# Full Bandwidth Transformer lost its lab _John Langford says Microsoft folded AI Frontiers into Microsoft AI, and he left over research versus scaling._ **Published:** 2026-11-08 **Source:** https://www.startuphub.ai/ai-news/public-companies/2026/full-bandwidth-transformer-lost-its-lab --- John Langford, the Microsoft Research scientist who helped found AI Frontiers, said he has left Microsoft AI after the group was absorbed in a [Hunch](https://hunch.net/?p=13764006) post. ChatGPT Shock 2023DriverFrom the articleLangford said AI Frontiers formed in 2023 after ChatGPT because one person could not move fast enough on baselines.sparkedAI Frontiers FoundedCoremicrosoft research group formed in 2023 to accelerate baseline progressFrom the article 5 mentionsJohn Langford, the Microsoft Research scientist who helped found AI Frontiers, said he has left Microsoft AI after the group was absorbed in a Hunch post.builtBuilt Core TechCoreautogen for orchestration plus fara for computer use capabilities builtthenFolded Into Microsoft AIDriverFrom the article 4 mentionsIn early September Microsoft folded AI Frontiers into Microsoft AI, a much larger group led by ex-Inflection leadership.promptedLangford Left MicrosoftOutcomefounder departed after absorption declared ai frontiers is over in hunch postFrom the article 2 mentionsJohn Langford, the Microsoft Research scientist who helped found AI Frontiers, said he has left Microsoft AI after the group was absorbed in a Hunch post.ChatGPT Shock 2023DriverFrom the articleLangford said AI Frontiers formed in 2023 after ChatGPT because one person could not move fast enough on baselines.sparkedAI Frontiers FoundedCoremicrosoft research group formed in 2023 to accelerate baseline progressFrom the article 5 mentionsJohn Langford, the Microsoft Research scientist who helped found AI Frontiers, said he has left Microsoft AI after the group was absorbed in a Hunch post.Shared Monorepo DefaultContextFrom the articleLangford said AI Frontiers shared a monorepo by default across Microsoft Research, including an edge scale pretraining pipeline from his COLM keynote.Built Core TechCoreautogen for orchestration plus fara for computer use capabilities builtFull Bandwidth TransformerCoreFrom the article 5 mentionsNew York work included Next Latent, Dion3 as a Muon successor, and the Full Bandwidth Transformer which adds power through recursion, according to the Hunch post.Folded Into Microsoft AIDriverFrom the article 4 mentionsIn early September Microsoft folded AI Frontiers into Microsoft AI, a much larger group led by ex-Inflection leadership.Scaling Philosophy ClashDriverleadership holds bigger models are better models versus research approachLangford Left MicrosoftOutcomefounder departed after absorption declared ai frontiers is over in hunch postFrom the article 2 mentionsJohn Langford, the Microsoft Research scientist who helped found AI Frontiers, said he has left Microsoft AI after the group was absorbed in a Hunch post. AI Frontiers is over. Langford said AI Frontiers formed in 2023 after ChatGPT because one person could not move fast enough on baselines. The group built Autogen for orchestration and Fara for computer use. New York work included Next Latent, Dion3 as a Muon successor, and the Full Bandwidth Transformer which adds power through recursion, according to the Hunch post. Langford said AI Frontiers shared a monorepo by default across Microsoft Research, including an edge scale pretraining pipeline from his COLM keynote. He said he knew of at least four model building efforts inside Microsoft, with only one shared, as of November 2026. In early September Microsoft folded AI Frontiers into [Microsoft AI](https://www.startuphub.ai/ai-news/public-companies/2026/tigress-s-ivan-feinseth-on-ai-ipos-and-tech-stock-picks), a much larger group led by ex-Inflection leadership. Langford said that leadership holds that bigger models are better models and that returns to scale beat returns to research. He called that scaling-only view misdirected and not fun. That cost claim for the Full Bandwidth Transformer now has two external reference points. Langford cited ByteDance experiments suggesting a Full Bandwidth Transformer variant can match a model twice its size. A listing per [Hugging Face](https://www.startuphub.ai/cybersecurity/hackerone-ceo-openai-ai-hack-was-responsible-testing) says Full Bandwidth Transformers match standard transformers trained with roughly 1.5 times more tokens with negligible per-token decoding overhead and shorter reasoning traces at equal or better accuracy. Jae Kwon, a commenter pitching startup gnosh.ai, said he is raising and will hire from the team. Langford said every other founding AI Frontiers leader reached a similar exit decision and that internships will not run next year. Who now pays for recursion work like the Full Bandwidth Transformer when scaling owns the budget? --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.