Microsoft on March 30 opened Copilot Cowork to customers in its Frontier early-access program, making the Anthropic-powered persistent agent feature available for the first time outside of a limited pilot. Copilot Cowork handles long-running, multi-step tasks within Microsoft 365 — delegating work across apps, reasoning over files and tools, and reporting progress back to the user.
The product was first announced on March 9 as part of Wave 3 of Microsoft 365 Copilot. Built in close collaboration with Anthropic, Copilot Cowork uses Claude as its reasoning engine and runs on the same agentic harness that powers Claude Cowork, according to Fortune. The critical difference: Microsoft’s version runs in the cloud within a customer’s M365 tenant, not locally on a user’s device.
“We actually don’t work locally, and that’s a feature, not a bug,” Jared Spataro, Microsoft’s chief marketing officer for AI at Work, told Fortune at the March 9 announcement. He described Anthropic’s Claude Cowork as “a fantastic tool” with “limitations” in corporate environments, citing its lack of access to cloud-based enterprise data and security concerns at scale.
What Copilot Cowork Does
Users describe the outcome they want, and Cowork creates a plan, reasons across tools and files, and executes — with visible progress and opportunities to steer, per the Microsoft 365 blog. Built-in skills from both Claude and Microsoft include calendar management and daily briefing. The system handles both one-off tasks and repeatable workflows like monthly budget reviews.
Capital Group, one of the early-access organizations, is already using Cowork for planning, scheduling, creating deliverables, and preparing for executive reviews. “This isn’t about generating content or answers. It’s about taking real action — connecting steps, coordinating tasks, and following through across everyday workflows,” Barton Warner, SVP of Enterprise Technology at Capital Group, said in the Microsoft blog announcement.
Because Cowork operates within the customer’s M365 tenant, it’s covered by Microsoft’s enterprise data protection and has access to what the company calls “Work IQ” — intelligence drawn from emails, files, documents, meetings, and chats.
Multi-Model Researcher Upgrade
Alongside the Cowork launch, Microsoft announced upgrades to Researcher that use models from multiple frontier labs. A new Critique feature separates generation from evaluation: one model plans and drafts, while a second model (from Anthropic or OpenAI) reviews and refines the output before the final report is produced. Microsoft reported a 13.8% improvement on the DRACO benchmark (Deep Research Accuracy, Completeness, and Objectivity), per the blog post.
A new model Council feature lets users compare responses from different models side by side, showing where models agree, diverge, and what each contributes uniquely.
Why It Matters
Copilot Cowork’s Frontier launch puts persistent agent capabilities into the hands of enterprise early adopters at Microsoft’s distribution scale. The Frontier program is invite-only, but Microsoft’s M365 install base means millions of seats could activate this within a quarter once it reaches general availability.
For the enterprise agent market, this is Microsoft’s clearest signal that the era of one-shot Copilot completions is giving way to sustained, multi-step autonomous workflows. The question is how quickly Frontier participants demonstrate ROI that justifies broader rollout — and whether the enterprise data governance frameworks announced at last week’s RSAC 2026 can keep pace with the agent capabilities Microsoft is deploying.