Jeff Dean is leaving Google after almost 27 years to start a new company called Discovery Loop, a move that pulls one of the company’s most prominent AI engineers into a startup built around automating scientific and engineering work. He is founding it with Sanjay Ghemawat, Oriol Vinyals and Quoc Le, all former Google AI leaders.
The timing matters because Dean had just been onstage on July 25 at Y Combinator’s Startup School in San Francisco, speaking to 6,000 would-be founders before his departure became public. There, he told the audience the best startup ideas come from work founders care about and that matters in the world, and he said he was drawn to an automated version of the scientific method.
Dean is not leaving from the margins. He cofound Google Brain, serves as chief scientist of Google DeepMind and Google Research, and is technical colead on Gemini. He and Ghemawat wrote major pieces of Google’s computing and search infrastructure, which is part of why the move lands as more than another senior executive departure. Google will also take a stake in Discovery Loop, giving the search company a financial interest in a startup that could eventually compete with the systems it helped build.
What makes the new company unusual is how quickly it came together. The idea for Discovery Loop surfaced only a few weeks before launch, even though the four founders are long-time colleagues who vacation together. That gives the startup a rare mix of speed and shared history: a fresh company, but one built by people who have already spent years solving problems together inside Google.
Discovery Loop’s first task is narrower than its ambition. Its autonomous experiment loops will initially be aimed at improving its own machine learning algorithms, with the company planning to use better software as the core of an AI system that can later move into chip design, biology, drug discovery and material design. Quoc Le said he is excited about automating machine learning, while Oriol Vinyals said the team expects to generalize once the core discovery loop is in place.
That is the real bet behind the departure. Google loses a foundational engineer in the middle of intense AI model competition, while Dean and his cofounders are wagering that the next leap will come from systems that can run thousands of experiments on their own and learn from the results faster than people can. If they are right, Discovery Loop is not just another AI startup. It is an attempt to turn the scientific method into software.
