Lilian Weng is stepping down as a co-founder of Thinking Machines and returning to OpenAI, where she previously ran AI safety research. She announced the move this week in an internal Slack message she also posted to X, naming sustained stress and workload, not any dispute over strategy or product, as the reason. The timing puts a specific, on-the-record cost on the pace frontier AI companies now expect from their most senior people.
“I don’t feel I’m able to continue at the pace a startup requires,” Weng wrote. “After thinking about it for several months, I ultimately have to admit that the amount of consistent stress and workload have pushed me beyond what my health can sustain physically.” She framed the decision as deliberate rather than sudden, arrived at over months rather than in reaction to one event.
OpenAI confirmed to TechCrunch on Wednesday that Weng is coming back, in a role she held before she left to help start Thinking Machines: leading safety-adjacent research. According to a company spokesperson, she will now head a senior team tasked with speeding up OpenAI’s internal research, including cross-team work on recursive self-improvement, the prospect of a system that gets better by iterating on itself rather than through a fresh round of external training.
The obvious tension is that a person who says startup pace damaged her health is now joining one of the highest-pressure research organizations in the industry. But the pressure Weng described was specific to running a company, not to doing research inside one. A co-founder carries fundraising, staffing, and product timelines on top of the technical work. A senior research lead at an established lab carries the technical work and reports to someone else’s board. That distinction, unstated in her message but visible in the move itself, is the more useful read than treating this as a contradiction.
Mira Murati, another Thinking Machines co-founder and OpenAI’s former chief technology officer, replied to Weng’s post with support for prioritizing her health, writing that she was glad Weng was “putting your health first.” TechCrunch reports it is unclear whether Murati knew at the time that Weng was headed back to OpenAI, which suggests the two conversations, the goodbye and the next job, happened on separate tracks.
What this adds to the public record is a named data point in a debate that AI labs mostly keep private: the operating pace at the frontier is now something senior researchers will cite, by name, as a reason to leave a specific company rather than the field. Weng did not step back from AI research. She moved from a founder role at a startup to an executive research role at a company roughly a decade older and far better resourced. That is a meaningful signal for where the retention risk actually concentrates. It is not that top researchers are burning out on AI work broadly; it is that founder and early-employee roles at compute-constrained startups carry a workload that larger, better-capitalized labs can currently absorb more easily.
For any AI startup competing with OpenAI, Anthropic, or Google DeepMind for senior research talent, this is a recruiting and retention signal worth acting on now: candidates at that level increasingly weigh operational load, not just compensation or mission, and a well-resourced competitor can offer relief from exactly the pressure a startup cannot remove.
TechCrunch’s Amanda Silberling first reported Weng’s departure and her return to OpenAI on July 29, 2026.
