Subvolts take Google's AGI chief details the path to Gemini and the frontier AI race. Sep 3, 2026 - Developers
- Learn about the engineering breakthroughs and ambitions behind Google's frontier models like Gemini and the development of coding agents.
- Business
- Understand Google's strategic approach to building AGI and how user collaboration shapes their AI development process.
This episode features a top leader from Google DeepMind discussing their AGI strategy and Gemini's progress, offering direct insight into a major lab's frontier work.
Koray Kavukcuoglu on frontier models, coding agents, and building AGI
Audio streams from Google AI. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Google DeepMind's Koray Kavukcuoglu discusses the journey from early reinforcement learning to Gemini and the pursuit of AGI. He shares insights on Gemini 3.7 Flash's breakthrough, Gemini 4's pre-training, and the challenges of frontier AI research. The episode also touches on co-building AGI with users and the unique environment at Google for this work.
Google's top AI architect talks about how they are building super-smart computer systems called AGI. He explains the progress with their Gemini AI and why building these advanced tools is a big challenge that requires constant innovation.
This episode offers a deep dive into the strategic thinking behind Google's AGI ambitions from a key leader. It's for AI researchers, engineers, and strategists interested in the practicalities of building frontier models and agents. Kavukcuoglu's perspective connects directly to the current race among labs like OpenAI, Anthropic, and Meta to achieve AGI, highlighting Google's approach to development and user collaboration.
- HostLogan Kilpatrick
- GuestKoray KavukcuogluSVP and Chief AI Architect, Google DeepMind
- Guest: Koray Kavukcuoglu, Google DeepMind SVP and Chief AI Architect
- Show: Google AI: Release Notes
- Length: 27 minutes
- Topics Covered: Journey from DeepMind's early RL to Gemini, building AGI, Gemini 3.7 Flash breakthrough, Gemini 4 pre-training, frontier AI research, co-building AGI with users
Transcript ↗ Key moments and transcript search come with a later piece.
- AGI
- Koray Kavukcuoglu
- Artificial General Intelligence, a hypothetical AI that can understand or learn any intellectual task that a human being can.
- reinforcement learning
- Koray Kavukcuoglu
- A type of machine learning where an AI learns by trial and error, receiving rewards or penalties for its actions.
39 minIntroducing Gemini Robotics 2
WithCarolina Parada, Stuart Bowers, Kanishka Rao, Jie Tan
42 minGemini co-leads on project origins and what's next
WithJeff Dean, Koray Kavukcuoglu, Noam Shazeer, Oriol Vinyals
40 min1 h 12Yossi Matias on the golden age of research
WithYossi Matias
Same guests, other shows
42 minGemini co-leads on project origins and what's next
WithJeff Dean, Koray Kavukcuoglu, Noam Shazeer, Oriol Vinyals
49 minKoray Kavukcuoglu: “This Is How We Are Going to Build AGI”
WithKoray Kavukcuoglu
26 minTowards the future
WithKoray Kavukcuoglu, Trevor Back, Raia Hadsell, Murray Shanahan
27 minLife is like a game
WithOriol Vinyals, Max Jaderberg, Raia Hadsell, David Silver
1 h 14When AI Improves Itself | Richard Socher (Recursive)
WithRichard Socher
Questions people ask
- What made Gemini 3.7 Flash a breakthrough?
- The show notes state Gemini 3.7 Flash was a breakthrough for engineers, but do not detail why.
- Why is Google the right place to build AGI?
- Kavukcuoglu believes Google is the place to build AGI because of its resources and environment for frontier AI research and development.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Google AI.