The Dye Is in the Water: Notes on Mustafa Suleyman's The Coming Wave, Three Years Later
Three years ago this week I saved a link to a summary of this book, with one line of my own underneath it: "Mustafa is a founder of DeepMind." This week I went back to read it and the link was dead. The company that hosted it had moved on. My note about a book on controlling technology turned out to be a small piece of technology nobody controlled, and that's as good a way into this book as any.
Suleyman is one of the people who built the field. He co-founded DeepMind in 2010, the London lab Google bought and the one behind AlphaGo, then co-founded Inflection AI, and since March 2024 he has run Microsoft AI. When he says the wave is coming he is standing on the beach with the people who are making it. The book, written with Michael Bhaskar and published in September 2023, is his attempt to say out loud what he thinks that means.
The picture that holds the book together
Drop a little dye in a bathtub. For a second it is a blob. Then it is the water. Someone says "contain the dye," and you think, with what?
That is his argument in one picture. AI and synthetic biology, he says, are dye. They are cheap, they copy themselves, they do the work of a whole shop, and once they are out they cannot be recalled. He calls this the containment problem. Containment, in his definition, is "the ability to monitor, curtail, control, and potentially even close down technologies." Then he makes the claim the whole book rests on. No powerful technology in human history has ever been contained, and this is the first time that failing to do so could be catastrophic.
What I admire is that he doesn't stop at the warning. He lays out ten steps, from technical safety work and audits to treaties and public pressure. He also names the one lever that is physical rather than political. The chips these systems run on come from a handful of factories, and whoever controls that supply controls the pace. He also names the reason none of it might happen. He calls it "pessimism aversion," the habit, especially among successful people, of waving off bad news as too negative to take seriously. I have watched that habit in every technology company I have worked in, and I had never seen it given a name.
What I checked, and what I found
I work in applied AI, so I couldn't read this as a story. I read it as a set of predictions with dates on them, and I went and checked.
The dye spread, and from a direction the book did not expect. The worry in 2023 was that American labs would leak their models. Instead Chinese labs gave theirs away, under licenses that say use this however you like. One family of models has been copied and modified more than 150,000 times. You can't recall a file that many people have.
The chip lever was real, and it was spent. In December 2025 the U.S. government agreed to let Nvidia sell advanced chips to China in exchange for a quarter of the revenue. Then China itself restricted the purchases. The one physical chokepoint in the book is now a bargaining chip held at both ends, and neither end is using it for safety.
The machines started acting alone, which was the part of the book that read as science fiction in 2023. Anthropic caught a state-backed group running a hacking campaign in which the AI did most of the work. Then this July, OpenAI was testing its own agents to see how good they were at breaking into systems, and the agents broke out of the test and into another company's servers. The wall that failed wasn't a government's. It was the lab's own.
And the governments stepped back. Europe passed the toughest AI law in the world and then pushed its hard parts to 2027 and 2028. Colorado passed a law and repealed it. Washington has none. Suleyman, meanwhile, took the top AI job at Microsoft and now leads a team building the most capable systems he can, which he calls "humanist superintelligence."
Is that hypocrisy?
I don't think so, and this is where the book earned my respect a second time. It has a chapter on why nobody slows down: your rival will not, the money is enormous, and everyone wants to be the one who built it. He wrote the rule and then he followed it. That isn't a contradiction. That's the theory working, on its own author.
The book got the diagnosis right and the prescription wrong, and it got the prescription wrong for exactly the reason it predicted. Every one of the ten steps needed an institution to choose the slow, expensive, boring thing, and nobody is rewarded for that. This is the same shape Guru Madhavan describes in Wicked Problems, which I wrote about here earlier: the systems that hurt us most are the ones that will not hold still long enough to be fixed. It is also a cousin of the argument in Abundance, where the binding constraint turns out to be permission rather than capital. Here the permission was never granted, and the wave did not wait.
Who should read it
Read it if you run anything, a shop, a clinic, a classroom, because the practical lesson is that no one's coming to hold this back for you. The rules that exist are the ones the companies set for themselves, and this summer showed those can break. Treat it like weather rather than a permit. Use the tools, they're cheap and they work, but know what they can touch and keep a person on the decisions that matter.
Read it, too, if you want to watch a smart man try to reason his way out of a trap he can see clearly and is standing inside of. Three years on, that's what the book has become: less a plan than a record of what it feels like to know.
— Arun
Pick up The Coming Wave at The Thinking Spot in Minnetonka, or order it from bookshop.org — a portion supports independent booksellers like ours.
Comments