Brian Reed: Eliza, something's been bugging me since August fifth — a company does not remove a Nobel Prize winner from running its most important division unless something has already gone wrong.
Eliza Ward: Wait, is this the Hassabis thing? Because I saw the headline and — okay, catch me up.
Brian Reed: Demis Hassabis is out as CEO of Google DeepMind. The CEO title for that division is just — gone entirely. Koray Kavukcuoglu steps up to SVP, gets day-to-day operational control of Gemini and all AI model development, reporting straight to Sundar Pichai. Hassabis becomes chair of Google DeepMind and chief scientist of Alphabet, which is a real title but not an operating one.
Eliza Ward: And the stock fell on the announcement.
Brian Reed: It did. Which — investors don't usually tank a stock because a company elevated someone to chief scientist. They tank it because they read something else in the move. And I think what they read is the same thing I read: Gemini 3.5 Pro was expected in June, it's now August, it still hasn't shipped. That's two months of slippage on the model Google needs to actually compete with OpenAI and Anthropic.
Eliza Ward: So the delay comes first, and the reshuffle is the response.
Brian Reed: That's how it reads. And then Jeff Dean — I mean, he's been Google's chief scientist for years — he announces he's leaving to start a new AI company, same week, with three other senior veterans. So you've got a product delay, a leadership overhaul, and a departure wave hitting simultaneously. The thing I genuinely can't answer yet is whether Kavukcuoglu actually has the authority to move faster, or whether he's just the new person holding the bag.
Eliza Ward: That's the whole question, right — authority versus title.
Brian Reed: And the authority-versus-title gap is actually easier to understand once you know what the 2023 merger actually was. Google Brain and DeepMind became — on paper — one thing called Google DeepMind. But the researchers didn't move. Brain people stayed in California. DeepMind people stayed in London. Same org chart, two different kitchens on opposite sides of the building, still cooking from their own recipes.
Eliza Ward: Wait, so the merge was organizational but not operational.
Brian Reed: That's it, exactly. You've got one dining room, two kitchens, and every dish that needs both teams has to travel across an ocean to get plated. It slows everything. And when a model like Gemini 3.5 Pro — a flagship release — needs components from both sides, the coordination cost just... compounds. June comes, nothing ships.
Eliza Ward: Hold on. June 2026. That missed date — is that when this became undeniable to people outside the building?
Brian Reed: That's the moment. Because before June, the friction inside Google DeepMind is, I mean — it's internal noise. After June, investors and rivals can see it. And the context that matters here is ChatGPT launching in late 2022 caught Google completely off guard. That's what started all of this — the pressure, the merger, the acceleration. Google's been playing catch-up to OpenAI and Anthropic for three and a half years, and then the flagship response to that race just... doesn't arrive.
Eliza Ward: So ChatGPT fires the starting gun in 2022, Google scrambles, merges Brain and DeepMind in 2023 to go faster — and the merger actually made it harder to ship.
Brian Reed: That's the uncomfortable read, yeah. And so when Gemini 3.5 Pro is still sitting in the kitchen in August, that's not a leadership style problem. That's an org structure that couldn't route the decision fast enough. A new SVP title doesn't automatically fix that — but it's what actually triggered the reshuffle. Not a strategy meeting. A missed date.
Eliza Ward: Right — and the question left dangling is whether Kavukcuoglu has any real mechanism to close that gap, or whether the two kitchens are still cooking separately.
Brian Reed: The two kitchens are still there — but here's the part that breaks the clean story Google's telling. Hassabis's new Alphabet chief scientist title, nobody can confirm it comes with dedicated staff or a budget. And without that, you're not repositioning him to frontier science. You're just moving him off the product line.
Eliza Ward: Wait — no dedicated staff confirmed?
Brian Reed: Not that anyone's reported. And so the thing that tells me something real — Hassabis keeps Isomorphic Labs. That's Google's AI drug-discovery subsidiary. He's still running that. Which is interesting because that's not Gemini. That's not the competitive consumer AI race. That's a completely separate scientific mission.
Eliza Ward: Couldn't that actually be smart, though? Like — he won a Nobel Prize. Reuters reported he'd explicitly targeted that as a DeepMind business goal. Maybe you want that brain on the longer-horizon bets, not the shipping schedule.
Brian Reed: I mean — yeah, the division-of-labor story is coherent on paper. But here's what's hard to square with it: Kavukcuoglu was already Chief AI Architect in 2025, already reporting to Pichai, already working on product integration. This SVP elevation isn't a new idea. It's Google finally writing down what was informally true. Which means the structure they're announcing as the fix was basically already running — and Gemini 3.5 Pro still didn't ship in June.
Eliza Ward: So the org chart caught up to reality, and reality still had a two-month delay.
Brian Reed: Right. And picture a senior Gemini engineer — it's Sunday evening, she gets a calendar invite from Kavukcuoglu for a Monday product review. She scans the attendees. Hassabis isn't on it. That's the exact moment the new power structure stops being an announcement and becomes operational. The person who built this place is now on a different calendar entirely.
Eliza Ward: And Sergey Brin's been increasingly hands-on in Google's AI work — so the pressure to subordinate research to product was already coming from above before any of this was official.
Brian Reed: Exactly — the org chart just made visible what Brin and Pichai were already pushing for. And the question that actually makes this more complicated — whether Jeff Dean's departure and the rest of that summer talent wave is genuine brain drain or people who were already sidelined by this structure. That's the part we need to pull apart.
Eliza Ward: Dean's the part that stands out to me — because that's not a sidelining story, that's four people walking out the door together to start something new. Jeff Dean, longtime chief scientist, plus three other senior veterans. That's a founding team, not a coincidence.
Brian Reed: And Anthropic and OpenAI were actively recruiting Google AI staff that whole summer — 2026. So Dean's group leaving to start their own thing is happening inside a broader pull where the best outside offers in years are landing in Google DeepMind inboxes simultaneously.
Eliza Ward: Wait — actively recruiting, meaning structured offers, not just passive poaching?
Brian Reed: That's what the reporting says, yeah. Both Anthropic and OpenAI pulling star staffers specifically that summer. Which — I mean, the timing isn't random. They see the Gemini delay, they see the org turbulence, and they move. But what actually complicates the clean 'Dean was already peripheral' counter-argument: Google Brain's core leadership has now largely scattered since the 2023 merger. Not just Dean. The institutional knowledge that built Google's AI foundation over a decade — it's dispersed. You can argue Dean's specific group wasn't touching Gemini's delivery. Fine. But Hassabis is supposedly now freed to pursue frontier bets. Who actually knows how to build those bets if the Brain lineage has been leaving in pieces for two years?
Eliza Ward: That's — okay, that's the part I want to separate. Because 'peripheral to Gemini' and 'irrelevant to what comes next' are completely different claims.
Brian Reed: Exactly. Imagine a researcher who's been running long-horizon safety architecture at Google Brain since 2019. She's not on the Gemini 3.5 Pro sprint. She's not blocking the June ship date. But she's the person Hassabis would call when the frontier bet he's supposedly now free to pursue needs someone who's already done seven years of foundational groundwork. If she took an Anthropic offer in July, that call doesn't get made.
Eliza Ward: So the reshuffle answers the near-term question — who ships Gemini — and leaves the medium-term question completely open.
Brian Reed: The org chart change is silent on retention two years out. And that's actually what makes the brain drain framing contested rather than settled — it's not wrong, it's just aimed at the wrong timeline. Real impact may not show up in Gemini's next release. It shows up in whatever Google needs to build to still be competitive in 2028.
Eliza Ward: Which brings me back to where you started — a company doesn't move a Nobel Prize winner off operations unless something already broke. That's still true. What's different now is I understand what broke. Not Hassabis. The org that couldn't turn his research into a shipped product.
Brian Reed: And Kavukcuoglu gets 18 months, roughly, before the question answers itself. Either Gemini 3.5 Pro ships and competes, or it doesn't. No org chart has ever fixed a culture — but that's the bet Google is placing right now.
Eliza Ward: An honest place to leave it.
