Demis Hassabis and Jeff Dean exit Google DeepMind

Demis Hassabis is stepping away from day-to-day operations at Google DeepMind to become Alphabet's Chief Scientist, focusing on strategic and global artificial general intelligence (AGI) matters and Isomorphic Labs. Koray Kavukcuoglu, the former chief technology officer and chief AI architect who has worked with Hassabis for over 13 years, will take over day-to-day management as Senior Vice President. Kavukcuoglu will report directly to Google CEO Sundar Pichai and oversee Gemini model development, frontier AI research, the Gemini app, and Google's AI developer platforms.

Simultaneously, chief scientist Jeff Dean is leaving Google after 27 years. Alongside Google Senior Fellow Sanjay Ghemawat and longtime researchers Oriol Vinyals and Quoc Le, Dean is launching Discovery Loop. This public benefit corporation aims to automate machine learning, science, and engineering, starting with large-scale machine learning experiments. The startup's seed round is led by Radical Ventures and Khosla Ventures, with backing from Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet, which will serve as Discovery Loop's cloud partner.

These high-profile exits follow other notable departures, including Gemini co-lead Noam Shazeer, who left for OpenAI in June, and longtime researcher David Silver, who departed in February. The leadership shakeup comes as Google faces intense pressure from rivals like OpenAI and Anthropic, particularly in coding benchmarks where Google's models have lagged. To close this gap, Google is currently training Gemini 4, a much larger foundation model, though no release date has been announced.

For AI practitioners and developers, this transition marks a critical pivot in Google's ecosystem. With Kavukcuoglu—who helped build DeepMind's deep learning team and drove breakthroughs like WaveNet and DQN—now steering Gemini development, developers can expect a highly technical, research-focused leadership style. Meanwhile, the launch of Discovery Loop introduces a powerful new player in automated research, potentially accelerating the tools available for automated hypothesis generation and machine learning experimentation.