Copilot Cowork is Microsoft’s new cloud-based AI agent built with Anthropic's Claude technology that can plan, execute, and deliver multi-step work across Outlook, Teams, Excel, PowerPoint, and SharePoint on a user's behalf.

The product is the centrepiece of what Microsoft is calling Wave 3 of Microsoft 365 Copilot — a sweeping platform update that also brings Anthropic's Claude models into mainline Copilot Chat and introduces new enterprise pricing tiers. It draws directly on the same agentic harness as Anthropic's Claude Cowork, which had already been released for Mac in January 2026 and Windows in February.

Unlike Anthropic's desktop version, which runs locally on a user's device, Copilot Cowork operates entirely in the cloud within a customer's Microsoft 365 tenant. This means it is covered by Microsoft's enterprise data protection framework and connected to what Microsoft calls Work IQ — an intelligence layer that draws on the full graph of a user's work data, including email threads, Teams conversations, calendar history, SharePoint files, and Excel workbooks.

Microsoft president of Business Applications and Agents, Charles Lamanna, described the shift in a company blog post: "Copilot Cowork is built for that: it helps Copilot take action, not just chat." Users describe the outcome they want, and Cowork breaks the request into a structured plan that runs in the background, surfacing checkpoints for approval before any changes are applied.

In a demonstration, Lamanna showed Cowork analysing a month of meetings with direct reports, compiling customer notes from a business trip, and generating a competitive analysis with an accompanying Word document and Excel spreadsheet — without requiring manual coordination across apps.

The Anthropic partnership

The launch reflects the deepening relationship between Microsoft and Anthropic, formalised through a US$30 billion Azure compute deal announced in November 2025. Anthropic has operated as a Microsoft sub-processor since January 2026, covered by Microsoft's Product Terms, Data Processing Agreement, and Enterprise Data Protection framework.

Microsoft chief marketing officer for AI at Work, Jared Spataro, said the multi-model approach was a deliberate differentiator: “Every 60 days at least, there's a new king of the hill. There's so much demand for a platform that doesn't feel like, 'I have to skip over to the next vendor.”

Claude handles the complex reasoning components of multi-step tasks, while Microsoft's own models manage integration with M365 applications.

Fortune reported that Spataro drew a direct distinction between Anthropic's consumer product and the enterprise offering, describing the desktop version as a "fantastic tool" but one with "limitations" in a corporate environment, given its lack of access to cloud-based enterprise data and governance controls.

Availability and pricing

Copilot Cowork entered Research Preview on 9 March with a limited set of customers. Broader availability through Microsoft's Frontier program is expected in late March 2026. Accessing the feature requires a Microsoft 365 Copilot licence at US$30 per user per month on top of existing enterprise subscriptions.

Microsoft also announced the Microsoft 365 E7 suite, which will be generally available from 1 May 2026 at US$99 per user per month. E7 bundles Microsoft 365 Copilot, Agent 365, the Microsoft Entra Suite, and Microsoft 365 E5 security capabilities under a single plan — the first new enterprise licence tier in approximately 10 years.

Agent 365, the governance and orchestration platform for AI agents, will also be available as a standalone add-on at US$15 per user per month from 1 May.

Claude models are currently available in Copilot Chat, Researcher, and Excel for tenants holding a Copilot licence, though administrators must enable the Anthropic model in Copilot settings. Microsoft notes that Copilot Cowork itself remains in Research Preview, and no firm commercial availability date for general rollout has been committed.

A specific gap for A/NZ users

For Australian and New Zealand organisations, the governance picture requires careful mapping before adoption.

The Anthropic sub-processor toggle was enabled by default for most commercial tenants on 7 January 2026. EU, EFTA, and UK tenants have it disabled by default. Government cloud environments have no access to the feature.

Microsoft made in-country Copilot processing available for Australia in late 2025, meaning standard Copilot interactions can be processed and stored within Australian borders.

However, Anthropic models are explicitly excluded from those in-country commitments. Any request routed through Claude — including Cowork task execution — falls outside that data residency guarantee.

For New Zealand organisations, the same commercial tenant rules apply: the sub-processor toggle is on by default, and there are no equivalent in-country processing commitments that would cover Anthropic model use.

John Hoy, Founder of Orbital Intelligence and a member of the Governance Institute of Australia, has posted a detailed analysis of the governance implications for Australian and New Zealand enterprises on LinkedIn.

Hoy identifies four questions that M365 Copilot organisations need to address before enabling the feature: whether the Anthropic sub-processor toggle has been reviewed; which Copilot features route to Claude; whether data flows have been mapped; and whether existing AI governance frameworks account for a multi-model environment.

"Your standard Copilot interactions can stay onshore, but the moment a request routes to Claude, that guarantee falls away," Hoy writes.

"If in-country processing underpins your compliance posture, this needs attention before adoption."

Hoy tested Copilot Cowork extensively in advance of the announcement and described the capability as a genuine productivity force multiplier, noting a file management task that would have taken three hours was completed by the agent in 47 seconds.

But he was unequivocal that capability should not pre-empt governance: "Impressive capability sure, but there's a real governance gap here worth getting right before adoption."

Governance and enterprise readiness

Independent analysis from enterprise consultancy ARO echoes those concerns. In a post-announcement commentary, Paul Ryder of ARO argued that while the capability is a significant step toward integrating agents into daily processes, most organisations are not structurally or operationally ready for this level of automation. "The readiness gap matters far more than the technology itself," he wrote, pointing to the need for data quality, clear automation boundaries, defined ownership models, and training investment before broad adoption.

Governance researcher and enterprise security firm Pragatix identified an additional technical limitation: Copilot Cowork executes actions using the identity of the user rather than a dedicated AI agent identity, creating ambiguity in audit trails.

When an agent acts autonomously on a user's behalf — sending an email, modifying SharePoint permissions, updating a project timeline — existing compliance frameworks that assume a direct link between a user identity and a system action may not adequately capture accountability.

Microsoft has addressed the broader governance architecture in its product documentation. Cowork runs in a sandboxed cloud environment within Microsoft 365's security and compliance boundaries, with identity, permissions, and compliance policies enforced by default. Actions and outputs are auditable. The agent cannot perform actions that fall outside the permissions already held by the delegating user's account.

For those evaluating adoption, the immediate priority is understanding which Copilot features within their tenant currently route to Claude models, confirming whether the Anthropic sub-processor toggle has been reviewed against their data governance framework, and determining whether their AI governance policies have been updated to address multi-model agentic environments.

Market context

Microsoft's launch follows significant market disruption caused by Anthropic's standalone Claude Cowork releases in early 2026, which triggered a combined US$285 billion selloff in enterprise software stocks as investors repriced companies whose core workflows overlapped with what the desktop AI agent could automate.

With paid Copilot adoption still representing a small fraction of Microsoft's 450 million commercial Microsoft 365 subscribers, Cowork is widely seen as Microsoft's most significant effort to justify the Copilot licence cost and accelerate enterprise take-up.

Analyst commentary collected by Winbuzzer described the launch as Microsoft absorbing a competitive threat, repackaging Anthropic's technology within Microsoft's enterprise governance architecture, and using the resulting product to address both AI adoption and licence conversion.

Microsoft has also confirmed that Anthropic's Claude models are now available across the full Copilot Chat experience for Frontier program users, not solely in the Researcher and Excel features where Claude was previously integrated — meaning Anthropic's models are becoming a general-purpose option for everyday Copilot interactions, not only for task execution in Cowork.