Microsoft Research
AI Frontiers
Who We Are
Update, Sept. 14, 2026 – Microsoft Research AI Frontiers is now part of Microsoft AI (opens in new tab)!
Our lab has long focused on advancing frontier research in efficient language models and agentic systems, while reimagining existing computing through an agent-native lens. Through the years, we’ve shipped AgentInstruct, Phi family of models (opens in new tab), Fara portfolio of CUA models (opens in new tab), the Agent Framework AutoGen (opens in new tab), Magentic-One, MagenticUI, Magentic Marketplace. We have also led pioneering work in agent red teaming, benchmarking, and safety research. Collectively, these contributions have delivered significant value to both the academic community and the broader open-source ecosystem.
We’ll pursue our same goal of frontier and cutting edge agentic and model research at Microsoft AI (MAI) starting now. MAI is a natural fit for our work, given MAI’s focus on humanist superintelligence and building AI systems focused on real problems and in service to people.
We’ll share more exciting news on this transition soon!
How We Work
At AI Frontiers, we combine the freedom to pursue high-impact ideas with a culture of focused, collaborative execution. We champion openness by publishing our research and contributing to open source, while driving work that advances science and delivers real-world impact.
Our work is guided by three principles: clarity of vision, execution excellence, and deep collaboration. Together, these values enable us to move fast, adapt boldly, and work as One Microsoft to advance science and empower the world.
Our Mission
We believe the true transformative promise of GenAI will only happen when we start looking at the world through an agent-native lens.
This means stepping back to envision a world where agents and humans co-exist and interact seamlessly across personal and work contexts—free from the constraints of existing paradigms and processes.
Our Research Focus
Adaptive Agents
Closing the capability gap with long-running agents that learn, reason, and adapt across tasks, contexts, and modalities.
Agentic Robustness
Closing the reliability gap to ensure consistent, trustworthy performance in real-world conditions
Agentic Ecosystems
Reimagining workflows for an agent-native world where autonomous agents collaborate seamlessly with humans.