Microsoft’s AI Frontiers research group is over. John Langford, the Microsoft Research leader who helped create the group, announced on November 6 that he has left the company and that the roughly 50-person team behind the AutoGen agent framework is finished. The announcement, published on Langford’s blog, Machine Learning (Theory), closes one of the most visible research efforts inside Microsoft Research and raises fresh questions about where agent research goes next inside the company.

“So AI Frontiers is over,” Langford wrote, adding that there will be no internships with the group in the coming year. The post describes a team that was set up to run research in a more organized way after ChatGPT arrived, then was moved out of Microsoft Research and into Microsoft AI in early September, a reorganization Langford says proved incompatible with the group’s purpose.

What AI Frontiers built

AI Frontiers was formed inside Microsoft Research in 2023, after the release of ChatGPT made clear to Langford and several colleagues that research needed more coordination than the lab’s usual structure provided. The group, about 50 people, focused much of its work on making AI agents work. Its best-known output is AutoGen, the open source orchestration framework for coordinating multiple AI agents, along with Fara, a computer-use agent that operates a desktop on a user’s behalf.

The New York team also pursued fundamental research. Langford lists the Next Latent objective, which enables extrapolation; Dion3, an improved version of the Muon optimizer; and the Full Bandwidth Transformer, which he describes as making models function substantially more powerfully through recursion. The group also built a shared monorepo for Microsoft Research, including what Langford calls a proven edge-scale pretraining codebase.

The AutoGen logo, the letters AG in metallic silver on a dark background

AutoGen, the multi-agent orchestration framework built by Microsoft’s AI Frontiers group. Logo: Microsoft, via the AutoGen project.

Why it ended

The turning point, in Langford’s account, was the move into Microsoft AI in early September. “The core philosophy there is scaling, and aggressively not research,” he wrote, describing the leadership that came from the former Inflection team. He argues that a scaling-only environment shifts the work toward collecting, making, and refining data, far from the research the group was doing, and that the new organization showed little interest in keeping the team.

“I consider the scaling-only focus to be both misdirected and not fun,” Langford wrote. “It is an understandable failure mode, scaling is straightforward and it is common for people to give up on a tradeoff between multiple competing ideals (like innovation and size) by short-sightedly adopting one in totality and dropping the other. But returns to research are real and significant.”

To support that point, he cites experiments from ByteDance suggesting a variant of the Full Bandwidth Transformer might match a model twice its size, which would imply large savings in training and inference. “Do the math, how much is it worth to halve training and inference cost given that the industry has investments approaching a trillion dollars?” he wrote.

Langford says he placed his team members as best he could before leaving. “So I situated my folks as best I could and left,” he wrote, adding that every other founding member of the AI Frontiers leadership team seems to have reached a similar conclusion. He also considered returning to Microsoft Research and decided against it, writing that he could not get innovation into use while inside Microsoft AI and that he sees a breakdown of what he calls the basic deal in Microsoft Research: do good work and good things happen. “As far as I can tell we did good work and were non-consensually reorged away,” he wrote.

On X, where he shared the post, Langford was blunter still: “There’s a bit of a train wreck here that’s painful to express.”

Building 92 on Microsoft's Redmond campus, with the Microsoft logo sign in front

Building 92 on Microsoft’s Redmond campus. Photo: Coolcaesar, CC BY-SA 4.0, via Wikimedia Commons.

What happens to AutoGen

AutoGen’s future was already narrowing before Langford’s post. On October 16, a change to the project’s repository stated that Microsoft is not providing bug fixes, security fixes, new features, or support for the project. The repository’s readme now describes AutoGen as archived and no longer maintained, and points new projects to the Microsoft Agent Framework, which Microsoft calls the enterprise-ready successor to AutoGen. The Fara repository remains available under an MIT license, with its last push in September.

Microsoft has not commented publicly on Langford’s departure or on the end of the group. The post says nothing about what happens to the remaining AI Frontiers researchers who were placed elsewhere in the company, or about the fate of the group’s other projects.

Why it matters

The closure is a small event in headcount terms, about 50 people inside a company of more than 200,000, but it is a pointed one. AutoGen was one of the most widely used open source agent frameworks, and its quiet wind-down, followed now by the departure of the group’s founder, signals how Microsoft is consolidating its AI work around products and scale rather than open-ended research. For developers who built on AutoGen, the practical message is already clear: the framework is unmaintained, and Microsoft’s own guidance points to the Agent Framework instead.

Langford’s broader argument is about the industry, not just Microsoft. If returns to scale are more predictable than returns to research, he writes, companies will keep choosing scale, and the research that could halve training and inference costs will have nowhere to live. His exit is a test of that claim in public: a senior researcher leaving one of the world’s largest AI employers because, in his words, the research he does has no path to real use there.

Sources