Eight People Started the AI Era. Now We Buy and Sell Them One at a Time.

Sometime in 2017, eight people at Google finished a paper and gave it a title that sounds, in hindsight, almost arrogant: Attention Is All You Need.

They weren’t famous. Most weren’t senior. A couple were passing through — an intern, a part-time academic. They were working on a narrow problem, machine translation, the kind of thing that earns you a conference slot and a pat on the back. What they actually did was lay the foundation for nearly every AI system you’ve touched this year. ChatGPT, Gemini, Claude, the tool quietly drafting a colleague’s emails — all of it traces back to the architecture those eight described. The Transformer.

I’ve been thinking about that this week, because the last of those eight just left Google. Noam Shazeer — who’d left once, been lured back, and has now left again — joined OpenAI. His departure quietly completed something nobody seems to have said out loud: every author of that paper is now gone from the company where they wrote it. All eight. Scattered.

Where eight people go

Trace them and it reads less like an org chart than like seeds blown off a dandelion.

Ashish Vaswani, the lead author, co-founded Essential AI; Niki Parmar went with him. Aidan Gomez built Cohere. Llion Jones went to Japan and started Sakana. Illia Polosukhin co-founded NEAR. Łukasz Kaiser landed at OpenAI. Jakob Uszkoreit did something quietly telling — he walked away from language models entirely and pointed the same ideas at biology. And Shazeer bounced out, got bought back at a price that made headlines, and bounced out again.

Here’s what gets me. These people made the rarest thing there is — not a clever product, but a genuinely new idea, the kind that doesn’t come from a roadmap or a quarterly goal. The kind you can’t order up no matter what you spend. And the response of the industry they created was to break them apart and price them individually. They’re recruited, poached, retained, re-acquired. They have market values. Somewhere there are spreadsheets with their names and a number beside each.

There’s something both touching and a little sad in that. Eight people were, for one stretch of months, in the same place working on the same strange bet. That kind of room isn’t repeatable. You can’t reconstitute it by hiring the same names back — which, and this is the part that should give us pause, somebody essentially tried to do, at enormous cost, and it still didn’t hold. The lightning was in the bottle once. After that, the bottle is just glass.

We turned a moment into a market

I don’t think the real lesson here is about retention packages, and I’m wary of the version of this that becomes advice for executives. The honest observation is simpler, and more human, than that.

We’ve taken the people capable of original invention and made them the most tradeable resource in the business — line items moved between buildings. The money is staggering, and the staggering money is exactly what makes it feel hollow. You don’t put a nine-figure price on something you understand. You put it on something you’re afraid you can’t get any other way. The price is a kind of confession: we don’t know how to make more of these people, so we fight over the few who exist.

Meanwhile the thing they built escaped on day one. That was the point of publishing it. The Transformer became everyone’s the moment it went online, free, while its authors got carved up and bid over. The idea was given away; the people were put up for auction. If you’d had to guess in advance which would turn out to be precious, you’d probably have guessed wrong.

The part that stays with me

There’s a detail I can’t shake. Asked on a panel what comes next, several of these authors openly hoped the Transformer gets replaced — that the thing they’re known for turns out not to be the best we can do. The people who lit the fire are already looking past it, while the rest of the world stays entirely dependent on it. One of them left language models for protein design. That’s not a man chasing a paycheck. That’s someone who already heard the next thing calling.

So that’s the shape of where we are, and I don’t think it’s mainly about Google, or one deal, or even about AI. Eight people made something that belongs to everyone. They no longer work together, no longer work where they made it, and some would be glad to see it surpassed. They’re spread across the map now, each carrying whatever it was — call it taste, restlessness, the knack for being right when everyone else is sure you’re wrong — that made the room work. And the rest of us are left buying and selling them at ever-higher prices, hoping whatever happened once can be made to happen again on command.

It probably can’t. That tends to be how it goes with the things that matter most. They arrive quietly, from people nobody was watching yet — and we only learn their worth by the size of the hole they leave.