Google redraws its AI leadership: Hassabis moves to Chair and Chief Scientist, Dean leaves to found Discovery Loop

On 5 August 2026 Google announced a new structure for its AI leadership in a post signed by chief executive Sundar Pichai and DeepMind co-founder Demis Hassabis. Hassabis steps back from the day-to-day running of Google DeepMind and takes on the roles of Chair of Google DeepMind and Chief Scientist of Alphabet; he will continue to lead Isomorphic Labs. In the official text Hassabis writes: "It's critical that we collectively get the next steps right to ensure this all goes well for humanity."

On the operational side, Koray Kavukcuoglu, until now CTO of Google DeepMind, becomes SVP of Google DeepMind reporting directly to Pichai: his remit covers Gemini model development, Frontier AI research and the Gemini app and developer teams. Parts of the press described the move as "CEO of Google DeepMind", but the wording that stands is the "SVP" used on Google's corporate blog.

In the same announcement Pichai confirms Jeff Dean's departure after 27 years: "After an incredible 27-year run, Jeff Dean is at a moment where he wants to try something new." Dean and Sanjay Ghemawat are starting "an independent public benefit corporation to accelerate breakthroughs in machine learning, science and engineering", with Google as "founding investor and Cloud partner". The official announcement does not name the new company. According to the company's own communications and the trade press, it is Discovery Loop, Inc., a Delaware public benefit corporation announced on 5 August 2026 and led by Dean as chief executive; the co-founders named are Dean, Ghemawat, Quoc Le (a founding member of Google Brain) and Oriol Vinyals (senior research scientist at Google DeepMind), and all four are leaving Google (sources: Radical Ventures, TechCrunch, 9to5Google).

Discovery Loop's initial round is co-led by Radical Ventures and Khosla Ventures, with participation from Lightspeed, Kleiner Perkins, Doerr Capital and Alphabet. None of the sources consulted states the amount raised or the valuation. Radical Ventures sets out the thesis: start from automating machine learning research and engineering — with the company as its own first customer — and extend into "engineering, medicine, materials design and clean energy"; the stated goal is to build systems that run "thousands of experiments in parallel". Dean sums up the operating assumption: "You will get both a higher quantity and a higher quality of experiments, and that will lead to scientific breakthroughs and advances." Radical's note also mentions an in-person team in Palo Alto and the start of hiring.

The reorganisation comes after the 2023 merger of DeepMind and Google Brain (now Google DeepMind, responsible for the Gemini model family) and involves people who built Google's distributed infrastructure — Dean and Ghemawat (MapReduce, Bigtable, Spanner, Google File System) and, for Dean, the launch of Google Brain and TensorFlow. The wider frame is the rise of scientific-research automation as an explicit investment line, alongside announcements from the same weeks such as OpenAI's results formalised in Lean and the US public programme Genesis Mission.

What has not been stated: - the amount and valuation of Discovery Loop's round; - the actual date the four researchers leave Google; - how Oriol Vinyals's work on the Gemini models will be redistributed; - Google's announcement does not name Discovery Loop, Quoc Le or Oriol Vinyals, nor does it detail the scope of Alphabet's investment (that information comes from Discovery Loop and the investors instead); - the effect of the reshuffle on the Gemini product roadmap cannot be verified as things stand.

A personal note: Alphabet is rebalancing the internal axis between frontier research and product delivery while funding, as founding investor and Cloud partner, a public benefit corporation focused on automating science. The interesting signal is operational rather than narrative: whether the promise of "thousands of experiments in parallel" turns into reproducible results is something we will see over the coming months — far from the superlatives, close to the verifiable metrics. — Olya