Google invented the transformer and half the core AI architectures in use today, but it's bleeding the people who can actually build with them.
The Summary
- Jonas Adler and Alexander Pritzel are leaving Google for Anthropic, joining a pattern of high-profile departures that includes Noam Shazeer and John Jumper
- The exits risk undercutting Google's position in AI just as commercial agent deployment hits critical mass
- Anthropic now has a proven playbook for pulling top talent from the company that wrote the Attention Is All You Need paper
The Signal
Jonas Adler and Alexander Pritzel's departures mark the latest chapter in Google's talent hemorrhage problem. They follow Noam Shazeer, who co-invented the transformer architecture while at Google before leaving for Character.AI and recently rejoining Google only to leave again. John Jumper, who won a Nobel Prize for his work on AlphaFold, also left. The pattern is clear: Google trains world-class AI researchers, and competitors harvest them.
Bloomberg sources describe this as a series of high-profile departures that threaten Google's competitive position. That framing undersells what's actually happening. This isn't about prestige. It's about execution speed.
"Google invented the transformer and half the core AI architectures in use today, but it's bleeding the people who can actually build with them."
Anthropic has become the destination for Google researchers who want to ship. The company has Claude in production, developer tools in market, and a clear product vision. Google has Gemini, Bard's corpse, and a labyrinth of internal politics that turns cutting-edge research into vaporware. When you're a researcher who spent years advancing the state of the art, watching your work get buried in committee review or killed by executive indecision gets old fast.
The deeper issue: Google's AI strategy is defensive. They're protecting Search, protecting Ads, protecting a business model built for Web2. Anthropic is building for agents. They're designing systems meant to act autonomously, interface with tools, and operate at the edge of what's possible. That's where the frontier researchers want to be.
Key competitive dynamics:
- Google has more compute, more data, more money
- Anthropic has faster iteration cycles and fewer legacy products to protect
- Talent follows the pace of shipping, not the size of the research budget
Google still has DeepMind, which remains a powerhouse. But DeepMind's best people have options now. When Anthropic comes calling with equity in a company valued at $18 billion and the promise that your research will ship in months instead of years, the math gets compelling. Especially when you've watched colleagues leave and thrive.
The Implication
If you're building on AI infrastructure, watch where the talent goes. Google will keep publishing groundbreaking papers. But the distance between their research and your ability to use it in production is growing. Anthropic, OpenAI, and the other research-to-product shops are closing that gap. The researchers moving there aren't chasing headlines. They're chasing the ability to see their work deployed at scale while it still matters.
For Google, this is a crisis of culture, not capability. They have the resources to win. But resources don't ship products. People do. And right now, the people are leaving.