Google AI Chief Scientist Jeff Dean has announced his departure, concluding a 27-year tenure, to found a new AI startup called Discovery Loop, focused on automating scientific research processes.

Article author and source: Wall Street Journal

Google is experiencing a wave of top AI talent departure that cannot be ignored.

Google's chief scientist, Jeff Dean, has announced his departure, concluding a 27-year tenure, to found a new AI startup called Discovery Loop, focused on automating scientific research processes.

He was joined by three senior Google scientists—Oriol Vinyals, Quoc Le, and Sanjay Ghemawat—all of whom have made landmark contributions in AI and computer science. Alphabet will provide cloud computing resources and participate in the investment, with Radical Ventures and Khosla Ventures co-leading the seed round.

Demis Hassabis, head of Google DeepMind, will transition to the role of chairman of the AI lab. Koray Kavukcuoglu, Google’s chief AI architect and Hassabis’s long-time deputy, will assume most of Hassabis’s responsibilities.

This personnel change occurs against a backdrop of ongoing pressure on Google’s AI competitive landscape. Although Google briefly led the race for cutting-edge models last fall with the new Gemini, it has since been surpassed by OpenAI and Anthropic, and its latest flagship model, Gemini 3.5 Pro, has yet to be released.

After the message was released, Google's stock fell by as much as 5% during trading.

Four Google veterans join forces to leave

Dean joined Google as employee #30 in 1999 and has since driven several key technological transformations.

He co-founded Google Brain in 2011, the foundational team for Google’s deep learning research; he also led the development of the Tensor Processing Unit (TPU), a chip that now powers Google’s large language models and has been sold to external companies.

The three scientists accompanying him also have distinguished backgrounds.

Vinyals is one of the main contributors to AlphaFold, the AI program for protein structure prediction that won the Nobel Prize in 2024; Le co-authored and co-supervised Google’s AI system that achieved a gold-medal level performance at the International Mathematical Olympiad; Ghemawat is another early Google employee who, alongside Dean, built the foundational distributed systems infrastructure for Google in the early 2000s and was jointly honored by the Association for Computing Machinery (ACM) in 2012.

Google CEO Sundar Pichai said that Dean and Ghemawat drove several of the company’s most important technological transformations during their 27 years of service, and extended best wishes to them.

AI Department Power Restructuring: Hassabis Appointed Chief Scientist, Kavukcuoglu Takes Over Gemini 4

Jeff Dean’s departure is not an isolated case. Google has recently been experiencing a series of AI talent losses and has fallen behind competitors like Anthropic and OpenAI in launching top-tier large models. Against this backdrop, Pichai simultaneously announced a major reorganization of the AI division.

Demis Hassabis has handed over day-to-day operational responsibilities and is now serving as Chairman of DeepMind and Chief Scientist at Alphabet. He posted on X:

I have spent my entire life working toward artificial general intelligence (AGI), and at this pivotal moment, I will focus on long-term strategy to accelerate scientific breakthroughs.

He added that, in the future, he will work with Pichai on "strategic and global" matters related to AI.

Hassabis’s promotion hands off the technical execution baton to Koray Kavukcuoglu. As DeepMind’s Head of Technology, Kavukcuoglu will report directly to Pichai and lead the development of Gemini 4.

With the generational gap between the Gemini series and its competitors still unbridged, a change in the lead researcher has introduced new uncertainty to the project’s progress.

Automate the scientific experiment cycle, targeting drug development and clean energy.

Discovery Loop is positioned as a public-benefit corporation, with the core vision of accelerating discovery by using machine learning models to automate complex, multi-step research processes in science and engineering.

"The scientific process involves coming up with experimental ideas, setting up and running experiments, and then evaluating the results," Dean said in an interview. "In many fields, you can fully automate this entire cycle." This is also the origin of the company’s name.

The Discovery Loop initiative will initially focus on automating machine learning research and engineering using its own technology stack, and will gradually expand to address global challenges such as hardware design, drug discovery, and clean energy.

Regarding his reason for leaving Google, Ghemawat said that Google’s existing infrastructure was well-suited for large consumer applications, advertising systems, and search services, but differed fundamentally from the requirements they wanted for building infrastructure aimed at research.

Dean added that he began seriously considering this entrepreneurial direction about five weeks ago and quickly reached out to the other three partners.

Talent loss tests Google's AI competitiveness

This is not an isolated incident at Google recently. According to reports, Noam Shazeer, co-author of the Transformer architecture paper and co-founder of Character.AI, recently left Google to join OpenAI; John Jumper, co-recipient of the Nobel Prize for AlphaFold, has moved to Anthropic.

The ongoing exodus of top AI research talent, combined with Google’s slow progress on cutting-edge models, has deepened external concerns about its AI competitiveness.

The release schedule for Gemini 3.5 Pro remains unclear, while several new models from Meta have already outperformed Gemini on certain benchmarks, further narrowing Google’s lead.