Barret Zoph Returns to Google DeepMind as VP of Research
Barret Zoph has been appointed vice president of research at Google DeepMind, returning to the company where he began his AI career as a research scientist six years before his tenure at OpenAI. Reuters reported the appointment on August 27, 2026. Zoph confirmed the move on his personal X account, noting that he started in AI through Google's Brain Residency class.
A Google spokesperson told the Wall Street Journal that the company looks forward to Zoph "returning and bringing his reinforcement learning and post-training expertise to Gemini." That phrasing is notable. Zoph's deepest technical reputation centers on post-training methodology, the phase of model development where a pre-trained base model, the raw system that has absorbed patterns from massive datasets, is refined through reinforcement learning, preference optimization, and related techniques to produce a deployable system that behaves in line with human expectations.
Zoph's career trajectory over the past two years reads as a compressed map of the industry's talent churn. He left OpenAI in October 2024 to co-found Thinking Machines Lab with Mira Murati, who had departed OpenAI the previous month. The startup, formally unveiled in February 2025, assembled a roster drawn heavily from Murati's former OpenAI colleagues. John Schulman joined as chief scientist, with Zoph serving as CTO. TechCrunch reported those titles at the time of the company's public launch. By July 2025, Thinking Machines Lab had reached a $12 billion valuation in a seed funding round, according to TechCrunch.
The startup's co-founder ranks thinned quickly. In January 2026, Zoph and fellow co-founder Luke Metz departed to rejoin OpenAI, as TechCrunch first reported. A March 2026 TechCrunch report indicated that Sam Schoenholz had also departed, making three co-founders who left for OpenAI in early 2026.
On his return to OpenAI, Zoph was assigned a different remit than his prior role leading model post-training. He was tasked with heading AI enterprise sales and remained roughly five months, departing in June 2026 (The Verge). His personal website lists his previous title at OpenAI as VP of Research (Post-Training), confirming the shift in responsibility between his two stints at the company.
Zoph's move to Google DeepMind closes a loop that began in Google's Brain Residency program and now places him in a research leadership position over Gemini, Google's flagship model family. The role positions him alongside the teams building Gemini's post-training pipeline, the same technical domain where he built his reputation at OpenAI before the Thinking Machines Lab detour.
The broader context here is one of persistent talent circulation among a small set of frontier AI labs. Zoph has now been employed by three of them, Google, OpenAI, and Thinking Machines Lab, and circled back to Google in under two years. His brief at OpenAI this year, enterprise sales rather than research, suggests that organization was testing where his strengths fit during his second tenure; the reassignment did not hold. Google, by contrast, is bringing him back into a research leadership role directly aligned with his core expertise.
One open question is what Zoph's arrival means for DeepMind's existing post-training and reinforcement learning leadership structure. Google has not disclosed details beyond the VP of Research title and the spokesperson's reference to Gemini. What is clear is that a researcher with hands-on experience in frontier post-training at OpenAI is now embedded inside the organization competing most directly with OpenAI in the Gemini-versus-GPT frontier model race. Whether that transfer of institutional knowledge translates into measurable improvements in Gemini's post-training pipeline will play out over subsequent model releases.
