Barret Zoph Lands at Google as VP of Research After OpenAI Exit

Barret Zoph, a co-founder of the AI startup Thinking Machines, has joined Google as vice president of research, according to a statement a Google spokesperson gave to the Wall Street Journal on August 27, 2026. The move caps a turbulent 18 months that included his departure from OpenAI in late 2024, a co-founding role at Thinking Machines alongside Mira Murati, a reported firing from that startup, a second short stint at OpenAI, and now a return to Google, where he had worked before.

What happened

Date	Event
October 2024	Zoph leaves OpenAI after two years to co-found Thinking Machines
January 2026	Zoph and co-founder Luke Metz leave Thinking Machines; it is later revealed Zoph was fired
January-June 2026	Zoph returns to OpenAI, heading AI enterprise sales, for five months
June 2026	Zoph leaves OpenAI again
August 27, 2026	Google confirms Zoph joins as VP of Research to work on Gemini

Google confirmed the hire with a direct quote: “We look forward to Barret returning to Google and bringing his RL and post-training expertise to Gemini,” a company spokesperson told the Wall Street Journal. The phrase “returning” is notable: Google is one of several organizations where Zoph has previously worked.

Thinking Machines was founded by Mira Murati, who had left OpenAI a month before Zoph did in late 2024. Zoph and fellow co-founder Luke Metz both departed the startup in January 2026 in what TechCrunch described as a dramatic exit. The firing detail emerged later.

Why it matters

Zoph’s specific skills are reinforcement learning (RL) and post-training, the processes used to fine-tune and align a base language model after its initial training. Both are central to how competitive frontier models like Gemini are built, so landing someone with that background is a concrete win for Google’s Gemini team.

The broader story, though, is the churn at OpenAI. According to TechCrunch, the company has lost a string of senior figures over the past eight months, including its COO and a top data center executive. A company preparing for an IPO while losing critical staff at this rate is a situation worth watching, especially for businesses that have built workflows or products on top of OpenAI’s API.

This pattern of short tenures also points to a structural tension in the AI industry: top researchers are in extreme demand, and the gap between a startup’s promise and its day-to-day reality can move fast. For anyone integrating AI tools into business operations, leadership continuity at the labs that build those tools matters more than it might seem.

Our take

Five months in an enterprise sales leadership role is a short run for anyone, let alone someone whose CV is built on research. It is hard to avoid the conclusion that the OpenAI role was not a great fit for Zoph’s actual skill set, and Google’s framing around RL and post-training suggests they are placing him somewhere closer to his strengths.

For businesses watching the frontier model race, Zoph moving to Google’s Gemini team is a small but real signal. Google has been competing hard with OpenAI and Anthropic, and adding post-training depth could show up in future Gemini releases. We track these moves on the Lumien AI news feed precisely because a change in a lab’s research leadership often precedes a meaningful model update.

The broader OpenAI churn is worth monitoring. If you are a business with significant spend or product dependency on OpenAI, it is not a reason to panic, but it is a reason to keep your integration loose and your options open.

What to do about it

1. Note which AI lab each key researcher lands at: that often signals where near-term model improvements will concentrate.
2. Review whether your current AI tooling is tied to a single provider’s API or whether you have a provider-agnostic layer in place.
3. Watch Gemini’s post-training quality over the next two to three model releases, as Zoph’s influence may start to show there first.
4. If executive churn at AI labs is making you rethink your stack, talk to the Lumien team about building integrations that are easier to swap out.

The researchers who build these models move around more than most people realize, and where they land tends to matter for which models improve fastest.

Source: TechCrunch · AI