Koray Kavukcuoglu

SVP and Chief AI Architect, DeepMind

Search every verified Koray Kavukcuoglu interview, podcast appearance, and on-the-record quote — each transcript cross-checked by AI and human review to confirm speaker identity. Koray Kavukcuoglu, SVP and Chief AI Architect at Google DeepMind, discussed AI agent autonomy and recursive self-improvement in a September 2026 interview with The Information's Jessica Lessin. He described recursive self-improvement as a staged process, noting that the current setup is not an end-to-end, human-free system and that human involvement remains part of the loop. He also spoke about the company's focus on frontier models, stating that being at the frontier is the primary priority for all decision-making and goal-setting at Google DeepMind. In an August 2026 appearance on the Release Notes podcast, Kavukcuoglu reflected on DeepMind's early reinforcement learning milestones, including the Atari project, and discussed recent model releases such as Gemini 3.7 Flash and the ambitions of the Gemini 4 pre-training run. He said there is no agreed-upon test for artificial general intelligence (AGI), and invited anyone with such a benchmark to share it so the team could run its models on it.