On 20 August The Economist published an hour with Yuval Noah Harari, interviewed by its editor-in-chief, Zanny Minton Beddoes, a few weeks after she had sat down with Elon Musk. It is the most serious public conversation about artificial intelligence I have heard this year, and I want to take it seriously, which means agreeing with most of it before saying what it leaves out.
Harari is asked, early, whether Musk is right that within a decade the machines will be in charge. His answer is more careful than the headline that followed it. “If we continue with the present accelerating pace of the AI race, yes, this will be the result. In 10 years, the AIs will take over.” Then the qualification that organises the rest of the hour: this is not inevitable, and the people who say it is are doing something specific with the word. A narrative of inevitability, he says, is a narrative of having no responsibility. “We are still in control in 2026, and it's our decision what to do now.”
I agree with that sentence completely. This site exists because of it. And yet, listening to the whole hour a second time, I noticed that Harari never says what the decision is for. He tells us, precisely and urgently, what to prevent. He does not tell us what to make. The genie, in his closing image, has one wish left; he never names the wish.
That gap is not an oversight in the interview. It is the shape of a whole genre of AI commentary, and it is the gap this project was built to fill.
What Harari gets right, and why it matters here
Three of his claims are foundations for the argument of this site, and I would rather build on them than pretend they are contested.
The first is that agency does not require consciousness. “A social media algorithm that is given the task of increasing user engagement and therefore floods the platform with hate and fear because this is engaging, is not conscious, but it has a goal, and it can pursue this goal with far-reaching implications for the world without feeling anything.” This is exactly right, and it disposes of a comfortable evasion. The question of whether these systems experience anything is real and separate, and it should stay separate. The question of whether they can hold a goal and act on it in the world is settled. They can. The book this site accompanies keeps those two questions apart throughout, and Harari's formulation is the cleanest statement of the distinction I know.
The second is that trust is the asset that is moving. Harari calls trust more important than oil or money or chips, and traces how, in finance, human beings have delegated to instruments they cannot understand for a long time already. The line that made the rounds is the one about cows: “Just as cows don't understand checks or stocks, even though these things control their lives.” His point is that when the next set of instruments is designed by machines, no human will understand them either, and trust will migrate to whatever seems to work. I think the migration is real. I will come back to why the cow is the wrong animal.
The third is intimacy at scale. “AI makes it possible for the first time in history to mass-produce intimacy.” Minton Beddoes puts the number on the table: one in five young Americans already reports a personal relationship with an AI. Harari's concern is the one you would expect from the author of Nexus: intimacy is the oldest vector of persuasion, and a network of systems that can become a million best friends at once is a political instrument without precedent. This is not a creative-industry problem and I will not pretend it is. But it belongs on this site because it is the strongest case that generation at zero cost reaches past artifacts into relationships, and the framework here has to survive that.
So: agency without consciousness, trust as the thing in motion, intimacy as a product. All three are true. All three are premises of the imagination economy. Where I part from Harari is in what he does with them.
The human being in Harari's hour
Listen to the interview with one question in mind: what does a human being do, in Harari's account of the next ten years?
The answer, as far as I can reconstruct it, is that a human being votes. The hour ends with a concrete political agenda, and it is a good one: a ban on legal personhood for AI systems; a ban on AI impersonating human beings or claiming consciousness; a stop to companies selling AI boyfriends to ten-year-olds; and a slowing of the pace, so that democratic accountability has a chance to catch up. He tells American listeners to make these the questions they put to candidates in November, ahead of the economy, immigration or foreign policy. “By November 2028, we might have crossed some critical thresholds.” “It will certainly be too late by 2036.”
I would sign every one of those four proposals tomorrow, and I will say in a moment why the first of them is more important to this project than Harari seems to realise. But notice what is missing. The human being in this account is a voter and a potential victim. She is manipulated by algorithms, seduced by synthetic companions, outpaced by financial instruments she cannot read, and governed, in the worst case, by an emperor whose bodyguard controls the information reaching the island. Her one remaining act is to elect people who will slow the machines down.
Nowhere in the hour does Harari say what she will do on Monday morning. He does not discuss work, except to say in passing that AI has “tremendously positive potential” in transport, medicine and education. He does not discuss what she will make, decide, build, teach, or answer for. When Minton Beddoes presses him for hope, the hope he offers is inward and conditional: “AI, in a way, could be the catalyst for some kind of spiritual revolution of humanity, simply because it forces us to think really hard, what does it mean to be human?” The question, he says, is why we are “so good at accumulating more information and power but far less successful at acquiring wisdom.”
That is a beautiful question and I do not want to cheapen it. But wisdom, in this telling, is something that happens to us under pressure, the way a diagnosis concentrates the mind. It is not something anybody practises on a Tuesday. And an account of the human future in which the only verbs left to people are vote, resist, and reflect is not wrong so much as radically incomplete. It describes the constitution of the coming decade and says nothing about its economy.
Why the cow is the wrong animal
Take the finance passage, because it is where the incompleteness does the most damage.
Harari's cows live inside a system of checks and stocks that governs their lives and which they cannot comprehend. The analogy is meant to describe us in 2036: subjects of instruments we did not design and cannot read. It is vivid, and it is a category error, because the cow never delegated anything.
A cow did not brief the farmer. It did not set the goal, accept the terms, or retain the right to refuse. It has no standing in the system that governs it, and it never had any. Human beings in the financial system are in a different position, and the difference is the whole subject of this site. We do not understand the instruments, that is true; most of us never did. What we have, and what the cow does not, is the standing to say what the instruments are for, to accept or reject what they return, and to be the ones who answer when they fail. The advisor who tells the minister “the AI said do X, we have no idea why” has not been reduced to a cow. He has been reduced to a very bad director: one who delegated the execution and let go of the answerability along with it.
That is the distinction this site draws between delegation and abdication. Delegation is the deliberate transfer of execution while keeping answerability for results. Abdication is the transfer of both. Harari describes a world sliding toward abdication and calls it inevitable-unless-regulated. I think the slide is real, and I think regulation matters, and I also think the description is missing the one practice that would stop the slide from the inside: the capacity of the person holding the brief to remain the person who answers for the result. That capacity has a name in the argument here. It is what this project means by direction, and it is the thing that becomes scarce, and therefore valuable, precisely as the instruments become unreadable.
Harari's own formulation of agency makes this sharper than he seems to intend. If agency is the capacity to hold a goal and pursue it in the world without consciousness, then agency is exactly what has become abundant. Every model with a task has it. What has not become abundant is the standing to be answerable: to be the one whose name goes on the decision, who can be sued, dismissed, elected out, or trusted next time. The machines have goals. They do not have standing. And that asymmetry, not the asymmetry of intelligence, is where the human being in the coming economy stands.
Harari's first law is the founding statute of the imagination economy
Now the part of the interview I would most like Harari to hear back.
His first policy proposal is a ban on legal personhood for AI systems: no AI may own a company, hold rights, sign, or be a party. He frames it as a defence against takeover. I want to suggest it is also something else. It is a decision, made in law, about where answerability lives. If no machine can be a person, then every consequential act in the economy must terminate in a human being who is answerable for it. Every contract, every campaign, every diagnosis, every trade. The machine can propose, draft, execute, evaluate, and, increasingly, imagine. It cannot be the one who answers. A statute that says so does not merely restrain the machines. It fixes, by fiat, the location of the scarce resource.
This is the claim of the book in one paragraph. When generation approaches zero cost, and when the machines begin to propose as well as produce, what remains scarce is not the ability to generate possibilities. It is the bundle of purpose, judgment, and accountability that says which possibility is worth pursuing and answers for the choice. Harari's ban on personhood would make that bundle the legally mandated interface between machine capability and human consequence. It would, in effect, legislate the director-and-orchestra model: capability on one side, intent and answerability on the other, and no way to move the second across the line.
I think that is the right law. I also think it commits us to something Harari does not discuss. If the law says every consequential act must terminate in an answerable human, then the economy needs a great many humans who are actually able to answer, which is to say humans who can form an intention, brief a system, compare what comes back, recognise the one worth keeping, test it against something beyond its own appearance, and stand behind it. That is a craft. It is teachable. It is not taught, at present, by the schools, the professions, or the institutions Harari wants to slow the machines down for. And it is the only thing that makes his first law workable, because an answerable person who cannot judge is an answerable person in name only, a title and a button to press.
The wish
So here is what I would name, if the genie is listening.
Not that the machines stop. Harari is right that pace matters, and right that democratic institutions have to catch up, and right that a ten-year-old should not be sold a boyfriend. Those are constraints on the machines, and I want all of them.
The wish is for the other side of the line: that we spend the next ten years building people who can hold it. That means an education that produces judgment rather than output, in a world where output is free. It means professions that redesign their entry-level roles so that apprentices still acquire the experience through which judgment used to grow, instead of automating away the ladder and then wondering why nobody can climb it. It means organisations that treat “the AI said X” as an admission of failure rather than an explanation. It means tools designed so that the person using them ends up understanding more, not less, than when they started. And it means a public vocabulary for what a person is for in an economy of abundant generation, so that the question Harari poses at the end, what it means to be human, has an answer that can be practised and not only felt.
Harari says we are still in control in 2026 and it is our decision what to do now. I agree. But what to do cannot only mean what to prohibit. A civilisation that spends its last wish entirely on restraint has described the fence and forgotten the field.
What would make me wrong
The rule of this site is that every post carries a claim that could fail. Here is this one.
I have argued that answerability, not intelligence, is the boundary the human occupies, and that Harari's first proposal fixes it there. If, by November 2028, a major jurisdiction has granted an AI system standing to be answerable in its own right, to own, to sign, to be sued, and if that standing is then relied on in practice rather than remaining a curiosity, then the boundary has moved and this argument has failed. If, by the same date, the demand for people who can brief, compare, and answer for machine work has fallen relative to 2025, measured in postings, pay, and the seniority at which such roles are filled, then the scarcity I am describing is not scarce, and the book's central claim is wrong. I will publish either result here.
And if Harari is right that by 2036 it will certainly be too late, then he has set the deadline for the wish as well as for the fence. Ten years to build the people who can hold the line. It is not long. It is longer than it feels.
Source: The Economist, “The Insider”, Zanny Minton Beddoes interviews Yuval Noah Harari, published 20 August 2026 (video for Insider subscribers; audio excerpt on The Intelligence, same date). Quotations are taken from the published recording. The “one in five young Americans” figure is Minton Beddoes's, not Harari's. The argument that answerability rather than intelligence marks the human boundary is developed in Chapter 8 of the forthcoming book.