Microsoft has launched Copilot Cowork within the Microsoft 365 Frontier program, enabling autonomous, multi-step task execution across applications such as Excel, Outlook, and SharePoint. The system is grounded in Anthropic’s Cowork technology and uses a multi-model approach combining OpenAI’s GPT and Anthropic’s Claude for more accurate research outputs.
The new capability is designed to handle long-running, multi-step workflows autonomously. Until now, Copilot has focused on generative tasks such as drafting emails or summarizing documents. These are single-shot outputs that still require a person to stitch the workload together to reap the benefits of automation. Cowork helps to actually automate certain tasks. Users describe a desired outcome, and the system draws up a plan and executes it across Microsoft 365 applications without needing a human to prompt each step.
The announcement was made today, and is part of Wave 3 of Microsoft 365 Copilot, which Microsoft describes as a turning point toward embedded, agentic AI in the workplace. Copilot Cowork is currently available through the company’s Frontier program, which gives enterprises early access to cutting-edge features. Ultimately, it will likely be a cornerstone of the new E7 tier of Microsoft 365, the highest echelon of the company’s subscription-based workplace suite.
Built on Anthropic’s Cowork
The technology underpinning the new feature is familiar, as is the naming scheme. Anthropic launched Cowork in January as an agentic tool for broader knowledge work, building on the same principles as Claude Code but targeting non-technical users. Plug-in support followed in February, broadening its enterprise appeal. Microsoft has since integrated that same technology platform directly into Microsoft 365 Copilot.
Within Microsoft 365, Copilot Cowork acts as an orchestrator, reasoning across files and applications including Excel, Outlook, Teams, and SharePoint. A task such as a monthly budget review, which normally requires jumping between spreadsheets, emails, and documents, can be handed off entirely. The system operates within the Work IQ framework, which is grounded in an organization’s data while observing its security and governance boundaries. Human oversight is preserved throughout: users can monitor progress and redirect the agent if it goes off track.
Barton Warner, SVP of Enterprise Technology at Capital Group, one of the first organizations to use the system, said: “It’s connecting steps, coordinating tasks and following through across everyday workflows.”
Multi-model Researcher
Copilot Cowork also introduces changes to the Researcher agent. It now uses a “critique” layer that pairs OpenAI’s GPT models and Anthropic’s Claude. One model generates an initial response; the other reviews it for accuracy and citation quality. Microsoft says this has lifted the Researcher’s score on the DRACO benchmark, the industry measure for deep research quality, by 13.8 percent. Roles can be reversed, and a new “model council” feature lets users compare outputs side by side.
Beyond benchmarks, however, the new tooling will have to prove to users it is more beneficial than the Copilot they know and (in many cases) don’t quite love. More truly agentic features that take over rote work will be appealing, at any rate, and tooling like OpenClaw have shown that autonomous actions can be varied and expansive. The Frontier program currently provides access to Cowork ahead of a wider rollout, hopefully to ensure a full release won’t feature the same security flaws seen in a tool like OpenClaw, instead offering enterprise-grade protection.