Microsoft Research's AI Frontiers Comes to an End: The Roughly 50-Person Agent Research Group Was Moved into Microsoft AI in September, and Founding Leader John Langford Says He Left over Incompatible Direction

Machine learning researcher John Langford, creator of Vowpal Wabbit, published "The End of AI Frontiers" on his blog hunch.net on November 6, saying that AI Frontiers, a roughly 50-person research group inside Microsoft Research, is over and that he has left Microsoft. He writes that the group, formed in 2023, built the AutoGen multi-agent framework and the Fara computer-use agent, and was reorganized into Microsoft AI in early September; he describes Microsoft AI, led by former Inflection leadership, as focused on scaling rather than research and incompatible with AIF, and says the other founding leaders reached similar conclusions. This is Langford's own account; no public response from Microsoft was found as of verification.

An ending announced in a blog post

John Langford, the machine learning researcher behind the open-source learning system Vowpal Wabbit, published "The End of AI Frontiers" on his personal blog hunch.net on November 6, announcing that the AI Frontiers (AIF) group inside Microsoft Research (MSR) is over and that he has left Microsoft. By his account, AIF was a sub-organization that several researchers set up within MSR in 2023, after the ChatGPT release, with about 50 people, to organize research into coherent efforts rather than individual projects. Much of its work was making agents actually work, including the AutoGen multi-agent orchestration framework and the Fara computer-use agent; the New York team also worked on fundamental capabilities such as the Next Latent objective, the Dion3 optimizer and the Full Bandwidth Transformer.

The split after the reorg

Langford writes that AIF was reorganized into Microsoft AI (MAI) in early September. He describes the core philosophy of MAI, led by former Inflection leadership, as scaling and aggressively not research, with the main effort going into getting and refining data, which he says was incompatible with AIF's direction, adding that "they just did not really want us." He says he situated his team as best he could and left, and that every other founding AIF leader seems to have reached a similar conclusion. He also chose not to return to MSR, citing two obstacles: his innovations could not get into use while he was at MAI, and the team was "non-consensually reorged away" despite doing good work. He says there will be no internships with the group's researchers in the coming year.

What remains unanswered

All of this is Langford's own account. As of verification, no public response from Microsoft was found, and neither where the team members went nor how AIF was formally handled in the org chart has been disclosed. The AutoGen repository's README currently carries an "archived" status badge pointing to Microsoft Agent Framework.

via: Machine Learning (Theory): The End of AI Frontiers, GitHub: microsoft/autogen