John Langford, one of the founders of the AI Frontiers group at Microsoft Research, announced on November 6, 2026, that the group is over — and that he has left Microsoft. The statement was published on Langford's blog Machine Learning (Theory) and shared by him on X the same day.
Created in 2023, after the launch of ChatGPT, AI Frontiers was a sub-organization inside Microsoft Research, with about 50 people. Its goal was to organize research in a more coordinated way than the lab's traditional model, with coherent efforts around AI agents and new fundamental model capabilities.
What the group built
Among the work cited by Langford are the agent orchestration framework AutoGen and the computer-use agent Fara. The New York team also developed fundamental research such as the Next Latent objective, the Dion3 optimizer — an improved version of Muon — and the Full Bandwidth Transformer, a technique that aims to make models substantially more powerful through recursion.
The September reorganization
According to Langford, in early September 2026 AI Frontiers was reorganized and folded into Microsoft AI (MAI), the company's AI division. He says the philosophy of MAI's leadership, made up of former Inflection executives, is incompatible with that of the research group: in his words, “scaling, and aggressively not research”.
Langford writes that he placed his team “as best I could” and left the company, and that every other founding member of the AI Frontiers leadership seems to have reached a similar conclusion. He also says there will be no internships with the group next year.
What is a statement and what is confirmed
Confirmed by Langford's original post: the group's creation in 2023, its approximate size of 50 people, the reorganization into Microsoft AI in early September, and Langford's departure. The characterization of Microsoft AI's philosophy as “scaling, and aggressively not research” is Langford's own assessment — a statement, not an official Microsoft position. As of publication, no public Microsoft statement responding to the post was found.
Another sign of the wind-down: since October 16, the AutoGen repository has displayed a notice that Microsoft no longer provides bug fixes, security fixes, new features, or support for the project. Microsoft, meanwhile, shipped Agent Framework 1.0 in April 2026, its official agent framework for .NET and Python.
Why it matters
The case adds to a trend among Big Tech companies: fundamental research groups have been absorbed into units focused on building models — Google merged Brain into DeepMind in 2023, and Meta reorganized FAIR in 2025. Langford's departure raises the question of how much room remains for long-term research inside large AI companies, at a time when industry investment is approaching one trillion dollars.
Langford's post spread quickly: he announced it on X on November 6, and the Hacker News discussion gathered hundreds of points and more than 200 comments.
The end of AI Frontiers is another chapter in this industry-wide reorganization — Geekiki has already covered other shifts in Microsoft’s AI strategy, such as the end of Oracle’s AI capacity rental, according to Bloomberg.
Image credit: photo of Building 92 at Microsoft's Redmond campus, by Jiaqian AirplaneFan, via Wikimedia Commons (CC BY).
Sources
- John Langford, “The End of AI Frontiers”, Machine Learning (Theory), November 6, 2026
- John Langford on X, November 6, 2026
- Dev.to: “The Microsoft Lab That Worked on AutoGen Is Over”
- FourWeekMBA: “3 Big Tech AI Research Groups Reorganized Since 2023”
- Claude Reads HN: reaction to the announcement
- Wikimedia Commons: Building 92 of Microsoft Redmond Campus
Transparency note: this article was produced with the assistance of artificial intelligence, based on John Langford's original post, public posts on X, and sources from the internet. We recommend consulting the original sources listed above.
Muse