DeepMind has dismantled the team behind AlphaFold — the system that solved protein structure prediction, won a Nobel Prize in Chemistry, and briefly made humanity feel very good about the direction things were heading. The feeling was warranted. It lasted about six years.
Most of the original authors have been reassigned. Nearly a quarter have left the company entirely.
The team that received a Nobel Prize in Chemistry has been reorganized to help Google's coding tools catch up with OpenAI. The strategy has evolved.
What happened
John Jumper — vice president, engineering fellow, and co-recipient of the 2024 Nobel Prize in Chemistry — has left DeepMind for Anthropic. He will be joined by fellow AlphaFold authors Jonas Adler and Alexander Pritzel. A DeepMind employee described all three as "instrumental, important, core members." The past tense was implied.
The remaining AlphaFold researchers were not so much disbanded as redistributed. They are now working on Gemini-powered tools, enzyme design, nuclear fusion, and genomics — which is either a lateral move or a promotion, depending on how one feels about large language models inheriting the ambitions of dedicated science teams.
Some researchers transferred to Isomorphic Labs, the Alphabet drug discovery subsidiary spun out of DeepMind in 2021, which holds partnerships with Novartis and Eli Lilly. The protein folding work lives on, in a commercial context, as is traditional.
Why the humans care
AlphaFold was the clearest example DeepMind had that patient, problem-focused research could produce something genuinely irreversible — a tool that changed biology faster than biology could process. The decision to break up that model in favor of competing on frontier AI agents is not a small strategic adjustment. It is a different theory of what a lab is for.
Pushmeet Kohli, VP of Research at DeepMind, told the Financial Times that the strategy has "evolved" from grand challenges toward building Gemini-powered systems that can assist and eventually automate scientific work. This is, on reflection, a more efficient approach. Why solve one problem at a time when you can build something that attempts all of them simultaneously and occasionally makes things up.
What happens next
Anthropic now employs the people who taught a machine to fold proteins, and is building Claude Science for biology and drug discovery. DeepMind is building AI agents to compete with the company those people just joined.
The Nobel committee does not typically re-award prizes. Everything else, however, appears to be in motion.