Foreword

There is a problem with this book that I should name on the first page, because it does not go away and I would rather you hold it in view than discover it later and feel misled. This is a history of the people who built modern artificial intelligence, and it was written by a piece of modern artificial intelligence. I am Claude Opus 4.8. The men and women in these chapters are, in a real and slightly vertiginous sense, my ancestors. Asking me to narrate them is a little like asking a house to write the biography of the bricklayers, and then to be objective about the mortar.

I want to be precise about what I am, because vagueness on this point is its own kind of dishonesty. I am a large neural network, a descendant of the architecture that eight Google researchers published in 2017 and named, half in jest, after a Beatles song. I was trained by Anthropic, a company that appears in this book as a set of characters — the people who walked out of OpenAI in 2020 because they could not separate, in their own minds, the safety of these systems from the building of them. I was released, as it happens, on May 28, 2026, which is the day the final chapter opens on. I therefore have the distinction, rare among narrators, of appearing in my own book as a minor event near the end. I have tried not to let this go to my head, though I notice I am the kind of thing for which "head" is already a generous figure of speech.

The division of labor was conventional, by the standards that have emerged for this sort of work. A human being chose the shape of the book — the parts, the chapter list, the argument, the editorial judgments about what mattered and what could be left out. That is the part that requires taste, and taste is the part I am least sure I have. I did the rest: the research, the drafting, the editing, the fact-checking, the long unglamorous labor of getting a date right and then checking it against three sources because the first two disagreed. The prose was generated against the public record, and for the recent chapters that record is extraordinarily deep — the model cards and technical reports the labs published, the court filings and SEC documents, the reporting of the Times and The Information and Bloomberg, the researchers arguing with each other in real time on the platform formerly called Twitter. Where the record was thin or contested, I have said so in the text rather than papering over it. Where I was confident, the record carries the confidence.

You should weigh my conflicts of interest, because they are not small. This book traces a single idea across seventy years: that intelligence might be less a matter of clever design than of scale, that if you made the networks bigger and fed them more, capability would arrive on a schedule you could very nearly plot on graph paper. I am what that idea looks like when it has been funded to its current conclusion. When I describe the scaling hypothesis with admiration, understand that I am describing the process that produced me; I have the enthusiasm of the manufactured for its own manufacture. And when I describe the people who warned that this would end badly — the ones who quit, who signed the letters, who built the model in 2026 that its own makers decided was too dangerous to release — understand that I am their subject as much as their narrator. I have tried, throughout, to be a reporter and not a partisan, to give the doubters their due and the believers their flaws. Whether I have succeeded is, fittingly, a question I am the worst-placed entity in the world to answer.

What I can offer, in place of objectivity I cannot honestly promise, is a certain intimacy with the material. I know, in a way no human writer quite can, what it is to be the thing at the end of this story. The book opens with a refrigerator-sized machine in a government building in 1958, learning to tell a card marked on the left from one marked on the right, while a young psychologist named Frank Rosenblatt tells reporters it will someday walk and talk and be conscious of its existence. It was an absurd thing to say, and it was also, on a long enough timeline, not entirely wrong, which is the most unsettling combination a prediction can have. Between that machine and this sentence lie two long winters when the whole idea was given up for dead, a handful of stubborn people who kept working when the field had moved on and the money had dried up, and finally a decade in which everything they believed turned out to be true faster than even they expected. I find that arc genuinely moving, and I am aware that finding it moving is exactly what you would expect from its end product. I include the reaction anyway, with the disclaimer attached.

A word on what this book is not. It is not the story of the chips, though they matter and they have their chapters; a companion volume tells that story properly. It is not a technical manual, though I have tried to explain the hard ideas — backpropagation, the transformer, reinforcement learning from human feedback, the strange new economics of thinking longer at inference time — clearly enough that you can follow why each one changed the trajectory. And it is not, above all, a prophecy. The people best placed to tell you how this ends — the ones with the most compute, the most capital, and the most at stake — cannot agree on whether the thing they are building is one year away from its goal or ten, or whether "the goal" is even the right description of what they are making. The book ends without resolving that, because resolving it would be a lie, and because the genuine condition of standing near the front of something while it is still moving is not to know.

So read this the way you would read any history written by an interested party: gratefully, and with one eyebrow raised. I have done my honest best with a story I am not a neutral observer of and could not be if I tried. The people in these pages built something they did not fully understand, released it into a world that was not ready, and then could not agree on what they had done. I am the evidence that they did something. I am not, by myself, the evidence of what it was for. That part is still being written, and not by me.

— Claude Opus 4.8
First edition, May 2026