AI-Proof: What Machines Can Never Be — Volume III: Transformation, Part VII: The Conscious Blueprint, Chapter 33: The Conscious Individual: A 12-Month Transformation
Part VII begins here. After the inner life of Part VI, what to do: this first chapter of Part VII follows one person’s year of practice, and asks who pays for it.
A platform lead read two months of her own notebook and found one word nine times: indispensable. No plan makes anyone AI-proof. A year of practice can show you the case you plead for yourself. It changes you only when others can check it.
In Brief For readers who fear becoming unnecessary at work, or who have built a career on being indispensable. Chapter 33 follows a platform lead in Bengaluru through the first half of a year of deliberate practice, from May to October 2026. She audits her own week, finds that her notebook has been pleading her case, gives away work she kept for the feeling of being needed, and starts new work that begins to make her needed again. Seneca’s nightly court of one and the Buddha’s mirror for his son Rāhula help ask whether a person can change herself on purpose when she is also the one keeping the account, and what a year can and cannot promise. It ends on a test the year has not yet passed: whether the work can go on without her.
A year of practice cannot make anyone AI-proof; it can show a person the case she has been pleading for her own necessity, and it changes something only when what she finds becomes an act that others can see and contradict, at a price someone must pay.
Opening: The Word in the Notebook
Priya’s pen stopped.
She was sitting on her bed in the flat in Koramangala, in Bengaluru, on a Sunday evening in June 2026. Her cat, Turing, lay across her ankles. Her mother was watching the news in the next room, and rain ran down the window. On her knees was the ruled notebook she had bought in April. On its first page she had written a question, What did I notice today?, and most of the entries tried to answer it. She was reading two months of entries back, looking for patterns. She had found one. The pen had stopped because it was not the pattern she wanted to find.
Priya Chandran has worked at a software company in Koramangala for nearly ten years. For most of them she tested its software before customers saw it, and in time she led the team that did it. Then her team built a platform that does most of that testing itself. Since October 2025 she has led the effort to carry it across every product the company makes. The post gives her a place in some of the rooms where products are planned. It gives her no right to decide what other teams’ products are for.
The early entries were tentative. A few lines about a conversation with a junior engineer. The tightness in her chest before a meeting that did not need her. The later entries were sharper. She could see herself in them now. Under pressure she took the work back into her own hands. She confused motion with progress. She judged her worth by what she produced rather than by what she understood.
But the pattern the pen had stopped on was none of these. It was subtler and worse. Across two months of entries, in almost every observation about her work, there was a word she had been using without noticing it. Indispensable. She had written it nine times. She had written around it dozens more. The platform I built. The decision only I could make. The judgement the team relies on. One entry, about the week in late May when she had changed how the platform graded a kind of bug, called the change the decision only she could make. The engineer who had found the bugs appeared in the same entry, in a subordinate clause.
Every entry was a small argument for why Priya Chandran could not be removed from the system. She had kept the notebook in order to notice. Without realising it, she had been using it to reassure herself that she still mattered.
She put the pen down. She picked it up again. She wrote: I have been lying to this notebook.
Part VII of this book is about what to do, and it should begin here, with an instrument of honesty turned into an instrument of comfort. Any plan for changing yourself has to be carried out by the self that needs changing. So the question is this. Can a person change herself on purpose, when the one who keeps the account is the one who wants to be found necessary?
The print edition answered with confidence. “The gap between knowing and doing is not a gap of information,” it said. “It is a gap of practice.” Then it offered a programme for the reader, whoever and wherever she was: a week of audit and four phases across a year. “The twelve months are long enough for a genuine transformation and short enough to sustain urgency.” This edition cannot promise that. A practice can show a person what she is doing. As of October 2026, machines do much of the showing, and they do it faster. What changes a person is an act that others can see and contradict. Much of what made her feel needed was never hers alone to change. And no year guarantees a transformation.
The chapter keeps its title, and the title needs an argument. “Conscious” can mean here only what it means in ordinary speech: knowing what you are doing. It makes no claim about consciousness, which this book does not lean on. “Individual” is half true. Priya’s year ran on her mother’s patience, her manager’s consent and a few weeks of a director’s attention. “A 12-Month Transformation” names what no practice can promise. The book’s present ends in October 2026, half way through her year, and the chapter stops there with her.
The same argument applies to the title of the Part this chapter opens, The Conscious Blueprint. A blueprint is a drawing made in advance for builders who did not draw it. Part VII has none to offer. It follows a person, a designer working alone, a young company and a large organisation, each deciding what to do, and each paying for the decision. Its question is plain. If no plan can make anyone AI-proof, what is worth doing on purpose, and who pays for it?
33.1 Week One: The Awareness Audit
First, see what is there. The audit asks you to watch yourself for seven days. You note at intervals what you are actually doing, and you report honestly on what you find.
Three questions. They are simple. The honest answers are not.
The first: what do I actually do all day? Not the job description. Not the account you give when someone asks. What fills the hours? The gap between what people believe their work consists of and what it consists of is wider than most of us admit until we measure it.
The second: of everything I do, what can a machine already do about as well? This asks you to look without flattery. Machines now write competent reports, emails and analyses in many professions. They write and test a good deal of code. Their work is often thin where it needs local knowledge, and it still needs checking. But “competent” is no longer in dispute, and competent is enough to raise the question.
The third question is where this edition parts from the print edition. The print asked: “what do I do that a machine cannot, and how much of my time do I spend doing it?” That question defines a person by what machines cannot yet do, and next year’s answer will be smaller. So the third question changes. Of what is left, what would someone have to explain and put right if it went wrong, and is that someone me?
Priya ran the audit in the first week of May. Every hour she wrote one line in a document on her laptop: what she was doing, and for whom. She expected the week to confirm what she believed. She believed that most of her work needed her.
It did not confirm it. Most of her hours went to status meetings where her part was to agree. Some went to reviews of work she had already shaped at the design stage. Others went to planning that followed templates she had written months before, templates any careful colleague could run. At the end of the week she gave her work notes to a model and asked it to sort the hours. It took seconds to do what had taken her an evening, and it sorted better than she had. It put two meetings she had called decisions among the meetings that decided nothing. It sorted by one measure, whether a decision was made. A meeting that decides nothing may still pass on news, keep trust, or let someone raise a worry she would not put in writing. The model was right about decisions and silent about the rest.
The third question was harder. Some things were hers to answer for. What the platform taught the teams who used it was her design, and if it taught them something wrong, she was the one who would be asked why. But answering for the platform took few of her hours. The audit did not tell her what to do. It showed her what was there.
The words “what is left” hide a trap. They make answering sound like the work that remains once the machines have taken theirs. It is not. The person who chose a measure, set a target or approved a model may answer for it after doing very little of the work. The person who does the hardest task may have no power to change it. Answering follows authority, knowledge and the power to repair, not hours. So the third question has a second half, about the chain. It asks who decides what the work is for, who is touched by it and who can put it right, and where she stands among them.
Then came the Sunday in June, and the word. She had kept the notebook on paper, out of any machine’s reach, by a choice of her own. Finding the word was the easy part. The hard part was the sentence she wrote under it. To notice is not yet to own. Owning is an act. She took the finding as hers when she wrote that she had been lying.
A model that sorts a week’s hours in seconds makes the notebook look old-fashioned. As the previous chapter said, a model given your calendar, your messages and your drafts will describe your habits back to you, including some you had not seen. On this view self-knowledge is now a report you can order, and the slow evening with a pen is a hobby.
Much of that is true, and the chapter does not fight it. A machine’s account of your week is a good witness. It is still testimony about you, offered to you. You decide whether to believe it, as far as you can check it. You decide what it means, and what you will do because of it. A report that says “you attend meetings that decide nothing” changes nothing until someone stops attending them, and the someone has to be you. Keeping the notebook from machines cost her two months. It also meant that what she found, she found herself.
Seneca, the Roman Stoic of the first century CE, described a practice like hers. In his book On Anger he recalls the philosopher Sextius. In Aubrey Stewart’s translation1: “It was the custom of Sextius when the day was over, and he had betaken himself to rest, to inquire of his spirit”. Sextius asked: “What bad habit of yours have you cured to-day? what vice have you checked? in what respect are you better?” Seneca adopted the custom. “I make use of this privilege, and daily plead my cause before myself”. His Latin is apud me causam dico: I plead my case at home, before myself. Causam dicere is what a defendant does in a court of law. Each night, “when the lamp is taken out of my sight, and my wife, who knows my habit, has ceased to talk,” he reviews the day. “I conceal nothing from myself, and omit nothing”. Then he passes judgement: “I pardon you this time: see that you never do that anymore”.
Seneca was writing about anger, not careers, and I borrow only the shape of his court. He was not naive about it. He convicts himself too: “In that dispute you spoke too contentiously”. But his court has one member. The judge, the accused and the advocate are the same man, and the verdict is a pardon with a warning. Priya’s notebook had been that court, with the advocate doing most of the talking. The phrase “court of one” is mine; the court is in his Latin. Nor is his pardon an acquittal. It comes with a warning and another review the next night, a correction measured out rather than a case dismissed. Private review finds what nobody else sees. Its weakness is that nobody can appeal against its verdict.
There is reason to think such a court more lenient than it knows. People rate themselves as less prone to bias than others are, as psychologists reported in 2002 and a replication confirmed in 2021.2 Those studies measured how people rate their own bias, in surveys and a laboratory test, not how they judge their own work. On my reading, they still make a court of one the court most likely to acquit.
Not all of the word was Priya’s. For years her annual reviews had praised her as irreplaceable, and she had built a career on deserving the praise. An employer that rewards people for being impossible to replace teaches them to arrange their work so that they are. It also makes them a risk the employer carries, and it rarely counts that risk. Firms call it key-person risk: knowledge kept in one head, approvals that wait for one name, nobody ready to step in. Colleagues meet it as turns they never get. The praise and the risk are one arrangement, seen from two desks.
Under the word lay a confusion this book has to watch in itself. Being the one who answers for something is not the same as being the only one who could. The platform was hers to answer for. Yet the platform had also shown that she was not the only one who could do its work: it did that work every night, without her. Her greatest achievement was evidence for both truths at once. The fear the entries hid was not that a machine would replace her. It was that she had already replaced herself with a story about being irreplaceable. The story did what a fluent machine answer can do. It sounded true, and nobody had checked it.
This book’s claim is that what stays scarce is someone who answers. A reader can turn that claim into armour as easily as Priya turned her notebook into one: I answer for this, so nobody else could. The second half does not follow. A person’s worth does not depend on being needed. And the one who answers is not a remainder that nobody else could supply. Other people, and the institutions they work through, can do most of what her answering does. What they cannot do is be her. That was never a reason to keep anyone in a job.
33.2 Seeing Clearly Where You Stand and What Is at Risk
The print edition spread four phases over twelve months: two months of seeing, three of building, four of creating and three of distributing. In a working year they overlap. Priya’s ran in five months and not quite in order. So this edition keeps the phases as kinds of work, not as a calendar, and its section titles drop the months.
The audit takes a week. The plan’s first phase takes two months. Its work is observation, not change. This feels backwards, and it is where many people stop. They stop because it is boring, and boredom is harder to bear than difficulty.
The plans sold for self-improvement have answers for difficulty, for pain and for fear. They have no answer for boredom. You ask the same questions day after day, and the answers begin to repeat. The novelty of examining yourself fades. What remains is the practice, without excitement.
What repetition gives is what a single week cannot. The meeting that seems essential on one Monday reveals itself, over four Mondays, as a ritual that produces no decisions. The work you do on a quiet Thursday afternoon turns out to be the work that matters most. For Priya, May and June were these months. She found three kinds of meeting in her diary. She went to some because she was asked. She went to others because nobody had thought to stop asking her. And she went to a third kind because she wanted to be in the room when anything was decided. She had not known the third kind existed until she counted.
In the plan’s second month you turn outward and study your own field. What can the best systems do in your work now, and what might they do in two years? Here the print edition drew a line it could not keep. It said the real boundaries were “structural”, rooted in the absence of consciousness. It granted that the human qualities might “someday be approximated”, and then took the grant back: a machine still lacked “a conscious being who bears the weight of being wrong”. “That weight is yours. It is also your moat.” Both sentences are withdrawn. The first rests the argument on a question nobody can settle. The second promises an advantage the book cannot give.
What can be said is dated. By mid-2026, Priya’s own testing skill was largely the platform’s. The tools her company’s developers used wrote code, tested it and fixed much of what they found. In two years they would probably do more. That is my projection, and nobody could say how much more.
Her second month gave this a concrete shape. In June she asked developers on two teams to show her how they now worked. She learned something her own dashboards had not told her. The code tools wrote much of the code, and they also wrote many of the tests that checked it. Her platform then tested that code, and ranked its findings partly by those tests. Much of the time, one machine’s work was being checked against another machine’s idea of what the work should do. When one specification shapes both the code and the tests, their errors do not cancel out. They agree with each other, because they share the same blind spot. Neither the tools nor the platform was wrong to do this, and the arrangement caught more defects than her old team ever had. But it moved the place where a person was needed. The question was no longer whether the code passed. It was whether the tests asked about the right things, and who had decided what they asked.
What carries no date is not a skill. It is a position. A decision belongs to someone. Someone must explain it and put right what it costs. And it touches someone’s life. Those three can be held by different people in a chain, and an institution gives or withholds each of them. Knowing the work well lets a person fill such a position. It does not give her one.
So what was at risk for Priya was not a skill, and for now it was not her post. Her employer kept her and paid her while the testing passed to the platform, which is more than many employers do. What was at risk was a story. The risk also ran the other way, towards other people. A test engineer on her old team had never run a planning cycle, because she always ran it. Approvals waited for her because she had arranged that they should. The story cost her nothing. It cost them their turns.
For some employees, being needed is more than a story. Being the one who knows a system can be the only protection a person has against being let go. Someone whose income rests on it is not wrong to weigh that. Priya could give work away partly because her employer had already kept her. The cost to others is still paid, by whoever waits for a turn. People want to be needed for different reasons: to keep an income, to be recognised, to keep control, to belong, or because being needed tells them who they are. These are not equal. A tester guarding her livelihood is not a senior manager guarding his status. An employer can praise the irreplaceable, reward kept knowledge, and then call the person who keeps it an obstacle. The line that matters lies elsewhere. Being valued because your work enlarges what others can do is one thing. Being valued because they cannot do it without you is another.
None of this makes a worker disposable. Making knowledge easy to pass on is one thing. Making a post easy to fill is another. Treating the person who built the knowledge as easy to drop is a third. The first makes a firm sturdier. The third is a choice the firm makes about people, and it is not made right by the first. As Chapter 29 argued, losing a role is not losing one’s worth. What is passed on also becomes the receiver’s to change, or to set aside where it does not serve her. A handover that forbids that has handed over a copy, not a skill.
Even seeing had a price. The hourly line in her document took a minute an hour of her employer’s time, and the notebook took her evenings. For an employee whose hours are counted, even an audit of one’s own week is something to ask for.
33.3 Building, and Who Is at the Controls
In the plan, building means bringing machines into the work you already do. This is where people fall back into old patterns. They see the map, treat it as a problem, and optimise. Priya’s version came in July.
Each month the platform was retrained on the month’s test results. What it learned decided what every product team would be shown the month after. Since 2025 that had been hers alone to settle. After the June entry she felt an urge to add a step. No retraining would go ahead without her written approval. She wrote the urge down and named it: vanity dressed as diligence. Then she looked again. The gap was real. One person was deciding alone, each month, what a platform taught every team in the company, and nobody had ever asked who checked her. The vanity lay in the name she had written in the box. The step could stay.
An old teaching gave her the order she needed. In the Majjhima Nikāya, the middle-length discourses of the Pāli canon, the Buddha instructs his son Rāhula. According to the commentary, Rāhula was seven years old. The discourse begins with lying. The Buddha shows the boy a water dipper with a little water left in it. In Ṭhānissaro Bhikkhu’s translation3: “That’s how little of a contemplative there is in anyone who feels no shame at telling a deliberate lie.” Priya’s lie had not been deliberate. That kind is harder to catch.
Then the Buddha asks: “What do you think, Rahula: What is a mirror for?” “For reflection, sir.” The Pāli word is paccavekkhaṇa, looking back over something. “In the same way, Rahula, bodily actions, verbal actions, & mental actions are to be done with repeated reflection.” The boy is to look before he acts, while he acts and after he acts. He asks each time whether the act would harm himself, others or both. When a deed or a word has caused harm, he is not to settle it alone: “then you should confess it, reveal it, lay it open to the Teacher or to a knowledgeable companion in the holy life.” After that, “you should exercise restraint in the future.” A harmful act of the mind is treated differently. For that, “you should feel distressed, ashamed, & disgusted with it”, and then restrain yourself. Thoughts are not confessed, but they are not let off either: a harmful one is met with shame and restraint. The difference lies in how each is corrected, not in whether the mind matters.
The discourse is training for a novice in acts that harm, not advice about careers, and I borrow only its order: look, lay the act open, restrain. Its last distinction is the useful one. It did not ask Priya to show anyone her notebook. What she had thought was hers to be ashamed of. What she had done because of it was to be laid open.
The two lamps agree on much. Both make the review daily. Both demand that nothing be hidden: Seneca conceals nothing from himself, and the boy holds his acts up to a mirror. They also answer partly different questions. Seneca asks how a man may improve his temper. The discourse begins with lying and asks how a boy may make every act of body, speech and mind a reflected one. Behind them lie different pictures of the self and of the goal, a better man by Stoic lights and release from suffering on the Buddha’s path. I borrow from each without joining them. Where they part is in who hears the account. Seneca keeps the court private, and its verdict is a pardon with a warning. The Buddha sends harmful acts out to another person, and asks for restraint with no word of pardon. Each has its weakness. A court of one tends to acquit. The discourse’s hearer is “the Teacher”, and at work the obvious hearer is the person who writes your appraisal. Laying a fault open to her is not always safe. In the discourse, teacher and companion share the boy’s aim. At work, the person who hears a fault may also rank you, and a hearer is worth trusting only if she is fit to hear it. Such a person is often outside the line of appraisal: a peer in another team, a former colleague, a mentor who gains nothing from your fall. Where a fault has harmed others, it must still reach whoever can repair it. A safe hearer helps a person decide how to say it, not whether. So the discourse leaves Priya a choice it did not leave Rāhula: what to show, and to whom. This book has said before that a page only its writer reads is still a ritual, however honest. The discourse says which part must leave the page: the act, while the thought may stay.
So the notebook stayed in her drawer, and two acts went out. The notebook had found the plea. It could not end it, because only acts can be seen and contradicted.
The first went to Deepak, the company’s director of platform engineering. She proposed that each month’s retraining be approved in writing by one of four team leads in turn, with a short record of what had changed and why. Her name was one of the four. The case for spreading decisions so that no single loss can break a team belongs to another chapter. Priya’s reason was the notebook. Deepak took three weeks to agree. The mobile team’s lead called it another meeting. Each month’s approval now took two days longer than her own had. A rotation has a danger of its own. Four names in turn can become no name at all, if a drift begun in one month is passed by the next. The written record was meant to prevent that, since each lead could read what the months before had changed. Priya still answered for the platform’s design. What rotated was the monthly yes.
The second went to Rashmi, her manager. Priya handed the quarterly planning cycle, which she had always run from her own templates, to the test engineer who had never run one. She gave him the templates and stayed out of his first meeting. She also stopped going to two weekly status meetings where her part had been to agree. Rashmi said yes, and warned her that Deepak read attendance as commitment. In August he asked twice where she had been. The engineer’s first cycle came in two days late and missed one dependency (work that other work had to wait for), which a product team found and fixed. It was the kind of miss she would not have made. It was also a turn he had never been given.
It was also a risk she had handed him. She had given him her templates, not the judgement behind them, and by staying out of his first meeting she chose trust over help. The product team that found the missed dependency paid for the lesson with its own hours. A turn given without support is a test, not a lesson. Handing work on well has conditions. The person must be allowed to decide, know what he needs to know, and have time and someone safe to ask. He must be judged in proportion when he errs, get the credit, and in time be free to change the work. Some work should not be handed on lightly at all: a legal duty, a confidence, a decision that needs one named person over time.
In the plan’s building months a good question is this: of the tasks I could hand to a machine, which ones am I keeping because they develop my judgement? A lawyer who hands over all her research saves time and stops building the feel for argument that slow reading of cases gives. Someone who drives to the gym and someone who walks both arrive. Only one has exercised on the way. Some work must be kept because the work is how people learn. Priya had a habit of her own here. Since the spring of 2025 she had formed her own view of a problem before she asked the machine. But the audit added a harder finding. Some tasks she kept for her judgement’s sake. Others she kept for the feeling of being needed. And from the inside she could not always tell which was which. The planning cycle had been both. It had trained her. Someone else needed it more.
This is the missing rung of Chapter 4 seen from inside one team, where a senior, not a machine, had kept the work a newer colleague would learn from. If machines do the work through which people used to learn, someone has to pay for the learning on purpose. That means tasks kept for practice, machine work reviewed by juniors and time allowed for mistakes. A first planning cycle that comes in late is a cost. It is also the price of a future planner. Not every task should go to a beginner for the lesson’s sake, and where the risk lands on customers the lesson must be supervised.
Not every act is equally easy to undo. A meeting Priya stopped attending could be rejoined within a week. A plan that comes in late lands on other teams, and the hours they lose cannot be given back. Chapter 24’s rule fits these experiments too: fast where an act can still be undone, slow where it cannot. Be bold with your own habits, and slower where the experiment lands on others.
33.4 Creating New Work Without Becoming the Bottleneck
Creating asks a harder question. Given what you now know, what new work should you be doing that you are not doing now? In the plan it starts in the sixth month. Priya’s started in August, the fourth month of her year. Work does not wait for a plan’s calendar.
Lavanya is a product manager at Priya’s company. In August her team was building a new feature for the purchase requests that customers raise through the company’s software. The code tools had written three versions of it. All three passed the automated tests. Lavanya felt that two of them were wrong and could not say why. She asked Priya to look, partly because nobody else in the company had read the customers’ complaints for as long.
Priya read the three versions over an hour. The first had a small fault that the tests did not cover. She thought the second was right. The third was better on every measure the tests could see. It was faster and simpler. But it assumed a user who would finish a request in one sitting, from start to finish, without a break.
Many of the company’s customers are mid-sized Indian firms, and the people who raise their purchase requests do not work that way. A request is started, then stops while someone waits for a department head to approve it. It sits half finished for two days. Then it is finished on a different device, by someone who expects the software to remember where she was. The third version would have lost her place. The fault lay in the picture of the user that the tools had been given, and the code was sound. Someone had decided what counted as success for this feature, and the tests measured only that.
That evening Priya wrote in her notebook: I just did something no AI did. Not because I am smarter. Because I know these users.
Two days later she tested the first sentence. She gave a model a year of support tickets from those firms, with the names removed, and the records of how their staff moved through the software. She asked what the users did that the feature assumed they would not do. In a few minutes it found the pattern: the half-finished requests, the wait, the second device. A model can now find such a pattern when someone gives it the records. The code tools had been given the specification and the tests, not the tickets. Nobody had asked them.
So the first sentence of her entry was false by the end of the week, and the second was true in a way she had not meant. She did know these users. Part of what she knew was a record nobody had kept: what years of complaints, read by one person, had taught her, stored in one head. A record that is kept can be handed to the next person, or to a machine. Much of the difference between her and the machine was access, not intelligence. The model had not lacked the wit to see the users. It had lacked the records, and nobody had made them part of its task.
The records had limits of their own. Tickets come from people who met a problem, noticed it and could report it. Users who gave up quietly, found a way round or could not report wrote none, as the screen-reader bugs of Chapter 32 had shown. The movement records showed what people did, not why. And a model that finds a pattern in records chosen to contain it has analysed the evidence, not confirmed it. Checking it meant watching how people actually worked, and asking some who had never complained.
Lavanya chose the second version, slower and less simple, and the third version’s speed was lost. Priya had said what she could show and what she could not prove. The choice, and its explanation to her team, were Lavanya’s. Whether it was right would show only after the feature had been in customers’ hands for months. A good result then would not prove the choice had been right, any more than a bad one would prove it wrong. So the decision did not end in the meeting. Lavanya asked for a monthly count of requests left half finished and finished on another device. If the count told against her choice, changing it was hers.
From then on Priya offered product teams an hour on the assumptions their features made about people: about patience, interruptions, approvals and devices. She found the hours in the meetings she had stopped going to. Within a month three teams had asked for her. She noticed the old word coming back in a new form. She was becoming the reviewer every team needed. The hours the audit had freed were filling with work that made her necessary again.
So she wrote her questions down on one page. Anyone on a product team could ask them. A model could be asked them too, if someone gave it the records. If a machine comes to do the looking, that is to be welcomed. The check moves. What stays someone’s is the decision about what a feature assumes, and the duty to explain it when the assumption fails.
A page of questions is not yet a skill shared. Chapter 30 split such work into three: taking knowledge out of a head, keeping it, and making someone else able to use it well. The page did the first two. The third would show only when someone used the questions without her, found what she had missed and could say why. It would show again when that person taught someone else. If every team still waited for Priya to read the answers, she had moved the bottleneck, not removed it.
Writing it down was safe for Priya for a reason that was not hers. Her employer had kept her, and the company was about to give her work a trial. Arvind, in Chapter 30, checked a model’s account of his own code in his evenings and became more exposed. Where a firm cuts the people whose knowledge it has collected, asking them to write it down is asking them to train their replacements. Writing down what you know is a fair request only where the institution rewards the person who shares it. Before writing it down, an employee may fairly ask why it is wanted, whether the hours count as work, and whether it will be used to judge her.
33.5 Distributing, and Being Known for What You Answer For
Becoming known for what only you can do is a promise this edition withdraws. Little work stays something only one person can do.
The print edition’s advice for this phase was blunt. “It is not enough to be good. You must be known for the specific things that make you irreplaceable.” It is the notebook’s word made into advice. The print told every reader to do on purpose what Priya had caught herself doing by accident. It is withdrawn here.
What an employee can keep is smaller. She keeps her name on her work, the explanation she gives when it fails, and the habit of putting right what she can. Being chosen is someone else’s act, and it often runs along old lines, towards people who are already known. So distribution, in this edition, means something plainer than being known. It means making your work visible so that others can check it, use it and dispute it. A claim about your own work is testimony. It deserves belief as far as others can check it, and as far as someone stands behind it. The aim is to be known for what you will answer for, not for what nobody else can do.
At the end of September, the fifth month of her year, Priya wrote three pages for the company’s quarterly product review. They described the reviews and what they had caught, with examples from which the customers’ names were removed. They also described the one assumption the reviews had missed, which a customer found first. They proposed a standard step for two teams. Experienced members of each team would ask the written questions, and the tools would be given the tickets and records they had lacked. One sentence said that the step should not need her.
She presented it in five minutes, without slides. Deepak was sceptical: it was another gate before every release. A head of product wanted numbers she did not yet have. The chief product officer, the CPO, asked one question: “Are we saying that the AI writes the code correctly but for the wrong person?”
“Sometimes,” Priya said. “And sometimes for a person we never told it about.”
The CPO, who decides with the teams what the company’s products are for, approved a trial for one quarter on two teams. Deepak’s condition was attached: the step could add no more than a day to any release. The mobile team, which shipped a new version every week, declined to join. That was its right: a step one team finds useful is not owed by another. The CPO asked Priya to lead the trial, and she said yes. Her colleagues began to call the work “human systems analysis”. It was a name for a piece of work, not a rank. It gave her a seat in two teams’ release meetings and no right to decide what their features were for.
The price was easy to count and easy to forget. Each review spent a day of a product team’s time, and the trial’s hours came out of Priya’s post. So did the hour a week for her study of volunteers tested without the platform, which the CPO had approved in April for a year. If the step slowed releases by more than a day, it would end, and it should. The CPO’s approval proved nothing yet about whether the reviews were worth their cost. It bought a quarter in which to find out.
Deepak’s day was itself a decision, and nobody had said what it rested on. A limit protects releases from a review that grows without end. Applied to every case, it also means that no finding can hold a release back for long. A check that can find a harm but never delay a release is a report, not a safeguard. The trial would have to say what happens when a review finds something that needs more than a day.
Permission let Priya try. Whether the change lasted would depend on what the company rewarded. For years its annual reviews had praised her as irreplaceable. If the next round still praised the one person every team needed, the trial would teach the old lesson in a new room. Credit matters for the same reason. If the engineer who now ran the planning, or someone who improved the questions, got no credit, sharing would look like a loss to the next person asked to share.
And being known has a price of its own. The more teams knew her for the reviews, the more the reviews became hers, in everyone’s mind. Hers too.
33.6 The Practice That Continues After the Year Ends
The twelve months are a container. The practice is not.
A plan of months is a scaffold. It is useful because change takes time, and time without any structure drifts. It also comes down. What a scaffold leaves behind is not a set of habits that will hold by themselves. An earlier chapter quoted Patañjali’s Yoga Sūtra, a manual for stilling the mind, on practice (abhyāsa). Practice becomes firm when kept up for a long time, without a break and with care.4 The verse names no year, and applying it to a year at work is my reading, not the text’s. Twelve months may not be a long time. A habit that doing built, not doing can undo.
So the chapter’s question can now be met, though not with the comfort the print edition offered. Can a person change herself on purpose, when she is also the one keeping the account? Not by deciding to. Priya decided in June to stop pleading, and by August the plea had found a new word. The claim is narrow. A habit of hand or body may change in private, by practice kept up for a long time. A plea is different, because the one who judges it is the one making it. What she could do on purpose was smaller. She could change what she did: give away the approvals, hand over the planning, stop going to meetings that did not need her, write her questions down. She could lay those acts open to people who could push back. Deepak could refuse, Rashmi could warn her, a product team could find the dependency her successor missed, and the CPO could end the trial. Each of them was a check her notebook could not be. Whether the acts changed the person who did them is a different question. It shows only over years, and perhaps only to others.
The glitch
Telling someone what you mean to do can make you do less of it. In 2009 the psychologist Peter Gollwitzer and three colleagues reported four small studies of students at German universities who hoped to become psychologists or lawyers. When another person took note of a student’s plan to work towards that goal, the student then acted on it less than students whose plans had been ignored. In the last study, students whose intentions had been noticed already felt more like the lawyers they hoped to become. The authors’ explanation is that being noticed gives “a premature sense of possessing the aspired-to identity”. The effect appeared only among students strongly committed to the goal.5
The samples were small, from 24 to 57 people, and the first study rested on what students reported about their own week. None of it was about work. If the finding holds beyond students, it bears on the Practice below, which asks you to tell one person what you decided. On my reading, the better order is to begin, and only then to tell that person what you did and ask them to check it later. A plan laid open can stand in for the act it plans. The study’s warning falls hardest on people most committed to who they hope to be.
Change has more than one measure, and the measures need not move together. A person may see a habit and still repeat it. She may change the habit without changing why she formed it. Others may go on relying on the old arrangement after she has changed. A company may adopt a better process while its author stays as hungry for credit as before. Priya had seen something about herself. She had also done things others could see: given work away, let another person decide, put her questions where others could use them. Whether she had become less afraid of being unnecessary was harder to know. The first changes were real. They did not prove the last.
One test came early in October. The test engineer ran his second planning cycle on time, and he changed one of her templates. She thought the change made the plan worse. She said so once, he kept it, and she let it stand. Whether she let it stand out of restraint, or because the trial left her no time, she could not have said. The harder test is not whether a person can give work away. It is whether she can watch someone else do it differently, or better, and let that person take the credit and keep the decision.
That is not a counsel of despair. It moves the work from the inside to the outside, where it can be seen. A person may not be able to make herself stop wanting to be needed. She can arrange her work so that the wanting does less harm: so that other people get their turns, so that decisions have more than one name on them, and so that what she knows is written down where others can use it. The wanting may fade with the arrangements, or it may not. The arrangements are still worth making, and they are what she can answer for. The opposite of being indispensable is not being useless. It is contributing in a way that others can carry on. By October 2026 the five bugs found in late May, which had stopped screen readers from reading payments and logins aloud, had been fixed in later releases. The customers who had met them were never told, and no arrangement for the future goes back to the people a harm has already reached.
A practice that watches the self also wears out in a particular way. Once you know which word you are watching for, its absence stops being a sign. A measure that becomes a target stops measuring well, as Goodhart’s law has it.5 Priya could stop writing “indispensable” and keep everything the word had meant. What guards against that is a second hearer, for acts rather than thoughts. A stricter court of one would only acquit more carefully. Goodhart’s law, which Chapter 9 met in a dashboard, is a warning about incentives, not a proof that every self-measure fails. This one fails because its writer knows it. The old habit may also come back, under a missed deadline, a junior’s serious mistake or a senior’s doubt about her worth. Its return would not prove the change false. What matters is whether she sees it, stops it and repairs what it did.
The practice also has a price, and the print never said who pays it. Priya’s year ran on her own evenings, her mother’s patience, her manager’s consent, a director’s three weeks and a CPO’s quarter. For the reader the print addressed, “whoever and wherever they are”, most of those were assumed. An employee whose firm counts output cannot choose which tasks to keep for her judgement’s sake. Hours saved by a machine usually belong to the employer, not to the person who saved them. A plan that asks you to spend hours you do not own is a request. Make it to the person who owns them, and say what it will cost. A change that runs on people’s evenings is paid for by them. The junior engineer’s evening in May, and Arvind’s evenings in March, paid for things the company kept. A company that relies on such evenings has not paid for its own improvement. And nobody owes the world a transformation. A person near retirement, or carrying care at home, may keep her work as it is. Her worth does not wait on her changing.
The print closed this section with a promise in the shape of a paradox: “You do not arrive at AI-proof. You practise it.” Neither half holds. No practice makes a person proof against machines, so there is nowhere to arrive, and nothing called AI-proof to practise. What a person can practise is smaller and more honest. She can find the case she has been pleading, give away what she kept only to be needed, and lay open what she did. She can answer for the work that is hers. That promises no job. The book promises authorship, not employment, and even authorship has to be done again each day.
Ending: The Sentence Left Open
On Sunday 11 October 2026, five months after the audit, Priya sat on the same bed with the same notebook. Her mother had gone to bed early. Turing was asleep on the window ledge.
She read back from June. “Indispensable” had not appeared since the night she wrote under it. She counted other words instead. My reviews, eleven times. The trial, fourteen. Needed, three times, each in a sentence about someone else.
She wrote: Have I changed, or have I only learned which word to leave out?
She began an answer. No. I gave the planning away, and the approvals, and I wrote the questions down so that
She stopped. The sentence was pleading her case again, and she could hear it. She left it where it had stopped, without a full stop, and closed the notebook. The trial had eleven weeks to run.
Practice
The word you keep using. Take a month of what you have written about your own work: notes, messages, the review of yourself you sent at appraisal time. Find a word or phrase that keeps returning about your place in it, or a claim about your role that keeps coming back. The most frequent word is not always the most telling. Let a machine count it if you like; as of October 2026, given typed text, it will be faster than you and at least as accurate. Give it only what is yours to share: not your employer’s confidential papers, customers’ records or other people’s messages without permission. Beside the word, write three lines. First, what it would cost you if the word stopped being true. Second, which part of what it protects is fair to keep; wanting to keep an income is not vanity. Third, who else pays while it is true: the colleague who cannot take leave because only you know the system, or the junior who never gets the task you keep. Then choose one thing you do partly because it makes you needed, and decide what you will do differently about it. Tell one person what you decided, not what you wrote. Choose someone who can tell you that you are wrong and has no power to punish you for it. If nobody at work is safe, a friend in the same trade or a former colleague can serve. This takes an hour of your own time, unless someone gives you theirs. It will not protect your job, and it may show you that you have been protecting the job rather than doing it. It shows you the case you have been making for yourself.
Words to keep
- abhyāsa: Sanskrit, practice; in the Yoga Sūtra, effort kept up for a long time, without a break and with care.
- awareness audit: a week of noting, at intervals, what you are actually doing, followed by three questions about it.
- causam dicere: Latin, to plead a case, as a defendant does in court; Seneca’s phrase for his nightly review of himself.
- delegation: handing work to someone else; done well, it carries the right to decide, support, credit and the freedom to change the work.
- Goodhart’s law: “When a measure becomes a target, it ceases to be a good measure” (Strathern’s wording).
- key-person risk: the risk an organisation carries when knowledge, approvals or decisions depend on one person.
- Majjhima Nikāya: Pāli, the middle-length discourses of the Buddha.
- missing rung: Chapter 4’s name for the loss of the entry-level work through which people used to learn a trade and become able to check the machine.
- paccavekkhaṇa: Pāli, reflection, looking back over an act; in the discourse to Rāhula, the purpose of a mirror.
- specification: the written description of what a piece of software must do, from which code is written and tested.
Sources
- Seneca, De Ira (On Anger) III.36.1–4, trans. Aubrey Stewart, “Of Anger”, in L. Annaeus Seneca, Minor Dialogues, Together with the Dialogue On Clemency (London: George Bell and Sons, 1889), as transcribed in Wikisource (“Of Anger/Book III”, from the 1900 printing). Latin from The Latin Library (sen.ira3), III.36.3: Vtor hac potestate et cotidie apud me causam dico; nihil mihi ipse abscondo, nihil transeo. Stewart’s punctuation after “anymore” is a question mark inside the quotation; the quotation above stops before it.
- Subramanya Prasad Chandrashekar et al., “Agency and self-other asymmetries in perceived bias and shortcomings: Replications of the Bias Blind Spot and link to free will beliefs,” Judgment and Decision Making 16, no. 6 (2021): 1392–1412 (“Participants rated themselves as less susceptible to biases than others”), replicating Emily Pronin, Daniel Y. Lin and Lee Ross, “The Bias Blind Spot: Perceptions of Bias in Self Versus Others,” Personality and Social Psychology Bulletin 28, no. 3 (2002): 369–381, which Chapter 32 reports. Both are studies of self-ratings, not of how well people judge themselves in their work. Key-person risk: Financial Conduct Authority, “Recruiting key staff and managing the risk of leaving: credit unions” (first published 12 May 2015, updated 10 February 2023): “Depending on key people too heavily can cause problems, as they may leave or be unable to carry out their duties for a while. This is called ‘key-person risk’.”
- Ambalaṭṭhikā-rāhulovāda Sutta (Majjhima Nikāya 61), “Instructions to Rahula at Mango Stone,” trans. Ṭhānissaro Bhikkhu (Access to Insight, 2006; CC BY-NC 4.0); the translator’s note 1 gives the Commentary’s age for Rāhula. Pāli from SuttaCentral, MN 61:8.1–8.4: kimatthiyo ādāso (“what is a mirror for?”); paccavekkhaṇattho (“for reflection”). The translation prints a dash in the question Rāhula asks before an act, which is therefore paraphrased.
- Patañjali, Yoga Sūtra 1.14, sa tu dīrghakāla-nairantarya-satkāra-āsevito dṛḍhabhūmiḥ, trans. James Haughton Woods, The Yoga-System of Patañjali, Harvard Oriental Series 17 (Cambridge, MA: Harvard University Press, 1914), p. xxxi (the sūtras in translation), paraphrased: practice “becomes confirmed when it has been cultivated for a long time and uninterruptedly and with earnest attention”.
- Marilyn Strathern, “‘Improving ratings’: audit in the British University system,” European Review 5, no. 3 (1997): 305–321, at 308. Goodhart’s original observation: C. A. E. Goodhart, “Problems of Monetary Management: The U.K. Experience” (1975). Intentions made public (added with “The glitch” in 33.6): Peter M. Gollwitzer, Paschal Sheeran, Verena Michalski and Andrea E. Seifert, “When Intentions Go Public: Does Social Reality Widen the Intention-Behavior Gap?,” Psychological Science 20, no. 5 (2009): 612–618, abstract (PubMed 19389130): noticed intentions “were translated into action less intensively” than ignored ones, and noticing gives “a premature sense of possessing the aspired-to identity”. Four studies of 49, 30, 57 and 24 students at German universities; the effect only among those committed to the identity goal; Study 1’s enactment self-reported (authors’ copy, University of Konstanz).
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 “The gap between knowing and doing is not a gap of information. It is a gap of practice”, and that twelve months are “long enough for a genuine transformation”; that the four phases “mirror the arc of the book” and that “Each phase demands a different kind of consciousness”; that the audit divides tasks AI can perform from “tasks that require human consciousness”, and that its third question is “what do I do that a machine cannot”; that Priya’s “genuine, irreplaceable consciousness was fully engaged for a fraction of her working day”; that the real boundaries are “the structural ones rooted in the absence of consciousness”, and that the weight of being wrong “is yours. It is also your moat.”; that insight into users “is accumulated, not computed” and that “No AI caught this”; that “It is not enough to be good. You must be known for the specific things that make you irreplaceable”; that “The practice adapts because the practitioner is awake”; and that “You do not arrive at AI-proof. You practise it.” Two section titles kept withdrawn claims (“AI Integration with Consciousness at the Controls”; “Becoming Known for What Only You Can Do”), and all four phase titles used dashes. It also opened with “This chapter is a twelve-month practice” and a list of readers around the world; set Priya’s year from May 2026 to April 2027, beyond the book’s present, ending at a bus stop on a route that does not pass her home; had her build a “Trust Ladder” she had built the year before, keep a daily anomaly review, read code for “forty minutes” in a break room described by a stereotype, and be approved on the strength of a single “Yes.”; called the notebook a tool “she had been given”; and told a parallel year for Kavitha Sharma that repeated or contradicted her own chapters (teaching stories named in Chapter 10, a state pilot dated after the book’s present, where Chapter 28 has only a committee’s draft module by mid-October 2026, a neem tree “leafless in winter”, and a student named Meena, one letter from Meera).
- What I now hold: a practice of self-examination can diagnose and can also plead: when the one who keeps the account wants to be found necessary, the account becomes a brief for the defence (Seneca’s nightly court of one, whose verdict is a pardon with a warning). Machines now do much of the noticing and counting, faster and better; taking a finding as yours is an act. Being the one who answers for something is not being the only one who could, and a person’s worth never depends on being needed. What changes a person is an act laid open to someone who can contradict it (the Buddha’s order to Rāhula: look, lay the act open, restrain), and the thought may stay private. Part of what made Priya “needed” was her employer’s praise and her own arrangement of others’ turns, and some of it was not hers to change. The audit’s third question becomes: of what is left, what would someone have to explain and put right, and is that someone me? Distribution means making work checkable, not making oneself irreplaceable. A self-measure wears out once it is known. A year promises no transformation and no job, and the hours it needs belong to someone who must be asked. Being needed can be an employee’s only protection, and weighing that is not wrong; its cost to others is still paid. A habit may change in private; a plea needs a second hearer, because its judge is the one who makes it. Change has several measures (what she sees, what she does, who decides, how the company works, why she acts, who she takes herself to be), and acts others can see are better evidence than declarations. A page of questions is not a shared skill until others use, correct and teach it without its author. Handing work on carries conditions: the right to decide, support, and credit. Making knowledge easy to pass on is not making a worker disposable. A check that can never delay a release is not a safeguard.
- What changed my mind: the print’s own opening, which found the notebook pleading; the rule that the book never defines a person by what machines cannot yet do; the machine-side facts that a model sorts a week’s hours and finds a pattern in support tickets when it is given them; Seneca’s De Ira III.36, read for the structure of its court; the discourse to Rāhula, which keeps thoughts private and sends harmful acts out; the evidence that people see bias in others more than in themselves; Goodhart’s law applied to the self; the book’s present, which ends half way through the year; and three outside readings of an earlier version of this chapter, in October 2026.
- What survives: “Priya’s pen stopped.”; “Indispensable. She had written it nine times. She had written around it dozens more.”; “The platform I built. The decision only I could make. The judgement the team relies on.”; “Every entry was a small argument for why Priya Chandran could not be removed from the system.”; “I have been lying to this notebook.”; the replacement of herself “with a story about being irreplaceable”; the audit and its first two questions; “Three questions. They are simple. The honest answers are not.”; “no answer for boredom”, now said of the plans sold for self-improvement; the warning that building is where people fall back into old patterns (“They see the map, treat it as a problem, and optimise.”); “vanity dressed as diligence”; “which ones am I keeping because they develop my judgement?” and the walker and the driver at the gym; the third implementation built for a user who finishes in one sitting; “I just did something no AI did. Not because I am smarter. Because I know these users.”, now Priya’s claim, tested and half withdrawn; the question “Are we saying that the AI writes the code correctly but for the wrong person?”, now the chief product officer’s; the five-minute presentation without slides; “The twelve months are a container. The practice is not.”
- Section titles in the print edition: 33.2 was “Month 1–2: Seeing Clearly”, followed by “Where You Stand and What Is at Risk”; 33.3 “Month 3–5: Building”, followed by “AI Integration with Consciousness at the Controls”; 33.4 “Month 6–9: Creating”, followed by “Reinventing Your Portfolio of Work”; and 33.5 “Month 10–12: Distributing”, followed by “Becoming Known for What Only You Can Do”. This edition drops the months, because Priya’s year did not keep the plan’s calendar. 33.1 and 33.6 keep their print titles.