TL;DR
Microsoft Research’s AI Frontiers group has ended following a reorganization that moved it into Microsoft AI in early September, according to a report by one of its founding leaders. The leader said he left because of a mismatch over research priorities, and that other founding leaders reached a similar conclusion. Microsoft has not provided a response in the supplied source material, and the group’s personnel and projects have not been fully accounted for.
Microsoft Research’s AI Frontiers group has ended after a reorganization moved it into Microsoft AI in early September, according to an account by one of the group’s founding leaders. The leader said he left after concluding that the organization’s emphasis on scaling models did not fit the group’s research-focused work, and that the other founding leadership members appeared to reach a similar decision.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
In a post published on hunch.net, the leader described AI Frontiers as a roughly 50-person sub-organization created within Microsoft Research after ChatGPT’s release in 2023. Its aim, he wrote, was to coordinate research more closely so teams could develop ideas, build stronger comparisons, and move work forward more effectively than individual researchers could on their own. He characterized its approach as neither strictly top-down nor bottom-up.
The account names work on AI agents, including the AutoGen orchestration framework and the Fara computer-use agent. It also describes research efforts in New York on model capabilities and training methods, including the Next Latent objective, Dion3, and the Full Bandwidth Transformer. The leader said AI Frontiers created a shared code repository for Microsoft Research and included a model-pretraining codebase intended for edge-scale use.
After the team was reorganized into Microsoft AI, the leader said he tried to find a place for his colleagues but left because he believed the new organization’s priorities did not allow a path from research to practical use. He wrote that the other founding AI Frontiers leaders appeared to have reached a similar conclusion. He also said the group would not offer internships in the coming year. The source does not identify all departing members or provide a complete account of where the group’s projects and staff went.
Research Priorities After the Reorganization
The group’s end highlights a dispute over how an AI organization should improve models: through research into new methods and capabilities, or by concentrating on larger-scale training and data. The leader’s account presents that divide as the main reason AI Frontiers could not continue its work inside Microsoft AI. That is his assessment; the supplied material does not include Microsoft’s explanation of the reorganization or its strategy.
The issue matters beyond the team because its work covered tools and methods that could affect how models are developed and deployed. The leader argued that research can reduce the resources needed to reach a given capability, citing a separate set of ByteDance experiments as suggesting that a version of the Full Bandwidth Transformer might match a model twice its size. That comparison is presented in his post, not as an independently verified result in the available material. The broader question is whether the reorganization changes the support, staffing, or route to deployment for such research.
For researchers and students, the end of the group also has a direct consequence: the leader said AI Frontiers will not host internships in the coming year. Whether other Microsoft teams will offer opportunities connected to the group’s work is not stated.
Products Worth Considering
Samsung 27-inch Odyssey OLED G6 QHD Gaming Monitor
Logitech MX Master 4 Ergonomic Wireless Mouse with Haptic Feedback, Graphite
LG UltraGear 34G600A-B 34-Inch Curved WQHD Gaming Monitor, 160Hz
As an affiliate, we earn on qualifying purchases.
From MSR to Microsoft AI
The leader said AI Frontiers was formed in 2023, after ChatGPT’s release, because he and colleagues believed AI research needed more coordinated effort. He said the group tried to organize shared research programs while working through disagreements and limited access to compute. In his view, the group was accomplishing pioneering work despite its relatively small size.
In early September, the group was reorganized into Microsoft AI. The leader acknowledged a possible rationale for combining AI organizations: concentrating effort and bringing together teams with different strengths. He also described Microsoft AI, which he said was much larger, as having a different focus from AI Frontiers. His account says the latter’s members pursued research, while Microsoft AI’s leaders favored scaling and work on data. The supplied source does not give the reorganization’s year, a formal company announcement, or an independent account from Microsoft.
“So AI Frontiers is over.”
— An AI Frontiers founding leader, writing on hunch.net
Staff and Project Status Remain Unclear
The available account does not establish how many AI Frontiers members left Microsoft, how many remain in other roles, or who now oversees the group’s research. It also does not say whether named projects, including AutoGen, Fara, or the model-development work, will continue under other teams. The claim that the founding leadership reached similar conclusions is the writer’s account; individual statements from those leaders are not included.
Microsoft’s reasoning for the reorganization and its response to the leader’s criticisms are also absent from the supplied material. The post describes a conflict over priorities and support, but it does not provide internal documents or a company statement that independently confirms those characterizations. The date and year of publication and the reorganization are not specified in the source excerpt.
Microsoft’s Next Research Structure
The next indication of what follows will be whether Microsoft or former AI Frontiers members publicly clarify where the team’s staff and projects have gone. A company statement or updated project information could establish whether the work is continuing elsewhere within Microsoft, moving outside the company, or stopping. No further announcement or timetable is given in the supplied account.
The immediate change identified by the group’s leader is the absence of AI Frontiers internships in the coming year. Further details about internship programs elsewhere at Microsoft, or any replacement for the group, remain unannounced in the material available here.
Key Questions
What happened to Microsoft Research’s AI Frontiers group?
A founding leader says AI Frontiers is over after the group was reorganized into Microsoft AI in early September. The source does not include a formal Microsoft announcement.
Why did the group’s leader leave?
He said he saw a mismatch between AI Frontiers’ research goals and Microsoft AI’s emphasis on scaling, and believed his work had no clear path to use. Those are his views, not a company explanation.
What research did AI Frontiers work on?
The leader cited AutoGen and Fara, along with research on model objectives, architectures, and pretraining systems. The account does not clarify which projects will continue.
Will AI Frontiers offer internships?
The leader said the group will not have internships in the coming year. The source does not say whether other Microsoft teams will offer related positions.
Source: hunch.net
