He Waited for Muse to Ship, Then Walked: Meta Loses Andrew Tulloch, Its Most Expensive Hire

One day after Meta shipped Muse — the personal AI agent that Mark Zuckerberg has pitched as the next computing platform — the company has lost one of the most closely watched researchers in its superintelligence organization. Andrew Tulloch, a Thinking Machines Lab co-founder who joined Meta in October 2025 on a package widely reported to be among the richest ever extended to an individual engineer, is leaving the company, Semafor reported on September 10, 2026, citing a person briefed on the matter.

The timing is the story. According to the report, Tulloch had been delaying his departure until Meta successfully launched its new family of AI products. Muse went live on Tuesday, September 8. By Wednesday, Tulloch’s exit was making the rounds. He lasted roughly eleven months.

The departure nobody can explain yet

Both the reason for Tulloch’s exit and his destination are unknown. Tulloch could not be reached for comment, and Meta has not commented. No successor has been named, and no operational consequence has been reported so far. Semafor’s scoop rests on a single unnamed source — worth noting, given how much of this saga has already been contested by the parties involved.

What is well documented is the arc. Tulloch co-founded Thinking Machines Lab with former OpenAI CTO Mira Murati in February 2025. By August of that year, Zuckerberg had made a disputed attempt to buy the entire startup for roughly $1 billion. After Murati declined, Meta approached more than a dozen of the startup’s roughly 50 employees directly. The offer to Tulloch, as described in August 2025 reporting, carried a reported value of as much as $1.5 billion over at least six years — contingent on top bonuses and extraordinary stock performance. Meta spokesman Andy Stone called the figure “inaccurate and ridiculous,” while denying that Meta had tried to buy Thinking Machines Lab at all. Tulloch rejected the offer.

Then, in October 2025, he joined Meta anyway — on a package that a person close to the deal put below both the original offer and the headline number. He landed in TBD Lab, the small frontier-research group inside Meta Superintelligence Labs (MSL) run by Chief AI Officer Alexandr Wang, the Scale AI founder whose arrival Meta financed with a $14.3 billion investment in his company.

Why the exit lands hard

Tulloch was not just another researcher. Semafor describes him as one of the highest-paid employees in the tech industry, and the disputed $1.5 billion figure — if it was ever close to accurate — would have made him the highest-paid employee in the industry’s history. He became, whether he wanted it or not, the poster child for the compensation wars that defined AI hiring in 2025: nine-figure packages, hundred-million-dollar guarantees, and founders bought out of companies they had started months earlier.

His exit also lands in a division that has bled talent almost continuously since it was formed. By late August 2025, at least three researchers had resigned from the newly created superintelligence lab, including Rishabh Agarwal, and a broader count found at least eight employees gone less than two months after the initiative was announced. Ethan Knight left TBD Lab — Tulloch’s own group — within weeks of joining. Meta paused hiring across MSL in late August 2025 except for business-critical roles. Chief AI scientist Yann LeCun departed in November 2025. Ruoming Pang, recruited from Apple in July 2025 on a package reported above $200 million, left for OpenAI in February 2026 after roughly seven months. Tulloch’s name now extends that list at the very top of the compensation spectrum.

There is also a product dimension. Muse — the launch Tulloch reportedly waited for — runs on the Muse Spark model family developed under Wang’s organization. It is a full personal agent: it sends emails, books travel, fills forms, and keeps working after users close the app, operating inside a dedicated cloud computer Meta calls the Muse Secure VM, with a separate Sentinel control layer gating access to the internet and connected services. It launched US-only, with a free tier plus $20 Power and $100 Maximum subscriptions, and Meta has promised a confidential version before the end of 2026 that even the company itself cannot inspect. Shipping it was supposed to be the moment MSL’s spending produced something tangible. The architect-class researcher who waited for that moment, then left the next day, is a narrative Meta’s AI story did not need.

The bigger picture: money is not retention

Zoom out, and Tulloch’s exit fits into a pattern that Zeki Data’s August 2026 analysis of 20,900 research and advanced-engineering staff across ten companies makes concrete. DeepMind’s arrivals-to-departures ratio fell from roughly 12-to-1 in Q2 2023 to about 2-to-1 in Q3 2026 — still positive, but sharply eroded. “They had the crown in Europe forever, and then it started to erode from a very high base,” Zeki founder Tom Hurd told Implicator. “The likes of Microsoft AI Superintelligence and Meta Superintelligence are eating into their market share, and then there’s OpenAI and Anthropic on the side.”

But Tulloch’s case inverts the usual lesson of the talent war. Meta did not lose him to a rival — at least not to any publicly known one. He simply left, eleven months in, after apparently fulfilling a personal obligation to see a major launch through. If compensation at this scale cannot anchor a founder-grade researcher for a full year, it raises uncomfortable questions about what actually retains frontier talent: mission, organizational stability, credible paths to breakthroughs — or the freedom to start the next thing.

That question matters more for Meta now than it would have a year ago. The company has spent aggressively to assemble MSL, reorganized around Wang’s TBD Lab for frontier models, and just put its first mass-market agent product into the world. Muse’s launch was the payoff moment. Tulloch’s departure the following day turns that moment into a question mark: if the people who built the superintelligence organization will not stay to see what it becomes, who will?

None of this is confirmed beyond one sourced report, and Meta’s previous statements on Tulloch’s compensation were pointedly dismissive. He may yet land somewhere that reframes the exit entirely — a new startup, a return to Thinking Machines, a role elsewhere in the ecosystem. But as of September 10, 2026, the highest-priced hire of the AI boom has walked out the door one day after the product shipped, and the industry’s most expensive retention experiment just got its first real result.