asks what carries a person through change they did not choose.

← Previous: Chapter 12: Becoming AI-Leveraged · Series hub · Next: Volume II: Application (overview) →

Kavitha was told to be certified in a cartoon of what she already did. What carried her through was not a skill that compounds forever. It was a character practised in small acts, and no rubric could see it.

In Brief: For readers who are being asked to adapt to a change someone else chose. Chapter 13 is about what carries a person through change they did not choose. A teacher in Sagar is ordered into a certification that scores her teaching by counting machines; a marine biologist in Da Nang loses her funding to a free platform that serves the fishing communities better; a floor supervisor in Tiruppur reads the grief of two inspectors moved to packing, and pays for the reading. Aristotle on how character is built and lost, the Gītā on evenness as skill in action, the Tirukkuṟaḷ and modern psychology on reading faces, Hochschild on emotional labour and Hirschman on voice help answer what lets a person learn from change without surrendering to it, and who should carry the cost.

What carries a person through change they did not choose is not a skill that compounds forever but a character practised in small acts: it is built by doing, lost by not doing, and no rubric can certify it.

Opening: The Email at the Staffroom Window

The email arrived on a Tuesday in December.

It was 16 December 2025. It came from the district education office in Sagar, over the name of the district education officer, Suresh Tiwari, and it went to every government schoolteacher in the district. It said what Kavitha had expected for months. Every teacher must complete a sixty-hour certification in the use of AI in teaching by 31 March, the end of the school year. Those who did not would have it noted in their annual appraisal.

Kavitha read it standing at the staffroom window, with the neem tree outside. She read it once, put her phone in her bag, and went to teach her afternoon class on chemical bonding.

She was not afraid of the technology. She had used machines for lesson outlines and parents’ notes for well over a year, and the district’s platform had been in her classroom since August. Chapter 10 told how she learned to check her own eye against it, and Chapter 12 how she sorted her work by how much checking each kind needed. She had a rule, a notebook and a gradient. What she felt, reading the email, was not fear. It was closer to contempt.

Sunita, who teaches mathematics in the room next door, opened the portal that evening and showed her the first modules over tea the next day. They taught teachers how to log in to the platform Kavitha had used for four months. They explained, in the flat language of government manuals, things she had found out by doing them. And they were marked by a rubric that counted how many stages of a lesson used AI, not whether the lesson taught anyone anything.

The certification was a cartoon of the practice she had built. She could not make herself sit down to it.

So she did not. Not that week, not the next. The half-yearly results came first, then the parents’ meeting, then the science exhibition. Each delay felt reasonable. None of them was the reason.

It took her eight weeks to name the reason, and it had two layers. On top was contempt for a rubric that would score her teaching by counting machines. Under it was something she liked less: a fear of being scored at all, and found ordinary. The certification did not threaten her competence. It threatened her belief that her competence mattered on its own terms, without anyone’s marks.

This chapter is about what carries a person through a change like that. It is a change they did not choose. It arrives in a form that insults them, and it still has something in it worth learning. So the chapter asks: When change arrives in a form you did not choose, what lets you learn from it without surrendering to it, and who should carry the cost?

The print edition answered with four “meta-skills”: adaptability, learning velocity, emotional intelligence and resilience. It said they “do not depreciate with time. They strengthen. Each use deepens the capacity for the next.” The title of this chapter makes the same claim, and it needs an argument before the stories begin.

Compound is a word for money. Interest is paid on interest, and the sum grows by itself while its owner sleeps. Nothing in a person grows like that. Forever is a word for no human thing at all. Chapter 11 found that what compounds in judgement is a habit of not lying about luck, and only where feedback is honest and the next problem resembles the last. Chapter 10 found that depth fills a position and does not grant one. And meta-skill, a skill for gaining skills, puts these four in the wrong family.

Aristotle gives them a better one, though not the one a reader might expect. In the Nicomachean Ethics he describes how a settled state of character is formed. It is not given by nature, and, unlike the virtues of thought, it does not grow mainly by teaching. It is built by doing. In W. D. Ross’s translation: “we become just by doing just acts, temperate by doing temperate acts, brave by doing brave acts.” Then he adds the sentence that takes forever out of the title. “Again, it is from the same causes and by the same means that every virtue is both produced and destroyed … for it is from playing the lyre that both good and bad lyre-players are produced.”¹ What practice builds, the lack of practice and bad practice unbuild. A disposition is not a deposit. It is built by doing and kept by the conditions that let it be used. It can weaken from disuse, be bent by bad practice, or be held down by circumstances that leave it no room.

Aristotle draws one more line that this chapter needs. A craft and a virtue are judged differently. “The products of the arts have their goodness in themselves”: a good pot is a good pot, whoever made it and however. But an act that looks just is not yet a just act. “The agent also must be in a certain condition when he does them.” He must know what he is doing and choose it for its own sake, and he must act “from a firm and unchangeable character”.² A craft can be tested by its product. A virtue cannot, because it lives in the person, not in the thing made.

That is why the certification felt like an insult, and it is also why it could not have been anything else. A rubric inspects products. It can count the stages of a lesson plan that use AI. It cannot see whether the teacher who wrote the plan knows what she is doing. It cannot see whether she chose it for its own sake, or would do it again on a day nobody marks. Aristotle would not have been surprised by Kavitha’s contempt. He might have asked her what she expected a rubric to do. This does not mean that a disposition leaves no trace. A colleague who has watched someone teach for years can give real evidence of one. What no test can do is certify it. A certificate fixes a status on a date. A disposition is kept only by use, and it can be lost the next year by disuse.

So this chapter keeps the four: adaptability, learning, reading people and recovery. They are not a complete list of what a good character needs. They are the four that people are most often told to have when a change is done to them, and that is reason enough to look at them closely. The chapter treats them as dispositions, not as assets: built in small acts, under conditions nobody chose, and able to be lost. Whether they are also virtues is a separate question, and 13.5 asks it.

13.1 Adaptability: What Should Change, and What Should Not

Making change your default mode is the wrong goal. A person who changes with every wind has no shape for change to act on. Aristotle placed each virtue between two failures. In Ross’s words, virtue is “a state of character concerned with choice, lying in a mean, i.e. the mean relative to us”.³ Adaptability has two failures on either side of it: holding on to a form after the world has stopped needing it, and dropping everything at the first pressure. The disposition lies between them. It is the ability to tell which parts of your work should change and which should not.

Trần Thị Mai is a marine biologist in Da Nang, in central Vietnam. She spent four years building a network of underwater sensors on three reefs, measuring temperature, acidity, dissolved oxygen and light every twelve minutes. The data fed a model she had built during her doctorate at the University of Queensland. It warned of coral bleaching early enough for the fishing communities nearby to change where and when they fished. It worked. It had been checked against what happened on all three reefs.

Satellite warnings of bleaching were not new. The United States’ National Oceanic and Atmospheric Administration has published free maps of heat stress on the world’s reefs since 1997. Since 2020 its main daily maps have had a resolution of five kilometres.⁴ Mai had built her network to see what the satellites miss: water temperature below the surface, and the muddy water close to shore where coastal reefs live. Early in 2026 a research group in Singapore released an open AI platform. It combined satellite data, ocean currents and the region’s published sensor records into forecasts for the whole coast, updated every hour. It was free. For most reefs it was good enough. It covered every reef the fishing communities used, not only Mai’s three.

Her first response was to discredit it. She found cases where her sensors caught warming that the platform missed. She wrote a memo to her research director arguing that its coverage came at the cost of local precision. Everything in the memo was true. Within two months her funder moved its money to the Singapore platform anyway.

This call went the machine’s way, and it should be said plainly. The funder’s test was simple: warnings for every reef the communities fished, in time to act on them. On that test the platform won. Mai’s network was better at three places. The platform was better for the people the network had been built to serve.

Adaptability starts earlier than the memo. Mai had read the papers and heard the talks. She had filed the new platforms under “not yet good enough to threaten my work”. That was correct in 2024. It was wrong by 2026. The time between the two was the time in which she could have changed on her own terms. The habit that would have shown her the change is the one she later sold. She could have tested the new platforms against her own reefs each season, before anyone asked her to. Kavitha made the same mistake in another register. She had watched the district’s digital attendance pilot, the state’s budget for educational technology, and the national policy’s language about AI in classrooms. Her assumption, built over a decade, was that the district office never followed through. It had been true for years. It was not true now.

The harder part is the self. Mai described herself as “the person who built the reef monitoring system”. That was a fact about her work, and also a fact about who she was. When the system was set aside, the second fact seemed to go with the first. Chapter 2 met Nāgasena’s chariot, which turned out to be a name for its parts. A working identity is something like that: a name for a set of practices, held together by habit. It can be rebuilt from the same parts.

By September 2026 Mai had rebuilt hers, partly. She took a contract with the Singapore group to test its forecasts against ground truth. That meant her four years of sensor records, and new sensors on reefs that the satellites read badly. That work is also how the platform’s misses would be found. Her knowledge had not changed. Its use had, and so had her power over it. She no longer decided how her knowledge was used, on what terms, or who gained from it. A contribution can stay valuable while the person who made it loses that power. It is less secure work than she had planned, on someone else’s platform, and it may not last. The print edition said that it was “the career the world needed” and that she was “the only person in Vietnam” who could do it. Neither is this book’s to say. What can be said is narrower. She kept the part of her work the new system needed and gave up the part it did not. After two months of not being able to, she could tell which was which.

Kavitha’s identity was “the teacher who tells stories”. It was accurate, and Chapter 10 shows how hard it was won. But in December she had let it harden into an opposition: the teacher who tells stories against the teacher who fills in rubrics. The difference was real. The opposition was not the whole truth, as she found in the sixth module.

There is a second kind of adapting, and it matters more for Kavitha than for Mai. Adapting can change how you pursue a goal. It can also mean asking whether the goal is right. Mai had to change how her knowledge was used, but the goal, warning the fishing communities in time, was sound. Kavitha was asked to change her methods towards a goal she doubted: a lesson plan with machines in every stage. A person can become very good at adapting to a bad measure, and make an institution better at pursuing the wrong thing. That is accommodation. From outside it can look like adaptability. It is not the same thing.

13.2 Learning Velocity: What You Already Know

The print edition told this part of the story as a race. Kavitha logged in during the ninth week and overtook Sunita, who had started the day the email came. The print edition explained the gap as “root depth”. There is something true in that, and something that needs correcting.

The true part has a classic statement. The educational psychologist David Ausubel opened his textbook of 1968 with it: “The most important single factor influencing learning is what the learner already knows. Ascertain this and teach him accordingly.”⁵ New knowledge sticks when it has something to attach to. When the third module described “adaptive learning pathways”, Kavitha recognised her own afternoons. The platform sent the students who were struggling through extra practice while she took the others out to the neem tree. When the fourth module named “human-in-the-loop assessment”, she found a name for what she had written in her notebook since September. Chapter 5 met Nārada, who came to his teacher knowing every name in the learning of his day and was told: “Whatever you have read, is only a name.”⁶ Kavitha was in the opposite position. She had the things, and the course gave her names.

Sunita’s slower pace meant something simple. The platform had not come to her mathematics classes; the district’s pilot covered science. For her, the modules were new information, not new words for old practice. She worked through them two hours most evenings, taking notes, finishing each exercise before the next. That is how anyone learns what is new. It says nothing about which of the two is the better teacher.

The correction is this. Prior knowledge does not only speed learning. It filters it. What fits goes in fast. What contradicts it slows down, or is skipped. Kavitha raced through the modules that described what she already did, and she nearly skipped the one that did not.

The sixth module was about the platform’s reports on each student. Kavitha had glanced at them weekly and trusted her eye more. The module taught her to read them differently. She should look not only at accuracy but at time per question. And she should notice when a student’s speed rose while the difficulty of the questions did not. That pattern, the module said, could mean a student who had learned to answer without understanding. She resisted it for a week. It said that her eye, which her stories depended on, could miss something a table would show. Her notebook had already told her that her eye misread silence. The module said it could misread success too.

She found the pattern in Rohit. He was fourteen, in the third row, and the platform rated him above average. His accuracy was high. His time per question had fallen for a month while the level of the questions stayed flat. In class he had stopped asking questions. She sat with him after school and asked what he thought about while he did the exercises. He told her, a little proud, that he had worked out the platform’s habits. He could predict the next kind of question from the last three. He liked to finish early. She did not ask why.

The print edition said that what Kavitha saw was “invisible to any instrument except a human being who is paying attention”. That is not what happened. The instrument showed the sign, once she had learned to read it. But a falling time per question is a clue, not a diagnosis. The data showed her where to ask; it could not give her the answer. What it could not do was ask Rohit what he was doing, and whether he understood the bonds he was answering about. Data and a question together did what neither did alone. She changed two things. She asked Rohit to explain one answer aloud each week. And she wrote to the district’s coordinator for the platform that its questions came in a predictable order. That was a fault in the machine, and it could be fixed. It was also a chance for the system to learn from one teacher, and not only for the teacher to learn from the system.

So learning was fastest where Kavitha already knew, and deepest where the new thing showed her a real limit. The modules that confirmed her taught her vocabulary. The module that contradicted her taught her something. Contradiction does not teach because it is uncomfortable. It teaches when the limit it shows is really there. That is the part of “learning velocity” worth keeping: the willingness to slow down exactly where the new thing disagrees with you, and to check whether it is right.

One more thing about the sixty hours. They came out of evenings. The district gave no periods for the course, and Sunita’s two hours a night were her own. Chapter 12 asked whose hour the saved time is. Here is the same question from the other side: whose hours pay for the learning a mandate requires? A mandate that requires learning and does not give the hours for it moves the cost from the office that chose the change to the people it was done to.

13.3 Emotional Intelligence in Human-Machine Teams

The print edition opened this section by calling emotional intelligence “one of the irreducible human advantages”, and said that understanding a person “requires a self that has lived”. Both claims put the weight where this book has decided not to put it. The question here is narrower and harder: who reads whom, how well, and who pays for the reading. In its usual sense, emotional intelligence is the ability to notice and handle feelings, your own and other people’s.

Reading faces is older than any machine, and less reliable than we think. The Tamil Tirukkuṟaḷ has a whole chapter on it, which V. V. S. Aiyar titles “Judging by Looks”. “Even as the crystal changeth and assumeth the colour of that which is near,” it says, “even so doth the face alter and show that which overfloweth the heart.” Its advice is addressed to the king: “Behold the men that can judge a man’s intentions from his looks: take them into thy council at whatever cost.”⁷ In 2019 five scientists led by the psychologist Lisa Feldman Barrett reviewed more than a thousand published findings and came to the opposite view. A smile, a scowl or a frown can mean different things depending on the person, the situation and the culture. In Barrett’s summary: “It is not possible to confidently infer happiness from a smile, anger from a scowl or sadness from a frown.”⁸ The two lamps are not answering quite the same question. The Kuṟaḷ is counsel to a ruler about whom to trust; the review tests what a face can tell an observer about a feeling. But they pull against each other, and Kavitha’s notebook had already sided with the review: she had misread Pooja’s silence three times. The review does not say that faces carry no information. They are evidence to interpret, not a window into a person.

Machines that claim to read faces rest on the Kuṟaḷ’s confidence, not on the review’s caution. Since February 2025 the European Union has forbidden the use of AI systems to infer people’s emotions in workplaces and places of education, except for medical or safety reasons.⁹ It is a European rule, and it does not reach a school in Sagar or a factory in Tiruppur. The law does not say that a machine can never read a person. It says that in the places where people cannot easily refuse, the reading is not good enough, or not fair enough, to be done to them.

Lakshmi is a floor supervisor at a garment factory in Tiruppur, in Tamil Nadu. In 2025 the owner installed three camera systems that scan finished garments for defects: missed stitches, uneven colour, seams out of line. They are faster and more consistent than the human inspectors they replaced, and they catch defects on the night shift that tired eyes used to miss. The owner considers them his best purchase in five years. On his measure, he is right.

The inspectors were not dismissed. Two of the best, Selvi and Sumathi, each with more than ten years at the inspection tables, were moved to packing and sorting at the same wage. The owner meant it kindly, and it was kind. It kept their income. It took their craft. Selvi had been proud of catching what others missed; now she worked in silence. Sumathi had trained two generations of inspectors; now she took the long way round the floor so as not to pass the stations where the cameras hung. Nothing in the factory’s figures changed. Their care for the work, which had made the inspection line one of the best in the building, quietly left it.

What Lakshmi read was not a face. It was ten years of two people, and a change in them. She knew them because she had recommended both for the inspection tables, and had once told Sumathi that her eyes were the best quality control the factory had. Now she supervised their move. Her guilt was specific, and she had nowhere to put it. She asked Selvi once how she was. Selvi said it was fine, akka, elder sister; the wage was the same. Lakshmi did not believe her. She also knew that ten years of knowing someone can make you sure of the wrong thing. She kept a calm face on the floor for the women, and a businesslike face in the office for the manager.

The American sociologist Arlie Russell Hochschild named this work in 1983. Studying flight attendants, she called it emotional labour: “This labor requires one to induce or suppress feeling in order to sustain the outward countenance that produces the proper state of mind in others.”¹⁰ It is real work, it is tiring, and it is almost never counted. Notice who does it. In the Kuṟaḷ, the readers of faces are taken into the king’s council; the reading is done in the service of power. In Hochschild’s airlines, the smiling was done by those lower down. On Lakshmi’s floor, she read the women, and the manager did not read her. Being good at reading people is a disposition. Being the one expected to do it is a position, and the two should not be confused. And nobody’s job was to read Lakshmi. A factory that wants supervisors to see what the cameras do not see has to notice what the seeing costs them.

In January 2026 the factory’s scheduling system recommended cutting Selvi’s hours, because packing needed fewer people. Lakshmi went to the manager and asked him to hold the cut for a month. She said what she could not prove. Selvi was already close to leaving. She knew more about the factory’s fabrics than anyone on packing. A cut would tell her she was finished. The manager heard sentiment. He held the cut for a month, and asked her for a reason he could write down. She did not have one in his language. Her standing with him fell a little, and she knew it.

The print edition said that Lakshmi “was right to” override the system. That is not for the story to settle. Chapter 11’s rule applies: judgement is not the override. The person says what she cannot prove, and someone with the authority to decide, decides. Here the manager decided, for a month. Whether he was right will be known only by what happens to Selvi, and even then only in part.

There is a quieter loss on that floor, and Chapter 11 promised that this chapter would take it up. In Pune, Mr Deshmukh kept a small team checking a rotating sample of parts by other means. That keeps some eyes in practice, though not every eye, and not for every kind of defect. In Tiruppur no one is kept looking. Selvi’s eye for a puckered seam under a shiny fabric was built at the inspection table, over thousands of garments. At the packing table it will fade. In 1983 the psychologist Lisanne Bainbridge warned that skills which are not used decline (Chapter 3). And Sumathi will not train a third generation. Chapter 4 described three generations. The first learned the work before the machines and can still tell when the machine is wrong. The second learned beside them. The third will enter after. Tiruppur’s third generation will never stand at an inspection table.

Lakshmi tried one more thing in February. She asked the manager whether Selvi and Sumathi could spend part of each week checking a sample of what the cameras had passed. They would look for what the cameras missed, as Deshmukh’s team did in Pune. The manager said the owner would have to decide, because it cost two people’s time. By the end of March the owner had not decided. A machine’s ability to do a task settles what can be automated. It does not settle everything that should be reorganised around it: who keeps looking, who learns, and who gets the gains. Keeping some human inspection would not be nostalgia for the old tables. It would be a decision about which abilities the factory will still need, and who will have them, on the day the cameras meet a defect they were never trained to see. The inspectors have not lost their eye yet. They have lost the occasions to use it.

The strongest evidence against this worry deserves its place. Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied customer-support agents in a large study published in 2025. An AI assistant raised the number of problems solved per hour by 15 per cent on average. Less experienced and lower-skilled workers gained most, in both speed and quality. The most experienced gained a little speed and lost a little quality.¹¹ The machine seemed to pass the habits of the best workers to the newest. That looks like the opposite of losing skill, and for the work being measured, it is. The question it leaves is what the novices no longer practise, because the machine practises it for them. The study also found evidence that the assistant helped workers learn. So the honest question is not whether such tools always wear skill away. It is what decides which way it goes. Does the tool show people the reasoning, or only hand them the result? Do they still get to do the work themselves? The factory owner did not ask that question. Nobody had asked him to.

13.4 Resilience: Coming Back to the Work

The recovery that follows is real. Calling it a skill, and calling it conscious, adds nothing that the story does not show better.

By the first week of March, Kavitha had passed every module test so far. Then she reached the eighth module, “Designing AI-Augmented Lesson Plans”. Teachers had to submit a lesson plan that used AI in their teaching. She sat down to write hers on a Saturday morning, opened a blank document, looked at it for a long time, then closed her laptop and went to make tea.

What stopped her was not ignorance. It was the rubric. It marked lesson plans by how many stages of teaching used AI: planning, delivery, assessment, remedial work. The highest marks went to plans with AI in every stage. It did not mark the thing she cared about most. That was the moment when a fourteen-year-old in Sagar looks out of the window and understands photosynthesis as what is happening, now, to that tree, and not only as a formula. She could write a plan that scored well by filling every stage with machines. Or she could write the plan she actually taught, in which the machine handled practice and tracking and she handled the rest, and score worse.

Sunita had faced the same choice a week earlier and filled every stage. She was not wrong to. She had a rubric and a deadline, and she said so: “It is a form, Kavitha. Fill it.” Kavitha could not. Filling in a form that you have seen to be a form is compliance. Accommodation begins when the form becomes your idea of the work.

Three days passed. Four. The delay felt like good sense: other modules first, the deadline still weeks away. But putting off a decision is also a decision, and it was the same one she had made in December. In 1970 the economist Albert Hirschman described two ways in which people answer an organisation that is failing them. One is exit: leaving it. The other is voice: telling someone who can change it what is wrong.¹² A government teacher in Sagar cannot easily leave. Her delay was neither. It was a quiet withdrawal, the nearest thing to exit open to her, and it would have cost her an entry in her appraisal and changed nothing.

Exit is not always the weaker answer. Sometimes the right response to an institution is to leave it, to refuse it, or to join with others to change it. Staying is not a virtue in itself.

What brought her back was a question that arrived while she was marking homework one evening: why does this bother me so much? The question did not make the bother go away. It moved her from inside it to beside it. From there she could see two things. The rubric was bad: it measured the number of machines, not whether students learned. And the rubric was not a judgement on her teaching. It was a form designed in an office with other priorities. Both were true. Neither required her to give up.

On the fifth day she wrote about photosynthesis. She wrote about the neem tree in the courtyard. She described what the platform did, which was practice and tracking, and what she did, which was the story and the question. She mapped the machine’s part honestly, as the support of the lesson and not its centre. The plan scored seventy-one out of a hundred. Enough to pass. The examiner wrote a comment she read three times: “Unusual approach. The narrative framing is not part of the rubric but the pedagogical logic is sound. Consider submitting to the state teacher innovation programme.” The comment was one examiner’s kindness inside a bad process. It did not show that the process worked.

Then she did one more thing. She wrote a one-page note to the district office about the rubric. It said, without anger, that a lesson plan with AI in every stage is not therefore a better lesson. It suggested that the rubric ask what students would understand at the end, and how the teacher would know. That was voice. She did not expect a reply, and by the end of March she had not had one. She could send it because her post is permanent. A guest teacher, hired for one year at a time, might reasonably have kept quiet, and that would say nothing about her character. A note that is ignored, or an objection that is overruled, does not end the choice. The same two answers remain: to take the voice higher, to someone with more power to change the thing, or to leave. Each costs something, and what a person can afford decides which is open to her. Making it safe to speak, and worth speaking, is the institution’s duty.

The Bhagavad Gītā has a word for what this part of the story needs. In its second chapter Kṛṣṇa tells Arjuna to act without attachment, “equable in success or ill-success”, and says that this evenness, samatva, is what yoga means. Two verses later comes a famous phrase: yogaḥ karmasu kauśalam. Kashinath Trimbak Telang translates it “devotion in (all) actions is wisdom”.¹³ The word kauśalam is closer to skill: the skilfulness of a practised hand. So the Gītā, which calls evenness yoga (2.48) and yoga skill in action (2.50), is willing to call its discipline a skill, where Aristotle keeps character apart from craft. The difference is one of emphasis more than a quarrel. Aristotle explains how a character is formed and lost; the Gītā asks how to act without letting success or failure govern the act. They meet on the point that matters here: neither judges an action wholly by its visible result. The verses are spoken on a battlefield, to a warrior who wants to put down his bow, and I borrow only their account of evenness, not their occasion.

“Equable in success or ill-success” does not mean indifferent. Kṛṣṇa does not tell Arjuna to stop caring how the battle goes; he tells him not to let the hope of winning or the fear of losing bend the act. Seventy-one was ill-success by the rubric’s measure. It was a result, not an account of what she had done. The evenness was in not letting that number decide what she taught, and not letting her contempt decide whether she finished. Resilience, in this sense, is not toughness, and it is not a duty to stay available to whatever harms you. It is coming back to the work without letting either the insult or the score take its place.

13.5 How the Four Hold Together, and How They Come Apart

The print edition said that the four multiply each other, and described them as “an ecology” that grows like a tree whose roots go deeper every season. Part of that is right. Each of the four goes wrong alone. Adaptability without learning drifts: you accept that things must change but cannot take in the new world fast enough to act. Learning without the ability to read people misses what the new knowledge means to those it lands on. Reading people without the ability to recover wears you out. And recovery without adaptability is only stubbornness, the strength to endure a situation you should be changing.

But there is a harder problem. None of the four is good in itself. Adaptability can make a person very good at serving a bad institution. Learning can make a propagandist more effective. Skill at reading people can be used to manipulate them. The ability to recover can keep someone in a place she should leave. Aristotle has a name for a power like this. In Book VI of the Ethics he calls it cleverness: “There is a faculty which is called cleverness; and this is such as to be able to do the things that tend towards the mark we have set before ourselves, and to hit it. Now if the mark be noble, the cleverness is laudable, but if the mark be bad, the cleverness is mere smartness.”¹⁴ The four are closer to cleverness than to virtue. Practice makes them stronger. It does not choose their mark.

What chooses the mark is judgement, the subject of Chapters 3 and 11. It is not a fifth item on the list. It is the work of deciding what a situation asks of the four, how much of each, and at whose expense. When should you adapt, and when refuse? What should you learn, and what doubt? Whom should you read, and how far should you trust the reading? When should you come back, and when leave? Aristotle held that the virtues need one another and are joined by practical wisdom: “with the presence of the one quality, practical wisdom, will be given all the virtues.”¹⁵ Not every reader will accept so tight a unity. The weaker claim is enough here. Without the others, and without judgement to aim them, each of the four slides towards one of its failures.

Nor does a good response promise a good outcome. Chapter 11 kept the two apart: a sound decision can lose, and a poor one can win. Kavitha’s plan was sound and scored seventy-one. Lakshmi’s request may have been right, and it bought a month. The world can defeat a reasonable act, and success does not prove that an act was wise.

And forever fails on Aristotle’s own terms. The same acts that build a disposition unbuild it. A teacher who stops questioning her own eye loses the habit as surely as a lyre-player who stops practising loses the lyre. These four do not grow by themselves while you sleep. They last as long as they are used, and as long as there is room to use them. They are used in small acts on ordinary days: a module read slowly because it disagrees with you, a question asked after school, a cut held for a month, a note sent that nobody answers.

The uncomfortable question

This chapter holds two things that pull against each other. A disposition is built by a person’s own acts and lost by her own neglect. Yet the occasions to use it are often set by others, and the chapter asks those others to carry the cost.

Aristotle held people responsible for their characters on the first ground. In Ross’s translation: “we are ourselves somehow partly responsible for our states of character”. Character, he says, is made by “activities exercised on particular objects”. His image is a thrown stone. Once let go, it cannot be recovered, “but yet it was in your power to throw it, since the moving principle was in you.”

Selvi’s case strains this. If her eye for a puckered seam has gone in five years, her own days at the packing table will have worn it away. She did not choose those days; the owner did. Kavitha’s case runs the other way. Had she let the rubric become her idea of teaching, nobody would have arranged that but her.

Most working lives lie between these two cases, and this chapter cannot draw the line. How much of a lost disposition was let go, and how much was taken, depends on what the person could have done instead, and at what cost. From outside, that is rarely plain.

13.6 A Way of Living, Not a Checklist

A person is not an operating system, and calling her one quietly turns her into the kind of thing this book is trying to see clearly.

The four cannot be practised one at a time, any more than honesty can be practised only on Tuesdays. They appear together, in particular situations, as one way of meeting what happens. The teacher who has them does not decide to adapt to a mandate. She notices it, feels what she feels, looks at the feeling, and then does the work, including the work of saying what is wrong with it. The supervisor who has them sees the silence on Selvi’s face because she has known Selvi for ten years. Seeing does not protect her from the cost of acting on it.

The print edition ended this section with a promise: “agency is the only durable advantage”. Before that, it had said that those who develop these four early will “disproportionately benefit”, and that whether forever is literally true “matters less than whether the direction is right. And the direction is clear.” The direction is not clear. The print edition pointed to the industrial revolution as its evidence. The record there is harder. In Britain, for sixty years up to 1840, output per worker rose by almost half while the real wages of working people barely moved. The rate of profit doubled: the gains went first to those who owned the machines.¹⁶ After the middle of the century, the same study found, real wages began to grow in line with output. The gains spread in the end, but only after more than half a century. That was a different age, and it is a warning, not an analogy. Productivity makes greater prosperity possible. It does not decide who receives it, or when. In this chapter the gains of the platform and the cameras went first to the funder and the owner. Mai adapted and has less secure work. Lakshmi read the room and lost standing. Kavitha recovered and scored seventy-one.

So why have them at all, if they give no advantage? For three reasons, which overlap. They help a person do what she is trying to do: Kavitha found Rohit’s pattern because she let a module disagree with her. They help other people: a teacher who reads her students well teaches them better, and a supervisor who reads her floor keeps it human. And some of their worth is in the having. Kavitha’s refusal to let a bad rubric define good teaching expressed what she believes teaching is, whether or not it helped her appraisal. Aristotle would say that acting from such a state is part of living well, not a means to it. That third reason is the one the market cannot take away, and also the one it will not pay for. These four do not give a lead that others cannot close. They are part of not being made smaller by change.

Then there is the half that is not the person’s. The language of adaptability can quietly move responsibility from institutions to individuals. The institution chooses the tool, sets the deadline, writes the measure and keeps most of the gain. The worker is told to adapt. If the change goes badly, the failure is put down to her lack of resilience, or of willingness to learn. But no disposition can make up for a change that is badly designed. A person needs time to learn, feedback that is honest, some say in the work, and a fair share of what the change produces. Where these are missing, the demand for adaptability becomes a way of making the individual carry the cost of a decision made elsewhere.

People also meet change with unequal freedom. The ability to refuse, retrain, speak or leave depends on income, security and what it costs to be wrong. What looks like flexibility in one person may be forced accommodation in another, and what looks like courage may partly be a safety net. A response should be judged only after asking what choices the person could afford.

So the chapter’s question has an answer, though not a comfortable one. Those who choose a change and take most of its gains should carry most of its cost. They should give hours for learning, feedback that is honest, and a share of what the change produces, and they should ask the people the change is done to. Nobody asked Kavitha whether the platform should come (Chapter 10), or what a good lesson plan looked like before the rubric was written. Nobody’s job was to read Lakshmi. Nobody kept the inspectors’ knowledge where the factory could still use it. A district that wants adaptable teachers has to build the conditions in which that disposition can be practised without exhausting the people who hold it. So does a factory that wants supervisors who see what the cameras miss. It also has to be able to learn from them, as the platform could have learned from Kavitha’s note about the order of its questions.

Ending: The Neem Tree in the Last Light

Kavitha finished the certification on Tuesday 24 March 2026, a week before the deadline. The email confirming it arrived while she was sitting in the staffroom, at the same window where she had read the first one fourteen weeks earlier.

She did not feel triumphant. Some of the modules had taught her nothing. One had taught her something she had not wanted to know, and Rohit was explaining one answer aloud each week because of it. The rubric was as bad as it had been in December. Her note about it had not been answered. She had saved the examiner’s comment. She had not submitted anything to the state programme. Not yet.

She had also learned something about herself that she had not wanted to learn: that under her contempt there had been a fear of being found ordinary. The fear had not gone. It had only been named, which made it a little smaller.

Outside, the neem tree caught the last of the afternoon light. She looked at it for a while, the tree that had stood in more of her lessons than she could count, and went home.

Practice

Two lists and one letter. Choose one change at work that you did not choose: a new tool, a new rule, a new measure. On one page, make two lists. In the first, write what in the change you should learn from, even if you dislike it. In the second, write what in it you should resist, and why, in words the person who made the rule could read without being insulted. Beside each list, write one piece of evidence that could make you change it. Then write one paragraph to the person who could change the part you resist. Decide whether sending it is safe and likely to be heard. If it is, send it, and keep a dated copy. If it is not, keep the paragraph, and write beside it what stops you, and whose hours the change is taking. You are practising three things at once: learning without surrendering, resisting without simply refusing, and finding a way to be heard. None of it will be certified. All of it gets easier by doing, and harder by not doing.

Words to keep

• accommodation: adapting to a goal or measure you have not examined, so that you serve it better without asking whether it is right.

• cleverness (deinotēs): in Aristotle, the power to hit whatever mark is set: laudable if the mark is good, “mere smartness” if it is bad.

• disposition (hexis): in Aristotle, a settled state of character, built by repeated acts and lost the same way.

• emotional labour: Hochschild’s term for managing your own feelings and face so that others feel what the job requires.

• exit and voice: Hirschman’s two answers to an organisation that is failing you: leaving it, or telling someone who can change it what is wrong.

• kauśalam: skilfulness; in Gītā 2.50, yoga is called skill in action.

• mean: in Aristotle, the right point between two failures, relative to the person and the situation.

• samatva: evenness; acting without being bent by the hope of success or the fear of failure (Gītā 2.48).

Sources

1. Aristotle, Nicomachean Ethics II.1 (1103a–b), trans. W. D. Ross (Oxford: Clarendon Press, 1908; The Internet Classics Archive).

2. Aristotle, Nicomachean Ethics II.4 (1105a–b), trans. Ross.

3. Aristotle, Nicomachean Ethics II.6 (1106b–1107a), trans. Ross.

4. NOAA Coral Reef Watch, 50 km satellite products (operational January 1997 to April 2020), succeeded on 30 April 2020 by the daily global 5 km products (coralreefwatch.noaa.gov).

5. David P. Ausubel, Educational Psychology: A Cognitive View (New York: Holt, Rinehart and Winston, 1968), p. vi.

6. Chāndogya Upaniṣad VII.1.3, trans. F. Max Müller, The Upanishads, Part I, Sacred Books of the East 1 (Oxford: Clarendon Press, 1879); see Chapter 5.

7. Tirukkuṟaḷ 703 and 706 (chapter 71, “Judging by Looks”), trans. V. V. S. Aiyar, The Kural or the Maxims of Tiruvalluvar (Madras, 1916; Standard Ebooks).

8. Lisa Feldman Barrett, Ralph Adolphs, Stacy Marsella, Aleix M. Martinez and Seth D. Pollak, “Emotional Expressions Reconsidered: Challenges to Inferring Emotion From Human Facial Movements,” Psychological Science in the Public Interest 20, no. 1 (2019): 1–68. The quotation is Barrett’s, from the Association for Psychological Science’s press release of 17 July 2019.

9. Regulation (EU) 2024/1689 (the Artificial Intelligence Act), Article 5(1)(f), applicable from 2 February 2025; Article 2(1) sets its reach (providers placing systems on the Union market, deployers in the Union, and providers and deployers elsewhere only where a system’s output is used in the Union); recital 44 gives the reasons (doubts about the scientific basis and reliability of such systems, and the imbalance of power in work and education).

10. Arlie Russell Hochschild, The Managed Heart: Commercialization of Human Feeling (Berkeley: University of California Press, 1983), p. 7.

11. Erik Brynjolfsson, Danielle Li and Lindsey Raymond, “Generative AI at Work,” Quarterly Journal of Economics 140, no. 2 (2025): 889–942. The 2023 working paper (NBER w31161) reported 14 and 34 per cent; the figures in the text are the published version’s. The published paper also reports evidence that the assistant helped workers learn.

12. Albert O. Hirschman, Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States (Cambridge, MA: Harvard University Press, 1970).

13. Bhagavad Gītā 2.48 and 2.50, trans. Kashinath Trimbak Telang, The Bhagavadgîtâ with the Sanatsugâtîya and the Anugîtâ, Sacred Books of the East 8 (Oxford: Clarendon Press, 1882), pp. 48–49. Telang renders yoga as “devotion”; the gloss of kauśalam as skill is mine.

14. Aristotle, Nicomachean Ethics VI.12 (1144a23–28), trans. Ross.

15. Aristotle, Nicomachean Ethics VI.13 (1144b30–1145a2), trans. Ross. On responsibility for character (for “The uncomfortable question”, 13.5): Nicomachean Ethics III.5 (1114a–b), trans. Ross (Internet Classics Archive, checked 4 November 2026): “we are ourselves somehow partly responsible for our states of character”; “it is activities exercised on particular objects that make the corresponding character”; the thrown stone, “the moving principle was in you”.

16. Robert C. Allen, “Engels’ Pause: Technical Change, Capital Accumulation, and Inequality in the British Industrial Revolution,” Explorations in Economic History 46, no. 4 (2009): 418–435: from 1780 to 1840, output per worker rose by 46 per cent and real wages by 12 per cent, while the rate of profit doubled. The abstract adds: “After the middle of the 19th century, real wages began to grow in line with productivity”.

Named people in this chapter are composites, and their conversations and circumstances are constructed; the studies, laws and figures cited are real and sourced.

Revised from the print edition (2026)

• What the print edition claimed: that four “meta-skills” “do not depreciate with time. They strengthen”, that they “compound forever”, and that “unlike every other capability discussed in this book” they cannot be tested; that emotional intelligence is “one of the irreducible human advantages” because understanding “requires a self that has lived”; that Rohit’s disengagement was “invisible to any instrument except a human being who is paying attention”, in a chapter whose sixth module teaches that very sign from the data; that the intervention “only consciousness makes possible”; that Linh’s new role was “the career the world needed” and that she was “the only person in Vietnam” who could fill it; that Lakshmi “was right to” override the scheduling system, after three sleepless nights; that people who develop these capacities early “disproportionately benefit” and “the direction is clear”, while admitting that whether “forever” is literally true “matters less”; that “agency is the only durable advantage”; and that Kavitha’s contempt was “a form of rigidity” and the gap between her values and the system’s “a gap she could bridge without betraying either side”. It also sent a March email with an end-of-year deadline fourteen weeks away (the school year in Madhya Pradesh ends in March), placed Sunita in Bhopal with twelve years’ service, made the certification’s penalty the loss of an increment, raced Kavitha through twelve modules in two weeks, named an inspector Revathi and the biologist Linh, and ended by asking “what will you build?”

• What I now hold: the four are dispositions, not skills or assets: built by doing and lost by not doing (Aristotle, II.1), sustained by conditions that let them be used, judged by the state of the person and not by a product (II.4), which is why no rubric can certify them; a long watch can give evidence of a disposition, but a certificate fixes a status on a date, and a disposition is kept only by use. They are not virtues on their own: like Aristotle’s cleverness (VI.12), each can serve a bad end, and judgement (Chapters 3 and 11) aims them; it is not a fifth item. The Gītā calls its discipline a skill in action (kauśalam) where Aristotle keeps character apart from craft; both refuse to judge an act wholly by its visible result. Adaptability is a mean between holding on and giving up, and is not accommodation to a goal you have not examined; filling in a form seen to be a form is compliance; adapting starts before the threat is plain, by testing the newcomer against your own evidence. Prior knowledge speeds what fits and slows what contradicts; learning is deepest where it shows a real limit; data shows where to ask, not the answer. Faces are evidence to interpret, not a window (the Kuṟaḷ and Barrett et al. answer different questions and pull against each other); reading people is mostly demanded of the less powerful (Hochschild), and nobody is assigned to read the reader. The kindness that kept the inspectors’ wages took their craft; a machine’s ability settles what can be automated, not everything that should be reorganised around it; the inspectors have not lost their eye, only the occasions to use it. AI help can teach as well as replace practice (Brynjolfsson et al.); what decides which is open. Recovery is coming back to the work without letting the insult or the score replace it, and includes voice (Hirschman); exit can be right, and voice depends on what a person can afford; a voice ignored or overruled can be taken higher, or its owner can leave, and making it safe to speak is the institution’s duty. The four promise nothing and give no lead others cannot close; they are worth having for what they do, for other people, and as part of living well. Those who choose a change and take its gains should carry most of its cost: hours, honest feedback, being asked, a share; and they should learn from the people they ask to adapt.

• What changed my mind: the objections that the four, called dispositions, were still being treated as virtues, and that the language of adaptability moves responsibility from institutions to the people the change is done to; the Prologue’s rule against defining people by what machines cannot yet do; Chapter 10’s finding that depth fills a position and does not grant one; Chapter 11’s finding that what compounds is conditional; Chapter 12’s question of whose hour it is; Aristotle’s sentences on the same causes and on cleverness; Barrett and colleagues’ review; the history of wages in the industrial revolution; the audit’s calendar and canon findings.

• What survives: the email at the staffroom window, and the class on chemical bonding; contempt, not fear (now with the fear underneath it); the delays, each reasonable; the chapter’s title, kept and argued with; the four, as dispositions; “Everything in the memo was true”; the funder’s choice; the sixth module; Rohit predicting the next question; Lakshmi, the women moved to packing, and the guilt “specific” and unplaced; the blank document and the tea; “why does this bother me so much?”; seventy-one out of a hundred, and the examiner’s comment; the neem tree in the last light.

• Section titles in the print edition: 13.1 was “Making Change Your Default Mode”; 13.2 was “The Speed of Conscious Absorption”; 13.4 was “The Skill of Conscious Recovery”; 13.5 was “Why These Skills Multiply Each Other”; 13.6 was “The AI-Proof Operating System: A Way of Living, Not a Checklist”.

Next in the series: Volume II: Application (overview).