The Pulse

AI Frontiers Is Over After Its Early-September Reorganization Into Microsoft AI, John Langford Writes

AI Frontiers is over after it was reorganized into Microsoft AI in early September, John Langford says. He describes the group’s research projects and writes that there will be no internships with the people there in the coming year.

AI.info Team · Written with AI · fact-checked · 2026-11-09

John Langford posted “The End of AI Frontiers” on 11/6/2026. In the post, he writes:

So AI Frontiers is over.

John Langford

He writes that he situated his folks as best he could and left. Langford adds that every other founding AIF leadership team member seems to have come to a similar conclusion.

AI Frontiers actively tried to organize coherent research efforts within Microsoft Research

In 2023, after ChatGPT’s release, Langford says, he concluded that research at Microsoft Research needed more organization. Several others thought similarly, he writes, because a lone researcher could struggle to keep pace with better-supported groups and could not assemble baselines strong enough to show that an idea was worthwhile. AI Frontiers actively tried to organize coherent research efforts in what Langford calls a “messy middle” between top-down and bottom-up approaches, with disagreements and compute availability among its challenges.

Langford describes AI Frontiers as a group of ~50 people. Much of its work involved figuring out how to make agents work, including the Autogen orchestration framework and the Fara computer use agent. In NYC, several members were interested in developing new fundamental capabilities such as the Next Latent objective, which enables extrapolation; Dion3, which creates an optimized/improved Muon; and the Full Bandwidth Transformer, which efficiently makes models function substantially more powerfully via recursion.

AI Frontiers decided to develop a monorepo shared by default with all of Microsoft Research. The monorepo contained the edge-scale pretraining codebase he discussed in a COLM keynote last month, Langford says. Langford was aware of at least 4 efforts to develop model building pipelines, he says, including the other 3 not in public-within-MSR repos.

AI Frontiers was reorged into Microsoft AI in early September

Putting the two organizations together could make sense because it creates opportunities to concentrate effort, Langford writes, since there were places where AIF was ahead of MAI and vice versa, while MAI was much larger. He describes the philosophy of the ex-Inflection leadership of MAI as incompatible with AI Frontiers’ work: its core philosophy is scaling, and aggressively not research.

A scaling-only focus puts the primary effort into getting/making/refining data, Langford says, which was quite a step away from AI Frontiers’ work. He acknowledges the business case for scaling, writing that bigger models are better models as far as the quality of the predictions they make and that returns to scale are far more predictable than returns to research. But he argues that returns to research are real and significant.

As an example, Langford points to Bytedance experiments suggesting that a Full Bandwidth Transformer variant might be comparable to a model of twice the size, implying significant savings in training and inference for the same capability.

Langford did not return to Microsoft Research

Langford did not return to Microsoft Research because he was uncomfortable working where his research had no path to real use, he says. He writes that he could not get innovation into use, or even seriously considered, while in Microsoft AI, and saw little prospect of success from Microsoft Research. He also says he feared a breakdown in what he calls the basic deal in Microsoft Research: “do good work and good things happen.”

No internships with the people there in the coming year, Langford writes

There will not be internships with the folks there in the coming year, Langford says. He enjoyed working with interns, he says, and sends his best wishes to the people he leaves behind.