Metis Reasoning

AI Reasoning · Enterprise AI · Robotics

Launch

AlphaGo Co-Creator Launches Compute-Efficient AI Startup Metis Reasoning

The startup is betting that verifiable reasoning, not bigger models, will win trust in regulated fields like engineering and medicine.

• Thore Graepel, co-creator of DeepMind's AlphaGo and a former Google DeepMind research lead, launched Metis Reasoning in London, reportedly raising tens of millions of dollars from VCs.
• The 12-person team pairs Graepel with serial entrepreneur Piers Ingram, Boeing-trained robotics specialist Michael Warren, and Cambridge Consultants AI chief David Berman, a UK government and NATO AI advisor.
• Metis separates an AI system's knowledge from its reasoning process, maintaining an explicit record of what it knows, what remains uncertain and what evidence supports each conclusion.
• The company argues today's LLMs reason unreliably and leave no audit trail, a gap it says blocks adoption in fields like law, medicine and engineering where work must be verified.
• Metis joins a cluster of London AI labs founded by senior researchers, including David Silver's Ineffable Intelligence and Richard Socher's Recursive Superintelligence.
• If a leaner, auditable architecture can match frontier LLM performance, it would undercut the premise that progress in AI reasoning requires ever-larger training budgets.