NP Nithin Palal

Chapter 4: Why This Matters Now

Part I: The Treasure You Already Carry

· Updated 15 November 2026 · 40 min read · Discuss · Plain text

AI-Proof: What Machines Can Never Be — Volume I: Awakening, Part I: The Treasure You Already Carry, Chapter 4: Why This Matters Now

Part I looks at the person. After the distinctions of Chapter 1, the traditions of Chapter 2 and the qualities of Chapter 3, this last chapter of Part I asks why the question of what you are cannot wait.

In 1930 Keynes wrote a letter to his grandchildren about a disease called technological unemployment. The letter was addressed to us. If a machine can do what you do, what are you? The terms are being set now.

In Brief For readers who fear what a machine that can do their job leaves them. Chapter 4 asks the question cheap intelligence puts to every working person: if a machine can do what I do, what am I? It sets Hannah Arendt’s distinction between what a person is and who they are beside Yudhiṣṭhira’s answer to the Yakṣa, and takes the honest objections seriously: work was often income, not identity, and work is where qualities are grown. Work bundled four goods, income, practice, standing and voice, and machines can take them apart. The gravest loss may be the missing rung: the junior work where people used to become trustworthy. The terms are being written now.

A machine can take over what you do. Who you are shows in what you answer for, and the terms are being written now.

Opening: The Nephew’s Saturday

Rajan Subramanian’s nephew came to see him on a Saturday afternoon in November, carrying a laptop and the bright face of a young man who believes he is about to change someone’s mind.

Rajan let him talk. He sat behind his desk in the office above his factory in Coimbatore. Through the glass he could see the floor: rows of looms, a hundred and forty people, and the steady noise that had been the sound of his life for more than thirty years. His father had first taken him to the mills as a teenager. He had run his own business since 1997.

Karthik was twenty-six and had just finished business school at IIM Bangalore. He was thorough, and he was kind about it. He showed a system that could grade fabric from photographs. A tool that planned production across all the looms at once. A program that wrote letters to buyers in English, German and Japanese, and changed its tone for each market. At the end he showed a spreadsheet that compared the cost of the software with the cost of the people who now did the same work.

Rajan looked at the spreadsheet for a long time. One column was salaries. He knew every name behind it. Muthu, his floor supervisor, had been with him for nineteen years. Below Muthu were four young men he was teaching to set up a loom and to see a fault in the cloth before it ran for a hundred metres. On the spreadsheet their jobs were the cheapest to replace.

He thanked his nephew. Karthik waited for more. It did not come.

What Rajan did not say was this: I have heard this before, and I am still here. He had heard it in 2005, when the old system of export quotas ended and a consultant from Chennai told him that machines would soon grade cloth better than any man. He had heard it around 2010, when a trade magazine said that online platforms would make personal relationships with suppliers worthless within five years. He had heard it in 2020, in an online seminar about forecasting software. Each time, the technology came. Each time, it was useful. Each time, it failed to replace something Rajan could not name but knew he had.

So he did what he always did. He went back to work. There was a shipment to arrange for Hamburg, with the buyer’s new standard contract clipped to the papers, unread. There was a dispute with a dye house in Tiruppur, and a daughter preparing for her engineering entrance exams, who needed a calm house. The laptop went home with Karthik.

This stubbornness had made Rajan’s career. It had carried him through the end of the quotas, the crash of 2008, the overnight withdrawal of most of India’s cash in 2016, and a pandemic that closed his factory for four months. Stubbornness had earned its keep.

But stubbornness is not the same as discernment. Stubbornness does not tell the difference between the threats you have survived and the one you have not yet understood. It says: I am still here, so my ways are right. It filters out every signal that does not agree with it. In the weeks after Karthik’s visit, Rajan’s stubbornness was working hard to filter out one feeling.

It was not a feeling about the technology. Rajan had never been afraid of machines. It was a feeling about himself. He could not say what he had that the software did not. For thirty years nobody had asked him. Now the question was there, quiet and persistent, like a sound in the walls that you cannot trace.

The question was this. If a machine can do what I do, what am I, and why must I answer that now?

4.1 The Letter from the Future That Already Arrived

In 1930, early in a world depression, the English economist John Maynard Keynes wrote an essay addressed to people who were not yet born. He called it “Economic Possibilities for our Grandchildren”. He was writing to us.

Keynes made two predictions. The first was about a new illness. “We are being afflicted with a new disease,” he wrote, “of which some readers may not yet have heard the name, but of which they will hear a great deal in the years to come, namely, technological unemployment.” He defined it as unemployment “due to our discovery of means of economising the use of labour outrunning the pace at which we can find new uses for labour.”1 Machines would learn to do our work faster than we could find new work to do.

The second prediction was hopeful. Keynes believed that within a hundred years, around 2030, “the economic problem may be solved, or be at least within sight of solution.” Machines and savings would make people rich enough that they would barely need to work. He thought three hours a day might be enough. And then, he wrote, “for the first time since his creation man will be faced with his real, his permanent problem: how to use his freedom from pressing economic cares, how to occupy the leisure, which science and compound interest will have won for him, to live wisely and agreeably and well.”1

We are close to his date. How did he do?

On wealth, he was roughly right. In the rich countries, and in much of the rest of the world, people produce many times what their great-grandparents did. On new work, the first prediction was too gloomy. Each wave of machines destroyed jobs, and each time new kinds of work appeared. One study found that about six in ten American jobs in 2018 were in kinds of work that did not exist in 1940.2 On hours, he was wrong. The working week in the rich countries did grow shorter over the century, but nowhere near fifteen hours.3 And the leisure that did come did not become the problem he named. People with plenty did not stop working, and many of those with the most choice chose to keep working long.

Why? Keynes himself half saw the answer. He feared that people with nothing to do would suffer “a general ‘nervous breakdown’.”1 Work had become more than a way to earn bread. It had become the way people knew who they were, and the way others knew it too. That is why his letter matters now. His problem was labour saved faster than new uses for labour could be found. Ours is wider. Machines now produce thinking faster than we can rebuild the places where people learn to think, and to be trusted. The question he left us is the one Rajan could not name.

Here is what has arrived, as of 2026.

In February 2024, the Swedish payments company Klarna announced that its AI assistant had handled 2.3 million customer conversations in its first month, two-thirds of all its customer-service chats. It said the system was doing the work of 700 full-time agents.4 That was the company’s own claim, made before it sold shares to the public, and it was widely repeated as proof that the replacement had begun.

Fifteen months later the same company changed course. Its chief executive, Sebastian Siemiatkowski, told a reporter: “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” He added: “Really investing in the quality of the human support is the way of the future for us.” And: “I just think it’s so critical that you are clear to your customer that there will be always a human if you want.”5

Read the two announcements together. The machine could do much of the work. Klarna still promised customers a person. The company did not say why customers wanted one; my reading is that some wanted someone who could be asked why, and who would answer. But notice what kind of person was offered. The new human agents were to work in what Siemiatkowski called “an Uber type of setup”, logging in remotely when needed.6 The evidence shows that a person stayed in the loop. It does not yet show that the person in the loop is someone who answers.

The wider evidence points the same way, with a warning attached. In August 2026, economists at Stanford looked at young workers, aged twenty-two to twenty-five, in the jobs most exposed to AI. By June 2026 their employment was about 19 per cent below where it would have been had it kept pace with similar workers in less exposed jobs. The gap came “primarily through reduced hiring of young workers rather than increased separations.” In jobs where AI helps people rather than replacing their tasks, employment was flat or rising. And the authors were careful to say that they did “not see widespread, economy-wide job displacement.”7 They also call their figures “early, descriptive indicators” rather than “causal estimates”. The gap holds when they allow for rising interest rates. It narrows when they allow for education, and some of it began before these tools reached the public.7 In October 2026, researchers at Yale found that across the American economy the mix of occupations was “not yet changing in ways that clearly align with the introduction of AI into the workforce”.8 The two findings fit together. Yale measured the mix of jobs across the whole economy. Stanford measured one narrow group, and a small group can shrink a great deal while the whole barely moves.

India shows the same pattern. In the financial year that ended in March 2026, the five largest Indian IT companies together employed about 7,400 fewer people than a year before. The year before that, they had added more than 12,700. They pointed to “an increasing focus on Artificial Intelligence (AI) and changing demand.” Several also cut the number of new graduates they planned to hire.9

So the honest picture, as of 2026, is neither panic nor comfort. Most jobs have not disappeared. But the first step on the ladder, the job a young person takes to learn the work, is being removed first, quietly, through hiring that does not happen. Section 4.3 explains why that matters more than it looks.

There are three clocks. The machines change in months. Institutions, laws, schools and habits change in years. And the formation of a person who can be trusted with hard work takes a decade or more. The third clock is the slowest, and it cannot be hurried. The distance between the clocks is where people get hurt, and it is also where the most important choices are made.

Rajan’s nephew brought the first clock into his office on a Saturday. Others hear it differently: in a memo about restructuring, in a dashboard that turned green without them, in a hiring freeze, in the slow realisation that the work they spent years learning can now be done by a system that costs less than a month’s salary.

The question these moments ask is always the same one. If a machine can do what I do, what am I?

4.2 What Happens When You Don’t Know What You Are

The most dangerous answer to that question is the quick one: I am my skills. I am the contract review, the radiology report, the production plan, the fabric grading. I am the work, and if the machine can do the work, the machine has what I have.

The print edition of this book called this answer a simple mistake, a confusion of the tool with the person holding it. I now think that was too quick, and too comfortable. The mistake is real, but it is not simple, and the people who make it are not foolish. To see why, we need a better way of asking the question.

The German-born American philosopher Hannah Arendt gave one in 1958, in The Human Condition. She distinguished between what a person is and who a person is. What you are is your “qualities, gifts, talents, and shortcomings, which he may display or hide.” Who you are is something else. “In acting and speaking,” she wrote, “men show who they are, reveal actively their unique personal identities.”10 The what can be listed on a CV. The who appears only in what you do and say among other people, and especially in what you begin, promise and stand behind.

Arendt added something strange and true. The who “remains hidden from the person himself,” though it “appears so clearly and unmistakably to others.”10 You cannot see your own who in a mirror. The people who deal with you can.

This gives a precise shape to Rajan’s question. A machine can match, and in some cases exceed, much of what a person is: their skills, their knowledge, their speed, even, as Chapter 3 showed, the outward signs of their qualities. That is the part of you that can be compared, priced and replaced. A machine can also describe you, predict you, and model your habits. Arendt’s point is that no such list, however long, is the who. The who is a particular person, acting and answering, in a world shared with others. And it cannot be produced on purpose, the way a skill can be displayed. It appears when you act.

Arendt drew one more distinction that matters here. She divided human activity into three kinds. Labour keeps life going and is used up as it is done: cooking, cleaning, answering the hundredth request for a refund. Work makes things that last: a table, a bridge, a bolt of cloth. Action is what passes between people: beginning something new, speaking, and making promises.11 Machines first took over labour and work. Action is where the who appears, and it is the hardest kind of activity to hand to anyone else, because it is the activity of being someone among others.

As I read her, action is not confined to a job. A promise kept to a neighbour, or a word said at a village meeting, is action in her sense, so a person without work still acts. When software agrees to terms on someone’s behalf, the promise binds that someone. The who that appears is theirs.

An Indian text asked a related question two thousand years earlier, and answered it in a way that still surprises. In the Mahābhārata, the eldest Pāṇḍava brother, Yudhiṣṭhira, comes to a lake where his four brothers lie dead. A spirit, a yakṣa, has killed them for drinking without answering its questions. Notice the shape of the scene. The lake is plenty. The condition of drinking is the question. The brothers reached the water and assumed that reaching it was enough. The Prologue began with the same image: when water is everywhere, someone still has to say which water is safe. You do not earn the water by having reached it. Yudhiṣṭhira agrees to answer. One of the questions is what makes a person a brāhmaṇa, the highest rank in the society of the epic: birth, study, or learning? Yudhiṣṭhira replies:

śṛṇu yakṣa kulaṃ tāta na svādhyāyo na ca śrutam | kāraṇaṃ hi dvijatve ca vṛttam eva na saṃśayaḥ ||

In Kisari Mohan Ganguli’s translation: “Listen, O Yaksha! It is neither birth, nor study, nor learning, that is the cause of Brahmanahood, without doubt, it is behaviour that constitutes it.”12

The word translated “behaviour” is vṛtta: conduct, the way a person lives and acts. Not kula, the family you were born into. Not svādhyāya, the texts you have recited. Not śruta, what you have learned.

Three things about this answer matter for this book, and the first is a warning against reading it too kindly.

The verse does not abolish rank. Read on, and Yudhiṣṭhira says: “One’s behaviour should always be well-guarded, especially by a Brahmana.” A man who has studied all four Vedas but behaves badly “is to be regarded as a wicked wretch scarcely distinguishable from a Sudra.” And: “He only who performeth the Agnihotra and hath his senses under control, is called a Brahmana!”12 The answer still speaks the language of rank, and still uses the lowest rank as an insult. What it does is make conduct the condition of holding a rank, rather than birth. That was a real correction in its own society, and that society mostly did not live by it. For centuries in India, birth decided rank in the law books and in much of daily life, including what work a person was allowed to do. Many family names are still the names of occupations. A book about identity and work, written from India, should say that plainly. The idea that you are your work, fixed from birth, is not only a modern office habit. It was written into law books and lived in villages for a very long time.

Second, I am applying the verse, not reporting it. Yudhiṣṭhira was asked who counts as a brāhmaṇa, not what a textile worker in Coimbatore is when a machine can grade cloth. Chapter 2 marked the same kind of stretch when it read the Analects’ “not a vessel” against the machine. With that said, look at the list he rejects. Study and learning are exactly what machines now have in abundance. A model has recited more texts than any priest and learned more than any scholar. If a person were their learning, the machine would now be the better person. Yudhiṣṭhira’s answer, like Arendt’s, points somewhere the machine’s abundance does not reach: to conduct, to what one does and stands behind among others.

Third, conduct must not become a new thing that a person is, a second CV. The print edition said you are not your intelligence; you are the one watching it. It would be just as wrong to say now that you are your conduct. Conduct shows who you are. It does not exhaust who you are, and it is not a possession you can point to. It is what you will be asked about. As Chapter 3 put it, you are not the only cause of what you do, and you still answer for your part.

So far the argument sounds like comfort. Here is the case against it, in full. It has three parts, and each is serious.

The first comes from this book itself. Chapter 3 argued that qualities are worn into us like a footpath, by our own repeated choices. Where do most adults walk that path? At work. Renata’s judgement was made at the heart-team table. Bora’s patience was made in twenty-seven seasons of tea. If a machine takes the work, it takes the field where the path was walked. So “you are not your skills” is half false. You are partly what your work has made of you. A person who loses their work may lose more than income. They may lose the place where they were becoming who they are.

The second comes from the American sociologist Richard Sennett. In 1998 he studied workers in the new, flexible economy, people moved from project to project, team to team, with no long story at work. He found that the deepest damage was not to their pay. It was to their sense of being needed. “’Who needs me?’ is a question of character,” he wrote, “which suffers a radical challenge in modern capitalism.”13 Priya Chandran asked a version of the same question at her desk in Chapter 1: If this works this well, what do they need me for? A philosophy that tells her she is more than her tasks does not answer it. Being needed by others is part of being someone.

The third objection is the sharpest, and it is about class. “What am I, beyond my work?” is a question for people whose work was ever their identity. For most people in the world, it never was. Work was how they fed their children. Of Rajan’s hundred and forty workers, many would say that their identity lies in their family, their village, their faith, their friends, and that the loom is how they pay for all of that. For them the machine does not threaten an identity. It threatens a wage. And a book that answers a threat to wages with a lesson about the self is offering philosophy to people who need income. That is worse than useless. It is an insult.

All three objections are right about something. Here is my reply.

For most of the last two centuries, one institution, the job, delivered several goods at once. Machines can take them apart. The word “lose” is hiding four different losses. A machine can take your income: the wage, the security, the ability to plan. It can take your practice: the daily field in which your qualities are worn and kept. It can take your standing: the sense that someone needs you, that what you do is asked for and counts. And it can take your voice: your say in how your work changes. A person can keep their wage and lose their voice, when decisions about their work are made in a spreadsheet they never see. These are different losses, and they need different remedies. Nor are they everything a job gives. It also shapes a day and gathers people, as Chapter 29 shows.

Three different questions hide here, too. What is a person, beyond their work? That is a question for philosophy. What is a person owed when their work goes? That is a question for politics. And who gets the gains the machine makes: customers, owners, or the workers whose skill it learned from? That is a question for economics and law. This book can answer the first. It insists that the third be asked. It cannot settle the second, and it will not pretend that a lesson about the self pays the rent. The rewrite of this book makes a narrower promise than the print edition did. It does not promise you employment. It promises to help you keep authorship: to know what is still yours, and to keep it.

The loss of practice is real, and the objection from Chapter 3 is correct. That is why the next section is about where the path is walked, and why its loss falls hardest on the young.

The loss of voice is the one Chapter 1 described in Priya: responsibility kept, authority taken away. Muthu has supervised Rajan’s floor for nineteen years. Karthik’s spreadsheet decided nothing yet, but nobody had asked Muthu what it should decide. Voice is the Prologue’s question “Who may decide?”, asked from the floor.

Standing is where Arendt and Yudhiṣṭhira help most. Sennett’s question, “Who needs me?”, is the right question. But look at what kind of need it is. People do not only need tasks done. They need someone who will do them for them, and answer to them. If my reading of Klarna is right, its customers did not need a person to process a refund; the machine could do that. Some of them wanted a person who could be asked why, and who would still be there if it went wrong. That kind of need does not go away when tasks are automated. It moves, from the doing to the answering.

Standing can also drain away inside a job, before the job itself goes. A junior who once learned beside a senior now checks a machine’s output that nobody asks her about. Her name is on the work, and her judgement is not wanted. Priya, in Chapter 1, knew that feeling at five o’clock on a Tuesday.

One caution, so that none of this is misread. A person’s worth does not depend on being needed, or on being able to answer. A child cannot answer for anything. Nor can a patient in a coma, or many people living with serious illness or disability. Their worth is not smaller. As the Prologue said of the third chair, someone must answer on their behalf. Standing is a good that people need and should not be stripped of carelessly. It is not the measure of what a person is worth.

This is where the print edition’s best sentence belongs, made exact. The print edition said: “The machine does not face consequences. Its output is weightless. Yours is not.” As Chapter 1 showed, machines in tests have acted to avoid being switched off, so it is not quite true that nothing is at stake for them. What is at stake for a machine was put there by its makers, and it can be copied, restored or changed. What is at stake for you cannot. So the sentence becomes: the machine’s stakes are installed and copyable; yours are lived, once. That is why, when something must be answered for, people still ask for a person.

One more correction. The print edition said that the Indian traditions warned against “mistaking the vehicle for the passenger: the intelligence for the awareness.” That had the image wrong. In the Kaṭha Upaniṣad, as the Prologue and Chapter 2 showed, the intellect is not the vehicle. It is the charioteer. The self is the one whose journey it is. The person who says “I am my skills” has mistaken themselves for the driver. When a better driver arrives, they have nowhere to sit. The person who knows they are the one whose journey it is can hand over the driving, and still decide where the chariot goes, and answer for it.

So the answer to Rajan’s question, as far as this chapter can take it, has two parts. What you are, a machine can increasingly match. Who you are shows in your conduct among others: what you begin, what you promise, and what you answer for. The first can be taken from you. The second can be pressed out of you by an institution, or given up by you, but it cannot pass to a machine without someone deciding who will carry what follows.

4.3 The Deficit: Why Most People Are Unprepared

If the distinction is so old, why are so few people ready for it?

The print edition called this “the consciousness deficit”. That was a mistake in words. Everyone reading this is conscious. Nobody lacks consciousness the way a car lacks fuel. What most of us lack is practice: practice in choosing which problem is worth solving, in refusing what we should not sign, and in answering for what we did. The deficit is not in our minds. It is in our training.

Look at what that training rewards. Rajan’s daughter is preparing for one of the hardest entrance examinations in the world. In January 2025 about 1.3 million young Indians sat the first session of the national engineering entrance test, and only a small share will win seats in the most famous institutes.14 To prepare, she solves thousands of well-posed problems, at speed, under the clock. Each problem has one right answer, already known to the examiner. She is very good at it. She has given two years of her life to it.

Notice what kind of skill this is. It is the skill of solving a clearly stated problem quickly and correctly. As the Prologue reported, machines now solve olympiad mathematics at the level of a gold medal. The exam is training her, with enormous effort, in exactly the ability that machines have learned fastest. It gives her almost no practice in the things that remain hers: choosing which problems matter, saying “I don’t know yet”, disagreeing with a confident answer, or standing behind a decision whose right answer nobody knows.

This is not her fault, or her teachers’, or even the examiners’. And the exam is not worthless. It builds concentration, persistence and real fluency in mathematics. For many of those 1.3 million families it is something more: the one gate that does not ask who your father is. That is its great merit, and it is why poor families make sacrifices for it. But it was designed for a world in which trained intelligence was scarce. It selects, from millions, the people who can solve known problems fastest, and it trains almost nothing else. The world changed. The exam did not. None of this is a reason for her to stop preparing. The gate is real. It is a reason to see what the exam does not train, and to find a place to practise that too.

Offices are built the same way. They measure output, speed and cost. A person who answers at once is seen as decisive. A person who says in a meeting, “I need to think about this before I answer,” is seen as slow. Chapter 3 showed what this does: the walkway is measured, and the walking is not. A whole working culture was built to treat fast cognitive output as competence. That made sense while only people could produce it. It no longer does, and the culture has not caught up.

Now the most serious part of the deficit, and the reason this chapter is called Why This Matters Now.

Recall the Stanford finding from 4.1. The effect of AI on jobs, so far, falls mainly on young people entering codified work, through hiring that does not happen. Experienced workers in jobs that depend on knowledge learned on the job have held their ground, or gained.7

At first this looks like good news for the experienced. Look again. Where did the experienced get their experience? From the junior work that is now not being hired for. Renata learned judgement by first being the junior who read scans and was corrected. Priya learned to find edge cases by first running the regression suites by hand. Rajan learned to read cloth by spending years on the floor, being wrong in front of men who were right. The first rung of the ladder was never only a job. It was where the footpath of Chapter 3 began.

If the first rung is removed, the path is never begun. The machine does the junior work well. Then, ten years later, there are fewer people who can check the machine, because nobody walked the path that checking requires. I will call this the missing rung: the loss, not of today’s jobs, but of tomorrow’s judgement. It does not show in any unemployment figure. It shows only later, when a firm looks for a senior person who can say “this is wrong” with authority and finds that nobody was formed to do it.

An economy makes people as well as goods. A junior lawyer becomes a senior lawyer by doing junior work. A young engineer becomes an engineer by fixing dull systems. A weaver learns cloth by spoiling some of it while someone better watches. Employment was never only a market for labour. It was also, without anyone planning it, the place where a society formed people it could trust.

The Prologue distinguished delegation from abdication. The missing rung needs a third word. Displacement is what happens when the machine does a task, so nobody learns to do it. Delegation keeps a person who answers. Abdication keeps nobody. Displacement keeps a person for now, and stops making the next one.

Think of three generations. The first learned the work by hand before the machines came, and can still tell when the machine is wrong. The second learned alongside the machines and knows both. The third will enter after the machines, supervising systems it never learned to do by hand. The danger is not that the third generation is less intelligent. It is that it may have nothing to check the machine against. Juniors also do something else that is easy to forget: they ask naive questions, notice odd things, and challenge their seniors. A firm without juniors loses its future seniors, and it also loses some of the questions that kept its present seniors honest.

The distinction between a task and a role helps here. In 2016 one of the founders of modern AI, Geoffrey Hinton, said: “People should stop training radiologists now.” Machines did become very good at reading scans. Yet the number of radiologists in the United States rose over the following decade, and training places reached a record. Diagnosis turned out to be only about a third of a radiologist’s time. The rest is overseeing examinations, talking to other doctors, teaching, and answering for the result.15 A machine took over a task; the role remained. The missing rung opens when institutions redesign the role around the machine, and stop hiring the people who would have grown into it.

This is why the deficit is not equally correctable for everyone. A senior professional can choose, as Chapter 3 suggested, to walk first and ride second. A twenty-three-year-old who cannot get the first job has no field to walk in. The advice “practise your judgement” is easy to give and hard to follow without a place to practise.

None of this was anyone’s plan. It is the sum of many sensible decisions: each firm hiring fewer juniors because the machine does junior work, each school training for the exam that exists, each worker producing the output that is measured. Each choice is reasonable. Together they remove the place where people used to become trustworthy. Economists have a name for this shape. Each firm hopes that some other firm will train the senior people it will need, and then hire them away. When every firm hopes the same thing, nobody trains them. It is the old problem of the common pasture, applied to judgement. Some governments answer it by making large firms pay towards training. Since April 2017, employers in the United Kingdom with a pay bill over £3 million have paid a levy of 0.5 per cent of it, less an allowance.15 In England the money pays for apprenticeships and, since April 2026, for some short courses too. Money for courses is not the same as a first job beside someone better, so whether a levy rebuilds the rung is still open.

The machine does not have to be the enemy of the rung. It can also rebuild it. Two of the best studies so far point the same way. In a trial with about a thousand secondary-school students in Türkiye, those given an AI that simply supplied answers did better in practice and then worse in the exam taken without it. Those given a version that offered only hints did much better in practice and no worse in the exam.16 In a 2026 study of junior software engineers, those who used AI to hand over the work learned less than those who wrote the code by hand. Those who used it to ask questions about the code, and to check their own understanding, kept more of what they learned.17 The lesson is not “keep the machine away from the young”. It is: let the machine do the work, and keep the young person responsible for understanding it. A junior doctor can read the scan first, then compare with the machine, then hear the senior explain both. A junior lawyer can check every clause the machine drafted. Then the machine becomes a training ground, not a replacement. The rung is not removed. It is redesigned.

This is a thesis, not a prophecy, and I should say what would prove it wrong. If, over the next five years, firms that use AI heavily hire more young people rather than fewer; if young workers trained with AI become trusted seniors faster than their predecessors did; if new kinds of work appear that form judgement as well as the old junior work did; then the missing rung will have been a passing fear. The evidence as of 2026 points the other way. It may not stay that way, and the book will say so if it changes.

The deficit can be corrected. People can learn at any age to choose, to refuse and to answer, and the practices in this book are meant for that. But the correction is not only private. The missing rung is a design problem, for firms, schools and governments. Here it is enough to see it, and to see that it is happening now, while the habits of the next twenty years are being set.

Receipts, please

Two studies carry the hope in this section, so their limits belong here too.

The trial in Türkiye compared nearly a thousand secondary-school students with students who never had the machine. Students given the plain chatbot scored 17 per cent lower in the exam than those who never had it. With the hint-giving tutor, that harm was “largely mitigated”, in the authors’ words. The tutor’s large gain came in practice, while students still had it.16 So the study shows that a well-built tutor need not damage learning. It does not show that it adds to it.

The coding study was smaller: 52 developers, most of them junior, learning one unfamiliar library. They were assigned at random to work with the assistant or without it, and that part is a fair test. The group with the assistant scored 50 per cent on the quiz; the group without it scored 67. The finding about asking questions is weaker evidence. It sorts people by how they happened to use the assistant, with two to seven people in each pattern, and the authors say it does “not draw a causal link”. The quiz came straight after the task, so it says nothing yet about what lasts. The study was run by the company that makes the assistant.17

My reading. A redesigned rung is a promising idea on a small base of evidence. It is a reason to try it and to measure it, not a result to rely on.

4.4 The Invitation

There is a difference between urgency and panic.

What this chapter describes is urgent. It is not a reason to panic. Most work has not disappeared. The skills people have built have not become worthless. They remain necessary. They are no longer enough on their own.

But urgency has a precise reason here, and it has a name. In 1980 the British scholar of technology David Collingridge described a trap that every powerful technology sets. “When change is easy, the need for it cannot be foreseen; when the need for change is apparent, change has become expensive, difficult and time consuming.”18 Early on, we can shape a technology cheaply, but we do not yet know how. Later, we know exactly what we should have done, and it is too late to do it cheaply.

As I write, in 2026, AI is in the middle of that window. Who answers when a system decides about a loan or a diagnosis; whether customers keep the right to a human; whether firms keep the first rung; who shares in the gains: these terms are being written now, in contracts, in hiring plans, in laws and in habits.

They are written in small documents. On the night after Karthik’s visit, Rajan finally read the Hamburg buyer’s new standard contract. Clause 11 said that disputes about quality would be settled by the buyer’s automated inspection report, and that the report would be final. The old contract had named a quality manager to phone. Nothing in the new one was dishonest. It was simply a term in which nobody answered. Rajan phoned Jürgen Weiss, who had bought from him since 2013, and asked for a name to be put back in the clause. He did not ask for the report to go. It might well judge cloth better than any manager. He asked for someone to phone when it was wrong. Weiss said he would ask his lawyers. Collingridge’s window is made of clauses like that one, signed or questioned in ordinary weeks. In India, a government planning body made an estimate in 2025. On present trends, it said, employment in IT services could shrink to about six million people by 2031. With the right policies, it could grow to about ten million.19 The gap between those two numbers is not a fact about machines. It is a choice about people.

The economist Erik Brynjolfsson has warned about a related trap. When machines are built to replace what people do, he argues, “workers lose economic and political bargaining power.” When they are built to extend what people can do, people keep a claim on the value they help create.20 The difference is not decided by the technology. It is decided by those who build it, buy it and govern it.

Hannah Arendt saw the stakes in 1958, before any of today’s machines existed. In the opening pages of The Human Condition she imagined automation emptying the factories, and asked what would become of a society that had made work its whole meaning. “What we are confronted with,” she wrote, “is the prospect of a society of laborers without labor, that is, without the only activity left to them. Surely, nothing could be worse.”21 That is Keynes’s “permanent problem”, seen from its dark side. If work was the only way we knew who we were, the loss of work is the loss of the self.

So this chapter’s invitation is not “find a skill the machine does not have”. That list grows shorter every year, and the Prologue promised never to define a person by it. The invitation is to decide, now, four things. Which of them you can decide depends on where you stand.

First: what will you keep answering for? Not every task. Most tasks can be delegated, and should be. But somewhere in your work there are decisions that should carry your name, where someone may one day ask “why?”, and where the honest answer must come from a person. Know which they are, before someone else decides for you that nobody needs to answer at all.

Second: what walking will you keep? Chapter 3 showed that qualities are worn by practice, and that tools can let the path grow over. Decide which practice you will keep doing yourself, so that you stay able to judge the machine’s work.

Third, if you are not the one who decides: where will you claim a voice? Muthu cannot choose whether the looms are automated. He can ask to be asked, and he can say what the spreadsheet does not know: which faults only a person catches, and who on the floor is learning to catch them. A voice is not a veto. It is a claim to be part of the deciding. A claim can also cost something. Where speaking up would cost a person their place, the duty runs the other way. The one who decides should ask.

Fourth, if you employ others: will you keep the rung, or redesign it? A firm that hires no juniors saves money this year and buys a shortage of judgement in ten. For Rajan this question has names. They are the four young men learning from Muthu, the cheapest line on Karthik’s spreadsheet. He knew every name in that column. Until now he had not seen which of them were the rung.

I must be honest about who this invitation reaches. A data-entry worker in Manila whose work has been automated is not helped by a distinction between what and who. The loss of income is real, and the work she did may not come back. This book does not pretend otherwise. What it can say is narrower. She is not the task she did. And the people who decided that her work could be done without her are answerable for that decision, as the Prologue insisted. What she is owed is a political question, and it must be put by people who can change the rules. That is not a job.

Keynes thought that when the economic problem was solved, we would face what he called “his real, his permanent problem”: how to live wisely and well. He was early. The machines are arriving at last, though not in the way he hoped. The question he left is the one this chapter has asked. What you can do is becoming cheap. Who you are shows in what you begin, what you promise and what you answer for. Institutions can press that part out of you, and you can give it away. It cannot pass to a machine unless someone decides who will carry what follows.


Six weeks after Karthik’s visit, Rajan is at his desk late in the evening. The looms are covered. The floor is quiet.

He has, despite himself, been using the software. Not much. Karthik had left him a login for one module that checks suppliers. In a moment of curiosity he would admit to nobody, Rajan used it on a weaving unit near Erode that could take his overflow orders. The module pulled the firm’s credit history, its delivery records, court filings, and reviews from buyers in three countries. Everything was clean. The prices were good. The sample cloth was excellent.

The data said yes. Rajan said: not yet.

The next morning he drove to Erode. Two hours each way. He did not phone ahead. He walked the corners of the place: the loading bay, the space behind the looms where waste yarn collects, the store room. In his experience the corners of a business tell you more than its showroom. He ran a length of finished cloth between his thumb and finger.

Then he sat with the owner over tea. The man was in his thirties. He had started the unit five years ago with a bank loan. His father had been a schoolteacher, not a mill owner, and nobody in the old Tiruppur families knew him. Rajan asked questions he already knew the answers to, because he was listening for something other than the answers. There was a pause before the reply about delivery times. There was an assurance about quality that came a little too fast. Once, before a question about payment terms, the owner’s eyes went to his accountant.

Rajan declined. He gave the overflow to an older unit he had used for years, at a price about six per cent higher.

On Sunday Karthik phoned, as he does every Sunday. Rajan told him what had happened. There was a silence on the line.

“The module had nothing on him,” Karthik said. “What did you see?”

“The loading bay,” Rajan said. “And his eyes.”

“Would you have seen the same thing,” Karthik asked, “if you had known his father?”

Rajan did not answer. After the call he sat for a long time. He had trusted this feeling for thirty years, and it had saved him more than once. He also knew, because he was honest, that it had been formed among men like himself, in a trade where trust ran through families who had known each other for generations. A feeling formed that way can read a man. It can also read only where a man comes from. Chapter 3 called intuition a judgement built from long experience. Experience compresses prejudice as well as knowledge. Chapter 3 also asked whether a surgeon’s memory was a broken leg, a real fact the formula had missed. Rajan’s question was harder. He could not tell whether he had seen the man or only the man’s newness. The module had blind spots of its own; it had never stood in a loading bay. Every kind of judgement, his and the machine’s, makes its own kind of error. The question is who is placed to catch which.

On Tuesday the owner from Erode phoned him. He was polite, and he was direct. He asked why.

Rajan could have said that the order had gone elsewhere, which was true and told him nothing. Instead he told him about the loading bay: the broken pallets, the cones of yarn left in the open where the damp would reach them. He did not mention the eyes. He did not know how to say that part honestly.

The young man was quiet. Then he said: “Come back in three months.”

Rajan said he would.

Hannah Arendt thought that action, the most human kind of doing, has two dangers and two remedies. The future is unpredictable, and the remedy is the promise: “the faculty to make and keep promises”. The past cannot be undone, and the remedy is forgiveness: release from “being unable to undo what one has done though one did not, and could not, have known what he was doing.”22 Rajan had just made the first. If what he saw in Erode was only the man’s newness, he would one day need the second, and it would not be his to give himself. Arendt wrote that “no one can forgive himself”. Only the young man could give it.22

Practice

Two columns. Take one ordinary week of your work and write down everything you do in it, in a column on the left. In a column on the right, write one thing beside each item: if a machine did this tomorrow, would anyone still come to me and ask why? Most items will get a “no”. That is not a verdict on you. Those are tasks, and many of them can be delegated. But if nearly every item gets a “no”, this exercise cannot help with the first of the four losses, your income, and it would be dishonest to pretend that it can. Look hard at the items that get a “yes”. Those are the places where you answer. Then ask one more question of each: what practice do I need to keep walking so that my answer there stays worth having?

If you have no first job yet, this exercise has little to work on. Then the question changes: where, outside a job, can you find a place to practise, with someone who will correct you? If there is none, that is not your failure. It is the missing rung.

Ending: Erode, Three Months Ahead

That evening Rajan opened his diary to a date three months ahead and wrote one word: Erode.

The software had not asked him for a promise. The young man had. It was the first thing in weeks that the machine could not have done for him, and he was not sure he would keep it.

Words to keep

  • answer for: to be the one who must explain a decision, undo it where it can be undone, and put right what can be put right.
  • entry-level rung: the first job in a field, where a young person learns the work by doing the simple parts and being corrected.
  • income, practice, standing, voice: four goods that a job used to deliver together, and that machines can take apart: the wage; the field where qualities are worn; the sense of being needed; and a say in how your work changes.
  • technological unemployment: Keynes’s phrase (1930) for work lost because machines save labour faster than new uses for labour are found.
  • the Collingridge dilemma: early on, a technology is easy to change but we cannot see what needs changing; later we can see it, but change has become expensive and hard.
  • the missing rung: the loss of the entry-level rung, and with it the loss of the experience that would have made tomorrow’s seniors able to check the machine.
  • three clocks: machines change in months; institutions, laws and habits change in years; forming a person who can be trusted takes a decade or more.
  • displacement: the machine does a task, so nobody learns to do it. Compare delegation (a person still answers) and abdication (nobody answers).
  • labour, work, action: Arendt’s three kinds of human activity: what keeps life going; what makes lasting things; and what passes between people, such as beginning and promising.
  • promise: binding your future self to another person; for Arendt, the remedy for an unpredictable future. Its pair is forgiveness, the remedy for a past that cannot be undone.
  • what and who: Arendt’s distinction. What you are is your qualities, gifts and skills, which can be listed and compared. Who you are shows in what you do and say among others.
  • kula: family, lineage; birth.
  • śruta: what has been heard and learned; learning.
  • svādhyāya: study, especially the recitation of sacred texts.
  • vṛtta: conduct; the way a person lives and acts.
  • yakṣa: a nature spirit; in the Mahābhārata, the questioner at the lake.

Sources

  1. John Maynard Keynes, “Economic Possibilities for our Grandchildren” (1930), first published in The Nation and Athenaeum, 11 and 18 October 1930; reprinted in Essays in Persuasion (New York: W. W. Norton, 1963), pp. 358–373. The original sets the “technological unemployment” clause off with a dash, given here as a comma.
  2. David Autor, Caroline Chin, Anna Salomons and Bryan Seegmiller, “New Frontiers: The Origins and Content of New Work, 1940–2018,” Quarterly Journal of Economics 139, no. 3 (2024): 1399–1465: about 60 per cent of US employment in 2018 was in job specialties that did not exist in 1940.
  3. Michael Huberman and Chris Minns, “The Times They Are Not Changin’: Days and Hours of Work in Old and New Worlds, 1870–2000,” Explorations in Economic History 44, no. 4 (2007): 538–567.
  4. Klarna, “Klarna AI assistant handles two-thirds of customer service chats in its first month,” press release, 27 February 2024: 2.3 million conversations; “the equivalent work of 700 full-time agents.”
  5. Sebastian Siemiatkowski, interviewed by Charles Daly, “Klarna Slows AI-Driven Job Cuts With Call for Real People,” Bloomberg News, 8 May 2025; quotations checked against the full Bloomberg text as syndicated by The Star (Malaysia), 8 May 2025, and against PYMNTS’s report of the same interview; the Bloomberg page itself is paywalled and was not seen in full.
  6. Charles Daly, “Klarna Slows AI-Driven Job Cuts With Call for Real People,” Bloomberg News, 8 May 2025.
  7. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, Stanford Digital Economy Lab, “Canaries in the Coal Mine” update, 12 August 2026: employment of workers aged 22–25 in the most AI-exposed occupations “about 19% below where it would be if it had kept pace with” that of less-exposed peers, as of June 2026 (a pattern the authors describe, not an estimated effect); the original paper (August 2025) reported a 13 per cent relative decline, and on the update’s own measure the shortfall was 15 per cent at the July 2025 data vintage, so 13 and 19 do not measure the same thing. Data from ADP payroll records. The paper (August 2026, abstract and introduction) calls its facts “early, descriptive indicators … rather than causal estimates”; the pattern holds “when controlling for exposure to interest-rate increases and for remote work”, attenuates “when controlling for education”, and shows “some divergent trends that predate ChatGPT”.
  8. The Budget Lab at Yale, “Tracking the Impact of AI on the Labor Market,” 14 October 2026, Key Takeaway 1: “The occupational mix is not yet changing in ways that clearly align with the introduction of AI into the workforce.” The page does not name its measure. The Lab’s first report, Martha Gimbel, Molly Kinder, Joshua Kendall and Maddie Lee, “Evaluating the Impact of AI on the Labor Market: Current State of Affairs,” 1 October 2025, measured change in the mix of occupations across the whole labour market with a dissimilarity index, and said the analysis would be updated regularly.
  9. IANS, “Top IT firms’ hiring turns negative as headcount falls over 7,000 in FY26,” 25 April 2026: combined headcount of the five largest firms down 7,389 in FY26 (April 2025 to March 2026) after a rise of 12,718 in FY25.
  10. Hannah Arendt, The Human Condition, 2nd ed. (Chicago: University of Chicago Press, 1998), §24, “The Disclosure of the Agent in Speech and Action,” pp. 179–80. Pages as cited to the 1958 Chicago printing, whose pagination the 1998 edition keeps, by Trevor Tchir, Existenz 14, no. 2 (2019), and in James Moody’s Duke lecture notes; the printed page itself not seen.
  11. Arendt, The Human Condition, Chapters III–V (labor, work, action), paraphrased.
  12. Mahābhārata, Āraṇyaka Parvan (Vana Parva), the Yakṣa’s questions. The verse quoted, and the lines that follow it, are in the vulgate (Nīlakaṇṭha) recension but not in the constituted text of the Pune critical edition, whose Yakṣa episode (3.297) does not include them; trans. Kisari Mohan Ganguli, The Mahabharata, Vana Parva, Section CCCXI (1883–1896), including the lines that follow on guarding conduct, the four Vedas, and the Agnihotra.
  13. Richard Sennett, The Corrosion of Character: The Personal Consequences of Work in the New Capitalism (New York: W. W. Norton, 1998), final chapter, p. 146. Page as cited from the Norton paperback (2000) in Mere Orthodoxy, “The Great Unmooring”; the printed page itself not seen.
  14. Allen (news report on National Testing Agency data), “NTA releases data of JEE Main 2025 January session exam,” 31 January 2025: about 13.78 lakh registered and about 13 lakh appeared (94.5 per cent attendance).
  15. Joshua Gans, “Radiologists are the canaries in a coal mine,” Substack, 26 September 2025 (Hinton’s 2016 remark; a record 1,208 diagnostic radiology residency positions in 2025; diagnosis “only about a third of their time”); Fortune, 4 May 2026, on radiologist employment rising about 10 per cent over the decade. On the levy: gov.uk guidance, “Pay Apprenticeship Levy”: “a rate of 0.5% of an employer’s annual pay bill”, reduced by an allowance of £15,000 a year, for employers whose annual pay bill is over £3 million; “Apprenticeship levy: how it will work” (gov.uk, withdrawn guidance): payable “from 6 April 2017”. Since April 2026, levy funds in England can also pay for “apprenticeship units”, “short, flexible training courses that upskill existing staff in critical skill areas”: Department for Education, apprenticeship service help centre, “Apprenticeship units” (available “from April 2026”; “the first time levy funds can be used for non-apprenticeship training”).
  16. Hamsa Bastani et al., “Generative AI without Guardrails Can Harm Learning: Evidence from High School Mathematics,” Proceedings of the National Academy of Sciences 122 (2025): e2422633122: about 1,000 students; unrestricted access raised practice scores by 48 per cent and lowered unaided exam scores by 17 per cent; a hint-giving tutor raised practice by 127 per cent with no exam harm. For the receipts in 4.3: the comparison is with students “who never had access” (abstract); the abstract says the negative effects were “largely mitigated” by the tutor, and reports no exam gain from it.
  17. Anthropic, “How AI assistance impacts the formation of coding skills,” 29 January 2026: 52 mostly junior developers; AI-assisted group 50 per cent on a comprehension quiz against 67 per cent for hand-coders; those who asked conceptual questions or follow-up questions retained more. The authors note the pattern analysis is not causal. For the receipts in 4.3: participants were randomly assigned to the two conditions; each usage pattern rests on two to seven people (“Our qualitative analysis does not draw a causal link between interaction patterns and learning outcomes”); the quiz came straight after the task, and the authors leave open “whether immediate quiz performance predicts longer-term skill development.”
  18. David Collingridge, The Social Control of Technology (London: Frances Pinter, 1980), p. 11, page as cited by the International Center for Ethics in the Sciences and Humanities, University of Tübingen; wording as quoted by Evgeny Morozov, in John Brockman (ed.), This Explains Everything (New York: Harper Perennial, 2013), p. 255; the printed page itself not seen.
  19. NITI Aayog, Roadmap for Job Creation in the AI Economy (New Delhi, October 2025): IT services employment of about 6 million by 2031 on current trends, or about 10 million with enabling policies.
  20. Erik Brynjolfsson, “The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence,” Daedalus 151, no. 2 (2022): 272–287.
  21. Arendt, The Human Condition, Prologue, pp. 4–5, as cited to the 1998 edition by secondary sources; the printed page itself not seen.
  22. Arendt, The Human Condition, §33 (“Irreversibility and the Power to Forgive”) and §34 (“Unpredictability and the Power of Promise”), p. 237, where both remedies are introduced; §34 follows. Page as cited to the 1998 edition by Marguerite La Caze, “Hannah Arendt and Self-Forgiveness,” Parrhesia 24 (2015); the printed page itself not seen. The line “no one can forgive himself” is in §33 (checked as quoted by Maria Popova, The Marginalian, 14 July 2021).

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 “legal services firms across London and New York have been reducing junior recruitment” (no source); that Klarna’s figures showed the direction of replacement (without its 2025 change of course); that “I am my skills” is a simple “category error”; that the Vedāntic traditions warned against “mistaking the vehicle for the passenger: the intelligence for the awareness”; that “Confucius warned against becoming a vessel”; that the machine’s “output is weightless”; that people suffer from “the consciousness deficit”; that those who make the shift will thrive, “not because they are smarter” but “because they are awake”; “You are not your intelligence. You are the one watching your intelligence.”
  • What I now hold: the evidence as of 2026 shows replacement concentrated at the entry level, through hiring that does not happen, and no economy-wide displacement yet; Klarna’s own reversal is part of the evidence. “I am my skills” is half true: work is where qualities are worn, and losing it is a real loss of income, practice and standing. A machine can match much of what a person is (Arendt); who a person is shows in conduct (Yudhiṣṭhira’s vṛtta) and in what they answer for. In the Kaṭha the intellect is the charioteer, not the vehicle. The “not a vessel” reading of the Analects is this book’s extension, as Chapter 2 says. The machine’s stakes are installed and copyable; a person’s are lived once. The deficit is in practice, not in consciousness, and its gravest form is the missing rung. The book promises authorship, not employment. Action, as I read Arendt, is not confined to a job. Where a voice would cost a person their place, the duty to ask falls on the one who decides.
  • What changed my mind: Klarna’s statements of May 2025; the Stanford, Yale and Indian IT data of 2025–2026; the audit’s finding that the print edition’s best line (“its output is weightless”) needed to be made exact; Arendt’s who and what; the Yakṣa’s question, read with the history of caste beside it; Sennett’s “Who needs me?”; a class objection to the print edition’s argument.
  • What survives: Karthik’s Saturday visit and Rajan’s silence; stubbornness that is not discernment; the question “If a machine can do what I do, what am I?”; the difference between urgency and panic; the honest sentence about the data-entry worker; Rajan’s two-hour drive and the corners of the business, now with a question he cannot answer.

↑ Back to top

Discuss

Loading the discussion…

More in Artificial Intelligence
  •  ·  26 min read

    Epilogue

    Late in the Gītā, the one who knows most says: ponder it thoroughly, then act as you like. A machine can now set out the reasons on every side. Saying “I answer for this” then is not pride. It is a reply, for your part, to those owed one.

  •  ·  37 min read

    Chapter 40: What This Book Cannot Tell You

    The print edition’s last chapter reduced the book to one sentence, and rested human worth on what machines cannot do. I withdraw both. A book can date its claims and correct them in public. It cannot sit in the room where you decide.

  •  ·  6 min read

    Part IX: The Unfinished Chapter

    The print edition’s last Part said that consciousness never concludes. This one is a single chapter that turns the book on itself: what a book can stand behind, what it owes those who read the print edition, and what it leaves open.