Langford: Microsoft's AI Frontiers Is Over

John Langford, a veteran Microsoft Research machine-learning scientist, said on Nov. 6 that he has left Microsoft and that AI Frontiers, the roughly 50-person research group he helped found in 2023, is finished. In a post on his blog, Machine Learning (Theory), he wrote that the group was reorganized into Microsoft AI in early September and could not fit its "scaling-only" approach.

• AI Frontiers was set up inside Microsoft Research (MSR) in 2023, after ChatGPT's release, and numbered about 50 people, according to Langford.
• Its work included the AutoGen multi-agent framework and the Fara-7B computer-use model; Langford says the group was moved into Microsoft AI in early September.
• Langford wrote that "every other founding AIF leadership team member" appears to have reached a similar conclusion, and that the group will offer no internships in the coming year.

Why Microsoft AI Frontiers ended

Langford said AI Frontiers was created because one researcher with a clever idea could no longer compete with the organized efforts at rival labs. The group tried to run coherent projects and shared a monorepo with all of MSR, including what he called a "proven-capable edge scale pretraining codebase."

After the September move, he said, the cultures clashed. He described the ex-Inflection leadership of Microsoft AI this way: "The core philosophy there is scaling, and aggressively not research." He added that "there was plenty of evidence that they just did not really want us."

"I consider the scaling-only focus to be both misdirected and not fun," Langford wrote, while conceding that returns to scale are more predictable for a business than returns to research. "So I situated my folks as best I could and left," he wrote. "So AI Frontiers is over."

What happens to AutoGen and Fara

The group's best-known output is AutoGen. Its GitHub repository is now marked as archived and "no longer actively maintained by Microsoft," and points new users to Microsoft Agent Framework, which it calls AutoGen's enterprise-ready successor.

Fara-7B, a 7-billion-parameter agent that operates a computer by clicking and typing, was released in November 2025 under an MIT license on Microsoft Foundry and Hugging Face. Microsoft's announcement said it was integrated with Magentic-UI, "a research prototype from Microsoft Research AI Frontiers."

The research Langford says is at stake

Langford pointed to architecture work from the group's New York team, including the Full-Bandwidth Transformer. In the arXiv paper, the authors trained 1-billion-parameter models on up to 400 billion tokens and report that they match or approach standard transformers trained on roughly 1.5 times more tokens. Langford also cited ByteDance experiments suggesting a variant "might be comparable to a model of twice the size."

Context: research vs. scale at Big Tech

The dispute is a familiar one in AI labs: whether spending goes to bigger models and more data, or to ideas that make models cheaper. Langford framed it in cost terms, asking how much it would be worth "to halve training and inference cost" in an industry with investments approaching a trillion dollars. AI Newsway found no public statement from Microsoft on the group's status as of Nov. 8.

What's next

Langford said he does not plan to return to MSR, saying he is "not comfortable being at a company where the research I do has no path to real use." He did not name a new employer. Developers still on AutoGen face a migration to Microsoft Agent Framework, since the archived project receives no bug or security fixes.

FAQ

Is AutoGen still safe to use? Microsoft says the archived repository gets no bug fixes, security fixes or support, and advises against production use. It provides a migration guide to Microsoft Agent Framework.

Who is John Langford? He is a longtime Microsoft Research scientist and a co-author of the Full-Bandwidth Transformer paper. He helped found AI Frontiers in 2023.

Has Microsoft confirmed the group is closed? The account comes from Langford's own post. AI Newsway did not find an official Microsoft statement as of Nov. 8.

References

• The End of AI Frontiers — John Langford, Machine Learning (Theory) (Nov. 6, 2026)
• GitHub - microsoft/autogen: A programming framework for agentic AI — GitHub
• Fara-7B: An efficient agentic small language model for computer use — Microsoft Research (Nov. 24, 2025)
• Full-bandwidth transformer — arXiv (Aug. 9, 2026; revised Oct. 1, 2026)