Case fileKevin Roose2026Farrar, Straus and Giroux
The Inside Story of the Race to Create an Artificial Superintelligence
A reporter's record of the decade when a few labs set out to build a mind, and argued all the way about whether they should.
They feared what they were building, and they built it
The core idea
Kevin Roose is a technology columnist at The New York Times and the author of Futureproof. The AGI Chronicles is his reported history of roughly the last decade of artificial intelligence, told through the companies racing to build artificial general intelligence: OpenAI, Anthropic and Google DeepMind. It draws on more than 150 interviews with people inside those labs, along with internal documents and research memos. Roose follows the labs where language models gained new abilities quickly, the split that led Anthropic's founders to leave OpenAI, Sam Altman's removal and return as OpenAI's chief executive, Google's push to catch up, and the officials trying to regulate a technology that moves faster than they do. Along the way he explains in plain terms how these models are built and trained, and he argues that the stakes are high enough that there may be little time left to make them safe.
The book reads as a history, not a forecast. Roose sets out who decided what, and when, so that a reader can see how a research curiosity became the most watched race in technology. The publisher frames it as this century's version of the Manhattan Project: a small group of people building something they believed could change everything.
What holds it together is a tension the builders never resolved. Many of them thought the technology could cure diseases, and many thought it could go badly wrong. Often the same people thought both. The race went on anyway, and the book is an attempt to explain why.
The three labs in the file
Thread 01
Plain explanations of how language models are built and trained, and why next-word prediction turned out to be so powerful.
Thread 02
Rivalries, splits and a board crisis, drawn from more than 150 interviews and from internal documents.
Thread 03
Cures on one side, catastrophe on the other, and Roose's view that the time left to make these systems safe may be short.
Interactive · the stance map
Six statements put you on two lines the book keeps returning to: how big the stakes are, and how fast this should go. Save a reading today and come back in a few months. Your saved readings become your own small chronicle, so you can see whether you moved.
Your position
Answer all six to place yourself on the map.
Today you read as
No readings saved yet. Save one today, then retake the map in three months.
HourLife's statements and labels, not the book's, and not a test with a right answer. Nothing you choose or type leaves this browser.
Concept anatomy
Two sketches to hold the book in your head. On the left, the cycle of building, releasing and answering that the labs were caught in. On the right, the plain steps of making a model, which the book explains for readers with no technical background.
Each lap of the loop makes the next one faster.
The process in four steps, simplified.
Step 1 · Gather
Books, websites, code and more. Where that text came from, and whether it was fair to use, is still argued over, and the book does not skip it.
Step 2 · Pretrain
Over and over, across the whole collection. The hope was that getting good at this would force the model to learn how the world works, not only how sentences end.
Step 3 · Shape
Feedback from people and written rules turn a raw text predictor into an assistant that follows instructions and refuses some requests.
Step 4 · Test and ship
Labs test for risky abilities before release. How long that testing should take is one of the fault lines the race runs along.
Community insights
Entry 01HourLife takeaway
This is the puzzle at the centre of the book. Some builders hoped the technology would cure diseases, some feared a catastrophe, and plenty held both views at once. Roose's reporting tries to explain how people who took the risk seriously still ended up speeding the race along.
Entry 02HourLife takeaway
The book walks through how the field moved from narrow systems built for single jobs, such as translation, to models trained to predict the next word across huge amounts of text. The bet was that doing this well would force a model to pick up a working picture of the world. Understanding that bet makes the rest of the story easier to follow.
Entry 03HourLife takeaway
Roose covers the split that led Anthropic's founders to leave OpenAI. A group that worried about how fast things were moving started its own lab, which then had to compete at the frontier to have a say. It is a clear case of how caution can still add speed.
Entry 04HourLife takeaway
The book covers Sam Altman's removal as OpenAI's chief executive and his return shortly after. Whatever you make of the people involved, the episode showed how much of the field's governance rested on a few boards and a few relationships.
Entry 05HourLife takeaway
Roose follows Google's effort to catch up once others had shown what large language models could do. Seen together, the labs behave less like separate projects and more like one race, where each move changes what the others feel they must do next.
Entry 06HourLife takeaway
The book includes the officials who tried to get a handle on the field. Their problem is timing: rules take years, and the systems they describe can change within months. Keep that gap in mind when you read about any new law or agreement.
Entry 07HourLife takeaway
Roose spends time in the San Francisco AI scene, where people who want to speed up and people who fear the outcome live side by side. Both camps take the technology seriously. Their argument is about pace, control and who decides.
Entry 08HourLife takeaway
This is the author's conclusion, and readers will weigh it differently. You do not have to share it to find the book useful. It is a strong case for following the question closely rather than treating it as science fiction or as marketing.
Entry 09HourLife takeaway
Much of what people say about AI is either a pitch or a warning. A dated account of what the labs said, built and did gives you a baseline. When the next big announcement arrives, you can ask how it compares with what happened before.
Action steps
In two sentences, write what you think AI will be able to do in five years, and how sure you are. Date it and keep it where you will find it again. It is the first entry in your own chronicle.
List ten things you do at work or at home each week. Mark each one: a current AI model does it well, does it badly, or cannot do it. Keep the list to compare later.
Give a current AI model a real task from your own field, one where you can judge the answer. Note where it is good and where it is confidently wrong.
Read one plain explanation of how language models are trained, such as the technical passages in the book. Then explain next-word prediction to a friend in three sentences without notes.
Pick three people to follow on AI: someone who builds it, someone who works on its safety, and someone who doubts the hype. Drop one source that only ever cheers or only ever panics.
Once a month, write three lines: what changed, what you tried, and whether it changed your view. A year of entries will tell you more than a year of headlines.
Every three months, run the stance map on this page again and save the reading. Look at whether you moved, and write one line about why.
Before you pass on a big AI claim, ask who is making it and what they gain if you believe it. That goes for warnings as much as for promises.
Once a month, talk about AI with someone whose view is far from yours. Try to state their position well enough that they would agree with your summary.
Many of the people building it feared it, and built it anyway. That is the story, and we are all in it.HourLife note after Kevin Roose
Pick setup actions and daily habits from The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence, choose a start date, and download a printable checklist with real calendar pages.
Kevin Roose is a technology columnist at The New York Times and the author of Futureproof. The AGI Chronicles is his reported history of roughly the last decade of artificial intelligence, told through the companies racing to build artificial general intelligence: OpenAI, Anthropic and Google DeepMind. It draws on more than 150 interviews with people inside those labs, along with internal documents and research...
HourLife’s key takeaways include: Many of the people racing to build AGI believed it could be dangerous, and they kept building it. A model trained only to guess the next word turned out to learn far more than anyone planned. Disagreements about safety did not slow the race. They created new competitors.
It's a strong pick for readers exploring Digital Literacy and History Worth Knowing. HourLife distills its core idea into key insights and practical actions worth trying.
Write down your AI forecast and date it — In two sentences, write what you think AI will be able to do in five years, and how sure you are. Date it and keep it where you will find it again. It is the first entry in your own chronicle.
About five minutes. The HourLife summary distills The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence into its core idea, 9 community insights, and 9 practical actions you can apply right away.
One idea from one book and one small habit from another, every week or once a month, on the day and around the hour you pick, where you are. Nothing else, and one link stops it.
Print it, pin it, post it. Ways to take The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence off the screen and into the world.
The core idea, key takeaways, and how to apply The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence, as one clean, readable page.
Read summaryEvery action from this page as a printable to-do list, with a 7-day tracker.
Download PDFA shareable card with the book and its top-voted insight, for saving or posting.
The summary card and every quote card in six sizes: Instagram feed, Story, Pinterest, YouTube thumbnail, phone wallpaper, and link preview.
Save the insights that land, build the habits that stick, and keep what you learn from The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence working long after you close the page — inside HourLife.
