Microsoft Research Disbands AI Frontiers Group as Founder Departs

What happened

The post appeared on Langford's personal blog, hunch.net, under the title "The End of AI Frontiers," and reached Hacker News on Nov 7. In it, Langford wrote that the AI Frontiers group is over, that there will be no internships with its researchers in the coming year, and that he has left Microsoft, with the other members of the founding leadership team reaching similar conclusions. He wrote that he is fortunate that many people in the world can see the value of research and innovation.

The group was formed in 2023 after the ChatGPT release, with about 50 people, aiming to organize research more coherently. Its work ran on two tracks: agent work such as the AutoGen orchestration framework and the Fara computer-use agent, and new fundamental capabilities explored by the New York team, including the Next Latent objective, Dion3, an optimized version of Muon, and the Full Bandwidth Transformer, which makes models substantially more powerful through recursion. The group also built a single shared code repository for all of Microsoft Research, containing proven-capable edge-scale pretraining code.

In early September the group was folded into Microsoft AI. Langford wrote that the merger could make sense on the surface, but Microsoft AI's leadership came from Inflection, whose core philosophy is scaling and aggressively not research, incompatible with the group's direction. He wrote that he could not get innovation into use while at Microsoft AI, and that returning to Microsoft Research offered little prospect, because the basic deal — do good work and good things happen — had broken down when the group was reorganized away without consent despite doing good work. He also noted the group had often faced severe compute bottlenecks.

Key facts

1. Author: John Langford, founder of Microsoft Research's AI Frontiers group, on his personal blog hunch.net, "The End of AI Frontiers"
2. Conclusion: the AI Frontiers group is over; there will be no internships with its researchers in the coming year
3. Formation: created in 2023 after the ChatGPT release, about 50 people
4. Work (agents): the AutoGen orchestration framework and the Fara computer-use agent
5. Work (fundamentals): the Next Latent objective, Dion3 and the Full Bandwidth Transformer
6. Shared infrastructure: a single code repository shared by default across all of Microsoft Research, including edge-scale pretraining code
7. Reorganization: folded into Microsoft AI in early September; Microsoft AI's leadership came from Inflection
8. Split: Microsoft AI's core philosophy is scaling, explicitly not research; Langford wrote he could not get innovation into use there
9. Departures: Langford has left Microsoft, and the other founding leadership members reached similar conclusions
10. Reach and constraints: the post reached Hacker News on Nov 7; Langford wrote compute availability was often a severe bottleneck