Jeff Dean and Top Google Researchers Depart to Launch 'Discovery Loop'

Jeff Dean, the engineer widely regarded as the most influential technologist in Google's history, is leaving the company after nearly 27 years — and he is not going alone. On August 5, Dean confirmed he is departing alongside three of Google's most decorated researchers to launch Discovery Loop, a public benefit corporation that aims to do something audacious: automate the scientific method itself.

Joining Dean, who will serve as CEO, are Sanjay Ghemawat, a Google senior fellow and one of the architects of the modern internet's infrastructure; Quoc Le, a founding member of Google Brain; and Oriol Vinyals, a senior DeepMind research scientist who most recently co-led the Gemini program with Dean. Between them, the four hold authorship credits on some of the most consequential work in computing — MapReduce, TensorFlow, sequence-to-sequence learning, AlphaFold, and the Transformer architecture that underpins today's large language models.

The departures land in the middle of an early-August reorganization at the top of Google's AI operation. Demis Hassabis has moved to Chair of Google DeepMind and Chief Scientist of Alphabet, Koray Kavukcuoglu has been elevated to SVP running Gemini, and London coding teams are relocating to Mountain View. Against that backdrop, the exit of four researchers of this caliber is less a routine talent loss than a signal about where the frontier of AI is heading next.

What 'automating the research loop' actually means

Science, at its core, runs on a loop: form a hypothesis, run an experiment, evaluate the result, and feed what you learned into the next hypothesis. That loop has always been executed by humans, in series, and it is slow. Discovery Loop's bet is that AI can run it instead — proposing experiments, executing them, learning from the outcomes, and iterating recursively, thousands of experiments in parallel rather than one at a time.

"The next great frontier for AI is to go beyond answering questions and to begin making discoveries," the founding team said in a joint statement. "By fundamentally accelerating how engineering and scientific discovery are conducted, we can deliver the benefits of transformative technologies to the world far sooner."

The team plans to start close to home, automating machine learning research and acting as its own first customer — a version of recursive self-improvement, in which AI is used to build more capable AI. From there, according to seed investor Radical Ventures, the company intends to point these systems at the physical world, targeting "grand challenges in engineering, medicine, materials design, and clean energy."

Dean framed the opportunity plainly. "We think there is opportunity for AI to more fully automate what has traditionally been a very human-intensive experimental loop," he told The New York Times. "You will get both a higher quantity and a higher quality of experiments, and that will lead to scientific breakthroughs and advances."

The money, and Google's unusual role

Discovery Loop's seed round is being co-led by Radical Ventures and Khosla Ventures, with participation from Kleiner Perkins, Lightspeed, Doerr Capital, and — notably — Alphabet itself. Google is not just watching its stars walk out the door; it is a founding investor and the startup's cloud partner, committing compute for the venture's first year. The funding total and valuation were not disclosed.

That arrangement is telling. Rather than treat the departures as a defection, Google has structured them as something closer to a spin-out it retains a stake in. For Alphabet, it is a hedge: if automated discovery becomes the next platform shift, the company keeps both equity and a cloud customer. For Discovery Loop, Google's compute removes the single largest cost barrier facing any frontier lab in its first year.

The symbolism is harder to manage. Dean was Google's 30th employee, hired in 1999, and helped build the crawling, indexing, and query-serving systems that made Search work — then helped build the AI research organization that produced Gemini. Losing him, Ghemawat, Le, and Vinyals in a single week strips out an extraordinary concentration of institutional knowledge at exactly the moment Google is reshuffling its AI leadership. Radical's partners called the group "legends in the field of artificial intelligence" and argued that "the scientific method is arguably the most powerful algorithm humanity has ever invented" — a framing that captures both the ambition and the hype risk of the venture.

What to watch next

The open questions are concrete. Can the founders show that an automated loop produces real, reproducible discoveries — starting with ML research — rather than a faster treadmill of low-value experiments? Watch whether more Google and DeepMind researchers follow the four out the door, how the London-to-Mountain-View consolidation reshapes Gemini under Kavukcuoglu, and whether the public benefit structure holds as investors seek returns. Discovery Loop is hiring an in-person team in Palo Alto. The first real test will be what it discovers.