The Take — Brin can't fix Gemini by showing up
Sergey Brin has taken direct oversight of Google's Gemini team, and the lab's most delayed flagship, Gemini 3.5 Pro, is reportedly being shelved to make way for a Gemini 4 push. I think Brin's return is the strongest signal yet that Google treats this as a code-red — and the wrong instrument for the fire. Gemini's disease was never a shortage of authority at the top; it was the organization underneath, and no amount of founder presence rewires that.
The pieces of the story are familiar by now — we covered the reports this morning (Brin takes over Gemini as 3.5 Pro is reportedly shelved). QbitAI and the Financial Times report that Brin stepped in urgently, sitting near new DeepMind operations chief Koray Kavukcuoglu as control of Google's AI effort shifts from London back to Silicon Valley. A SemiAnalysis report, relayed by IT之家, says Google may have shelved 3.5 Pro internally to clear the runway for Gemini 4. Google has not commented and still lists the model as "coming soon."
Nothing here is confirmed — but everything here is plausible, and that plausibility is the story.
The argument: the failure was organizational, not strategic
Start with what the Fortune investigation found. Gemini 3.5 Pro was unveiled at I/O in May with a promise to ship "next month." It missed its June deadline, missed a mid-July deadline, and remains unreleased. Engineers trace the delay to a decision last summer to deprioritize coding while Anthropic and OpenAI raced ahead on agents. That is not a hard technical problem that defeated Google's best people; it is a prioritization error made inside the org, by people who are still there, under incentives that are still there.
The rest of the picture is about people, not plans. Staff described 60-hour weeks and burnout. The leadership vacuum was literal — Brin was visible on the floor, Demis Hassabis was not — and then Hassabis moved up to chairman and Alphabet chief scientist. The talent ledger is the part no memo can reverse: Gemini co-lead Noam Shazeer left for OpenAI in June; AlphaFold co-inventor John Jumper went to Anthropic; Jeff Dean is departing after 27 years to found Discovery Loop with three other senior researchers. SemiAnalysis now ranks Gemini eighth on the Artificial Analysis Intelligence Index, behind open-source Chinese labs.
Read that list again and notice what kind of problem it describes. The bindings are people who left, a co-lead seat that is empty, a team asked to work 60-hour weeks for a year, and a second structural reorg — the shift from London to Silicon Valley — landing on top of them. Authority was never the missing input. The people at the top of Google have had the power to change any of this for the past year, and the org drifted anyway. A founder with a floor presence does not change the incentives that produced the drift; it changes the reporting lines, which is the one thing a burned-out team needs least right now.
The counter-case: founder mode has a real record
I want to steelman this properly, because "founder mode" is not a buzzword — it has a genuine track record. Steve Jobs returned to Apple in 1997 and reversed a dying company by fiat: killed projects, merged teams, forced one product narrative. Founders can make the brutal calls no employee can, because nobody can fire them and nobody can outrank them. If the Gemini problem is a strategy call — "coding is priority one again, effective immediately" — a founder can make it in a day, and the whole org will believe it in a way they would not from a new VP.
Brin is also the right kind of technical founder: he built the company's original ranking systems and spent years inside its moonshot side, so he can read model quality numbers himself. And the visible-on-the-floor detail cuts both ways — people work differently when the board-level founder is in the room. If the shelfing of 3.5 Pro is a deliberate portfolio decision — kill the delayed flagship, concentrate everything on Gemini 4 — that is exactly the kind of ruthless call a founder is uniquely positioned to make.
None of that survives contact with the exit list. The coding decision can be reversed in a memo; the trust broken by three missed deadlines takes quarters to rebuild; and the co-lead whose job it was to ship is now at OpenAI. A founder's mandate does not compress that timeline — and this company's history is not on the playbook's side. Larry Page returned as CEO in 2011 with the same founder-mode logic, and the organizational drift that produced this moment still happened on his watch. The failure mode is structural: fragmented teams, competing incentives, process that outlives purpose. Those are exactly the things a founder at the top is least equipped to dismantle personally, because the organization has spent two decades learning to route around the top.
Why the take still holds
Because of the asymmetry between what the signal does and what the problem needs. Brin showing up is a signal the market can price in hours — the clearest possible statement that the board treats Gemini's slide as existential, and that accountability now sits at the founding level. That is genuinely worth something, and I don't want to wave it away. But the fix needs the opposite shape: a stable, accountable operator who owns shipping dates for a year, who can name one plan and be judged on it. A firefighter who owns the floor for a quarter is not that operator — and every additional reorg raises the odds that more senior names follow Shazeer out the door, the one outcome the lab cannot afford.
What would change my mind
Three things, any one of which would materially shift this. First, a ship date: if Gemini 4 launches on a named date with coding capability restored to frontier level, the founder's reset worked, and I'll say so. Second, longevity: if Brin commits to the role for the long term and names a single accountable day-to-day leader for the lab, that is a structural fix, not a firefight. Third, the talent ledger: the strongest signal would be senior names returning to Gemini — or, at minimum, the exodus stopping. And one revision clause: if the 3.5 Pro shelfing turns out to be a deliberate portfolio move that genuinely shortens time-to-market, then Brin is managing a strategy rather than rescuing a project, and my argument dissolves. Until one of those shows up, I'll read the founder's return the way the reports describe it: a rescue mission for a lab whose real problem — the people and process around it — a founder cannot fix by showing up.
Can a co-founder's presence undo a year of drift and three departed leaders — or is the exit list the only number that matters? Tell us in the comments.
Sources: QbitAI · Fortune · The New York Times · Financial Times · IT之家 (via Sina)