Microsoft Research's AI Frontiers (AIF) sub-organization, formed in 2023 with approximately 50 researchers to coordinate agent-focused efforts like the AutoGen orchestration framework and FARA computer-use agent, has been dissolved following a reorganization into Microsoft AI (MAI) under ex-Inflection leadership. AIF pursued fundamental capability research including the Next Latent objective for extrapolation, the Dion3 optimizer improving on Muon, and the Full Bandwidth Transformer enabling efficient recursion; Bytedance experiments suggest the latter could match models twice its size, potentially halving training and inference costs. AIF also built a shared monorepo with a proven edge-scale pretraining pipeline to reduce duplication across MSR. The author argues MAI's "scaling-only" philosophy—prioritizing data refinement over architectural innovation—is incompatible with research-driven advancement, citing predictable returns to scale versus uncertain research returns. After failing to integrate AIF's work into MAI and seeing no viable path to production from MSR due to the non-consensual reorganization breaking the implicit "good work yields good outcomes" contract, the entire founding leadership departed. The author contends returns to research are significant given industry investments approaching a trillion dollars, and expresses skepticism about MSR's ability to organize effectively without leadership changes.