Meta just poached three more engineers from Mira Murati's $12 billion startup, bringing the total raid to eight people in under a year.
The Summary
- Meta hired two founding members and one senior researcher from Thinking Machines Lab, Mira Murati's AI startup that raised $2 billion last year
- Seven of Thinking Machines Lab's founding team members now work at Meta, including the engineer who shipped their flagship product Tinker
- The talent drain reveals how Big Tech wins the AI race: not by building better products, but by hiring the people who already did
The Signal
Thinking Machines Lab raised $12 billion in valuation with $2 billion in funding. That should buy you some retention. Instead, Meta has systematically dismantled the founding team. Mark Jen and Yinghai Lu were both founding members. Lu specializes in inference optimization, the unglamorous work that makes AI chatbots actually reason instead of just pattern-match. Tianyi Zhang isn't a founder but has co-authored widely cited papers, the kind that become footnotes in breakthrough announcements.
None of them updated LinkedIn. That's not modesty. That's NDAs and quiet departures orchestrated to avoid feeding a media narrative. But the pattern is loud: Meta hired Joshua Gross, the engineer who built and shipped Tinker, Thinking Machines Lab's flagship product. When you lose the person who shipped your main thing, you're not losing an employee. You're losing institutional knowledge about what actually works.
"When you lose the person who shipped your main thing, you're not losing an employee. You're losing institutional knowledge."
Here's what makes this different from normal poaching. Thinking Machines Lab was supposed to be the place where top talent could build without the bureaucratic drag of Big Tech. Murati left OpenAI as CTO to start it. Soumith Chintala, creator of PyTorch, left Meta to become CTO. These were people voting with their careers for the indie lab model.
And Meta is systematically reversing that vote. Eight people in a year. That's not coincidence. That's a targeted raid on a competitor's core capacity. The startup model promised ownership, autonomy, and the chance to build something new. Meta is offering something that apparently beats all three.
What could that be? Three possibilities:
- Compute access: Thinking Machines Lab has funding, but Meta has data centers that can run experiments most startups can only dream about
- Distribution: Building a great model means nothing if no one uses it. Meta has 3 billion daily active users across its platforms
- Certainty: Startups burn through runway. Meta has $60 billion in annual revenue and isn't going anywhere
The inference specialization matters here. Lu's focus on inference optimization is where the real product work happens in AI. Training models gets the headlines. Inference is what makes them fast enough to ship. If Meta is hiring the people who know how to make AI reason efficiently, they're not just talent-hoarding. They're building production capacity for agent-based products that need to think, not just respond.
The Implication
If you're building an AI startup, this is your warning shot. Raising billions doesn't protect you from talent flight if Big Tech decides your team is the product they actually want. The agents being built at Meta right now aren't coming from Meta's internal R&D. They're coming from the people who already proved they could ship at Thinking Machines Lab.
Watch what Meta ships in the next six months. If it looks suspiciously like what Thinking Machines Lab was building, you'll know exactly where it came from.