Google AI Pioneers Jeff Dean and Three Senior Researchers Exit to Launch Discovery Loop
Foundational AI architects leave Google to build a public benefit corporation automating the scientific method.
Four of Google's most influential AI architects have departed the company to launch Discovery Loop, a venture designed to automate the entire scientific process. The exit of Jeff Dean, Google's chief scientist and 30th employee, marks the end of a 27-year tenure and signals a strategic shift in how artificial intelligence is applied to fundamental research.
Dean left the company on August 5, 2026, alongside three other senior veterans: Sanjay Ghemawat, a Senior Fellow; Oriol Vinyals, formerly the VP of Research at Google DeepMind; and Quoc Le, a co-founder of Google Brain. The group has incorporated Discovery Loop as a public benefit corporation. According to the founding team, the venture's primary objective is to move beyond answering questions and begin making original discoveries by automating the proposing, running, and evaluating of scientific experiments at scale.
A Leadership Transition at Google
These departures coincide with a broader leadership reshuffle within Google's AI divisions. Demis Hassabis has stepped back from the day-to-day management of Google DeepMind to assume the roles of chair and Alphabet's chief scientist. In his stead, Koray Kavukcuoglu has taken over as the Senior Vice President of Google DeepMind.
Despite the exit of these key figures, Google remains closely tied to the new venture. The company is serving as a founding investor and Cloud partner, providing the critical compute power necessary for Discovery Loop's operations throughout its first year.
The End of the Human Bottleneck
This move represents a significant migration of foundational talent toward a philosophy where AI is not merely a tool for researchers, but the researcher itself. By targeting recursive self-improvement and the automation of the scientific method, Discovery Loop aims to remove the human bottleneck in innovation. The implications extend across multiple high-stakes fields, including machine learning research, chip design, biology, and materials science.
Quoc Le highlighted the potential for the technology to disrupt current AI paradigms, stating, “I’m very excited about automating machine learning. It might be that we will discover a different transformer architecture.”
The Path to Autonomous Discovery
As Discovery Loop begins its operations, the industry will be watching whether the team can successfully translate the theoretical automation of the scientific method into tangible breakthroughs. While the partnership with Google Cloud provides a stable technical foundation, the startup's success will depend on its ability to generate verifiable scientific results without human intervention. The venture now stands as a primary test case for whether AI can independently drive the next generation of scientific discovery.