Microsoft's Copilot Cowork is now available to Microsoft 365 Frontier program members, marking the shift from announcement to live product for one of the year's most significant enterprise AI agent releases.

What Cowork Does

Cowork handles long-running, multi-step work inside M365. Users describe an outcome, and the system builds a plan, reasons across files and tools, and carries work forward autonomously โ€” from monthly budget reviews to calendar management and executive briefing prep. Early adopter Capital Group says the system is already delivering value on planning, scheduling, and deliverable creation.

Claude Does the Reviewing

Cowork ships with skills from both Microsoft and Anthropic built in. The new Researcher Critique feature splits deep research into two phases: one model plans and drafts, a second (Claude) reviews for factual accuracy, source reliability, and completeness before the final report is delivered. On the DRACO benchmark โ€” the industry standard for deep research quality across 100 tasks in 10 domains โ€” Researcher with Critique scores 13.8% higher than the best prior system in the reference paper.

A companion Council feature lets users compare responses from multiple models side-by-side, with a summary highlighting where models agree and where they diverge.

Wave 3

Microsoft calls this part of "Wave 3" of Microsoft 365 Copilot โ€” a push toward agentic, multi-step AI workflows that can operate autonomously over hours, not just generate single responses. Frontier access is available now; broader rollout timelines have not been announced.