Microsoft's AI Frontiers Lab Is Over, Says Departing Researcher John Langford

"AI Frontiers is over," Langford wrote in a post dated November 6 on his long-running blog, hunch.net. He said he had placed the people who reported to him as well as he could and then left, and that every other member of the group's founding leadership had come to a similar conclusion. There will be no internships with the group's researchers in the coming year, he wrote.

The post is a rare public account from inside a big technology company of the argument over whether AI money and talent should go to open-ended research or to building ever larger models. Langford's account is that Microsoft's AI division has chosen scale, and that the research group did not fit.

Microsoft Research formed AI Frontiers after the release of ChatGPT, according to the post, as a sub-organisation of about 50 people. Langford wrote that he had argued at the time that research needed more organisation if it was to compete. The group built agent tooling, including the Autogen orchestration framework and the Fara computer-use agent. Its New York team worked on more fundamental capabilities, which Langford listed as the Next Latent training objective, Dion3, an improved version of the Muon optimiser, and an architecture called the Full Bandwidth Transformer. The group also built a monorepo, a single shared code repository, used across Microsoft Research.

A note on Microsoft's AI Frontiers web page, dated September 14, 2026, reads: "Microsoft Research AI Frontiers is now part of Microsoft AI!" Microsoft AI is the division led by Mustafa Suleyman, who joined Microsoft in 2024 with much of the team from his start-up Inflection AI.

Langford wrote that the merger took place in early September and that the scaling-focused approach of the former Inflection leadership was incompatible with the research focus of his group. "There was plenty of evidence that they just did not really want us," he wrote.

He argued that research still pays off, pointing to experiments at ByteDance that he said suggest a variant of the Full Bandwidth Transformer can match a model twice its size.

Langford also explained why he did not move back into Microsoft Research. He gave two reasons: that research there has no clear path into real-world use, and that he worries the bargain of "do good work and good things happen" has broken down. He wished the colleagues he left behind well and said he hoped Microsoft could again become an attractive place for research.

Langford was one of the founding members of Microsoft Research New York. He is known for his work on reinforcement learning and contextual bandits and for creating the Vowpal Wabbit machine learning system. His departure, and his account of the other founding leaders' decisions, removes a senior group of research managers from Microsoft as the company builds up its own frontier models under Microsoft AI.

The post does not say what Langford will do next. Microsoft's AI Frontiers page still describes the lab's mission as viewing the world through an "agent-native lens".