Barret Zoph, the Thinking Machines co-founder who returned to OpenAI, is now at Google as VP of Research

2026-08-28·6 min read

On August 27, 2026, per TechCrunch, the game of musical chairs for AI executives saw another reshuffle: Barret Zoph, co-founder of Thinking Machines who dramatically returned to OpenAI earlier this year and left a few months later, has found a new home: Google. Zoph will serve as Vice President of Research at Google. A Google spokesperson told the Wall Street Journal: 'We look forward to Barret returning to Google and bringing his RL and post-training expertise to Gemini.' For those familiar with Zoph's résumé, this feels like a homecoming — he is, after all, a former Google employee.

Walking through Zoph's trajectory over the past two-plus years reveals just how surreal the top AI talent market has become. Zoph spent two years at OpenAI and left in October 2024 to co-found AI startup Thinking Machines with Mira Murati, who had departed the lab the month prior — a company that quickly became one of the hottest startups in AI. In January 2026, Zoph and fellow co-founder Luke Metz dramatically departed Thinking Machines to return to OpenAI; it was later revealed that Zoph had actually been fired, not a voluntary returnee. His 'return' did not stick for long either: this time Zoph spent only about five months at OpenAI, tasked with heading AI enterprise sales, and left again in June. Now his next stop is confirmed — Vice President of Research at Google.

Why does Google value Zoph so highly? The answer is hidden in the spokesperson's words: reinforcement learning and post-training expertise. In the current AI competitive landscape, architecture gaps are narrowing, and what determines a model's ceiling increasingly depends on the post-training phase — RL, alignment, data curation and fine-tuning strategy. Zoph is a veteran in this domain: he has been involved in training and post-training frontier models across OpenAI and Thinking Machines, leaving technical fingerprints everywhere. By slotting an expert with both startup experience and familiarity with frontier lab processes into the Gemini team, Google is clearly doubling down on the post-training track. For Google, the Gemini team is in a critical phase of hand-to-hand combat with OpenAI and Anthropic — an executive who knows RL cold brings not just technique but first-hand experience with frontier training methods.

Zoph's move, viewed against the broader backdrop, is a microcosm of accelerating talent flow in the AI industry. Since 2026, top AI talent has been moving between labs, startups and giants at an increasingly fast pace: some left OpenAI to found companies, some returned from startups to big tech, and some like Zoph completed a full 'OpenAI — startup — OpenAI — Google' loop. This churn is a sign of a healthy industry — top talent searching for the position where they create the most value; but it also reflects the white-hot nature of AI competition: every giant is desperately filling its technical gaps, and talent is the scarcest resource. For Google, landing Zoph is a shrewd deal: one executive position in exchange for one of the most senior voices in RL. For Zoph, the Google VP of Research seat offers a stage where he can keep digging deep into the technical frontier without shouldering startup risk.

📌 Source: TechCrunch (August 27, 2026) — 'Barret Zoph, the Thinking Machines co-founder ousted before joining OpenAI, is now at Google' by Max Zeff. Google spokesperson comment via the Wall Street Journal. Link: techcrunch.com/2026/08/27/barret-zoph-the-thinking-machines-co-founder-who-defected-to-openai-is-now-at-google/

🤔 Frequently Asked Questions

Q1: Who is Barret Zoph?

Barret Zoph is a veteran AI researcher who spent two years at OpenAI, co-founded Thinking Machines with Mira Murati in October 2024, returned to OpenAI in January 2026, left again in June, and has now joined Google as VP of Research. His expertise is reinforcement learning and post-training.

Q2: What role does he hold at Google?

Zoph serves as Vice President of Research at Google. A Google spokesperson said the company looks forward to him bringing his RL and post-training expertise to the Gemini team.

Q3: Why did he leave OpenAI?

Per TechCrunch, Zoph's January 2026 return to OpenAI came after being fired from Thinking Machines, not as a voluntary move; after returning he headed AI enterprise sales, stayed only about five months, and left again in June. The specific reason for departure was not publicly disclosed.

Q4: What does this mean for Google's Gemini?

Zoph's RL and post-training experience is exactly the critical link in current model competition. His arrival could help the Gemini team accelerate in post-training, alignment and fine-tuning strategy — a major talent boost for Google against OpenAI and Anthropic.

🛠️ Recommended Tools

  • AI Image Generator - Experience cutting-edge generative AI and feel the real impact of model iteration
  • Text Summarizer - Track the dense flow of AI talent and product news with quick key-point extraction
  • Video Cutter - Efficiently handle AI-generated multimedia and turn industry insights into creative material

Summary

Barret Zoph joining Google is the latest footnote to the violent churn in the AI talent market. The technical heavyweight's career arc — two years at OpenAI, co-founding Thinking Machines with Mira Murati, a dramatic return to OpenAI that was later revealed to be a firing, another departure five months later, and now landing at Google — is itself a condensed chronicle of the AI industry. Google pays little for a VP of Research seat yet gains one of the most senior practitioners in reinforcement learning and post-training; at a moment when architectures are converging and post-training determines the ceiling, the value of this talent deal cannot be overstated. For industry observers, Zoph's destination reaffirms a trend: AI competition has fully shifted from parameter comparison to a combined contest of talent, compute and post-training craft. And every top-tier talent move is quietly reshaping the fate of the next generation of models.