Microsoft has lost the research group behind one of its best-known AI agent frameworks, after its founder declared the team over and quit the company.
John Langford, a partner research manager at Microsoft Research in New York, announced on 6 November that AI Frontiers, the roughly 50-person group he helped create in 2023, no longer exists. In a post on his blog, Machine Learning (Theory), he wrote: “So AI Frontiers is over.”
The group was set up inside Microsoft Research after the release of ChatGPT, when Langford and colleagues concluded that research needed to be organised more coherently to keep pace with better-funded rivals. Much of its work focused on making AI agents work, including the AutoGen framework for coordinating multiple agents and Fara, an agent that operates a computer on a user’s behalf. Its New York team also pursued fundamental work on new model capabilities, including the Next Latent objective, the Dion3 optimiser and the Full Bandwidth Transformer.
The break came in early September, when AI Frontiers was moved out of Microsoft Research and into Microsoft AI, the product-focused division led by former Inflection executives. Langford wrote that the two sides’ philosophies were incompatible. “The core philosophy there is scaling, and aggressively not research,” he said, adding that there was “plenty of evidence that they just did not really want us.”
He argued that a scaling-only approach is understandable — bigger models reliably perform better, and returns to scale are far more predictable than returns to research — but called it “both misdirected and not fun”. He pointed to experiments by ByteDance suggesting a variant of the Full Bandwidth Transformer could match a model twice its size, implying large savings in training and inference costs for an industry investing close to a trillion dollars.
Langford said he placed his team members as best he could before leaving, and that every other founding member of the group’s leadership appears to have reached a similar conclusion. He also said there would be no internships with the group in the coming year.
He considered returning to Microsoft Research but decided against it, writing that he was “just not comfortable being at a company where the research I do has no path to real use” and that the basic deal in Microsoft Research — do good work and good things happen — appeared to have broken down. “As far as I can tell we did good work and were non-consensually reorged away,” he wrote.
Note: this account comes from Langford himself. Microsoft has not publicly commented on his version of events, and the company has not described the move as a closure. The fate of AutoGen underlines the shift: in mid-October, Microsoft updated the project’s page to say it is no longer providing bug fixes, security fixes, new features or support.
Sources: John Langford, “The End of AI Frontiers”, Machine Learning (Theory), 6 November 2026 (https://hunch.net/?p=13764006); The Brief, “Microsoft’s AI Frontiers Research Group Closes After Leader Departs”, 7 November 2026 (https://www.thebrief.news/en/standard/article/41133/microsofts-ai-frontiers-research-group-closes-after-leader-departs).
