Google is facing its most consequential leadership churn in AI since ChatGPT forced Sundar Pichai to declare "code red" three years ago.
What happened?
1. Jeff Dean exits: 30-year Google veteran, chief scientist, and architect of the transformer, is leaving to co-found Discovery Loop - an AI startup focused on automating scientific and engineering research. Sanjay Ghemawat, Quoc Le, and Oriol Vinyals are joining him. Google is backing it as investor and cloud partner. Dean will be CEO.
2. Demis Hassabis steps down: CEO of DeepMind is moving to become Alphabet’s chief scientist and chairman of DeepMind. He’ll still run Isomorphic Labs.
3. Koray Kavukcuoglu promoted: DeepMind’s CTO becomes senior VP and takes charge of Gemini model development.
The Analytical Problem for Google-
Google invented the core tech. The transformer. The large language model stack. Gemini. DeepMind. Yet it lost the first-mover advantage in products. Why? The last 2 years show a pattern: the AI race stopped rewarding research seniority and started rewarding "shipping speed". OpenAI built developer tools and coding assistants first.Anthropic built enterprise trust first. Google, despite having the deepest research bench, struggled to convert that into products developers reach for first.Losing Dean + Hassabis stepping back signals that Google is losing senior technical talent at the exact moment it needs to execute. A steady trickle has already gone to OpenAI and Anthropic.
Google’s New Strategy: Hedge, Don’t Fight-
Instead of trying to keep everyone in-house, Google is backing spin-outs. Discovery Loop is the example. Google Ventures + Alphabet have done this before with smaller bets. But backing a chief scientist is a much bigger public version of the same instinct.Why? Because frontier AI is increasingly happening outside a trillion-dollar bureaucracy. Faster, riskier, capital-intensive. If it happens outside, Google would rather own a stake than watch from the outside.Isomorphic Labs is also doing adjacent work on drug discovery under Hassabis, outside Google’s main walls. Discovery Loop complements that.
The Real Test-
The article ends on this: Nobody knows if Discovery Loop will work. It’s capital-intensive and slow to show returns. But Google’s real test is not this startup. It’s whether Google can ship something in the next few quarters in coding, agents, and whatever comes after chat, that makes developers choose Gemini first. Not as an alternative to Anthropic or OpenAI, but as the default.Koray Kavukcuoglu now owns that job. Hassabis becomes Alphabet’s most senior technical voice at a bad moment.
Bottom Line-
This is a structural shift. Google is moving from "one unified DeepMind" to a "looser constellation of Google-backed ventures orbiting its cloud". It’s an admission that research alone isn’t enough. The next 6-12 months will decide if Google can turn research leadership into product leadership.