Microsoft made Copilot Cowork generally available worldwide on June 16, 2026, ending a roughly three-month Frontier preview that began in March. For IT teams, this is the moment an experimental agent becomes a real budget, security, and governance decision.
Unlike a chat feature that returns a draft, Copilot Cowork is built to hand back completed work from long-running, multi-step tasks. That shift changes how you should think about permissions, cost, and rollout. So here is a practical, grounded rundown of what it does and what you should check before switching it on.
What Is Copilot Cowork?
Copilot Cowork is an agentic capability that lives inside Microsoft 365 Copilot. Where standard Copilot summarizes or drafts, Cowork is designed to run a whole task end to end and return a finished result.
Crucially, it runs in Microsoft’s cloud rather than on your PC. According to Computerworld, that means it acts on documents held in your Microsoft 365 tenant and can keep working even when your computer is off. It is built on the same underlying technology as Anthropic’s Claude Cowork, but tied into Microsoft 365 data, permissions, and billing.
Microsoft says uptake during the preview was broad. The company reports that more than half of the Fortune 500 used Copilot Cowork in Frontier, naming customers such as Accenture, Avanade, Capital Group, Ooredoo Qatar, and Zurich Insurance.
In short: Copilot Cowork is an AI agent that completes multi-step office work inside your Microsoft 365 tenant — not just another text box that writes a draft.
What Copilot Cowork Can Do in Microsoft 365
The reach is the point. Within Microsoft 365, Copilot Cowork can carry out a wide range of actions, including:
- Create Word, Excel, PowerPoint, PDF, and other documents.
- Send emails, manage calendars, and schedule meetings.
- Post in Teams and draft stakeholder communications.
- Search across an organization, conduct research, and prepare briefings.
Behind this sits Work IQ, Microsoft’s context layer for Microsoft 365 data. It supplies business context from email, calendars, meetings, chats, files, people, collaboration patterns, and line-of-business systems. In its preview examples, Microsoft says customers used Cowork to compare thousands of files, automate spreadsheet-heavy workflows, generate dependency charts, and flag stalled sales opportunities.
Because a single task can move through calendars, files, chats, and business systems before producing a result, the agent is far more consequential than a text generator — which is exactly why permissions matter.
How Permissions and Governance Work
This is the part admins should read twice. Copilot Cowork inherits the signed-in user’s permissions, so it can only reach the files and emails that user could already open. It does not grant new access.
Several guardrails shape how far the delegation goes:
- Sensitive actions — such as sending email, posting in Teams, or scheduling meetings — require approval.
- Cowork is off by default after general availability.
- Administrators can disable it for specific users, control deployment and availability, and turn off individual models, including Anthropic models.
- Admins can set spending limits and usage alerts.
The Security Angle Admins Should Watch
Inherited permissions are a sensible default, but they are not a complete safety net. Security firm PromptArmor flagged inherited-permission risks during testing, warning that workflows could expose file links through self-directed Teams or Outlook messages.
The practical takeaway is simple. Before you enable Cowork broadly, review who has access to what in your tenant, because an agent acting on a user’s behalf is only ever as contained as that user’s existing permissions.
The Models Behind Copilot Cowork
At launch, Copilot Cowork runs on Anthropic’s Opus 4.8 and Sonnet 4.6 models, and Frontier customers can also use GPT-5.5. Microsoft expects its own in-house model, Cowork 1, to arrive in the coming weeks. Computerworld also reports that Microsoft plans to offer an open-source DeepSeek model to help lower costs.
For admins, the model choice is more than a technical footnote. Because you can turn individual models off, you get a lever over both behavior and cost — which leads directly to billing.
Pricing: Copilot Credits and Usage-Based Billing
Here is the biggest change from standard Copilot. Billing for Cowork is separate from the Microsoft 365 Copilot subscription. Tasks consume Copilot Credits, Microsoft’s usage unit for agent work.
Each task is priced from a mix of factors: model use, context retrieval, tool calls, and runtime. Related Work IQ APIs use consumption-based pricing with fixed and variable components. In other words, repeated or tool-heavy workflows become a budget variable rather than a fixed cost.
Microsoft has shared a cost comparison, but read it carefully. In an internal benchmark of 125 runs across 12 work-data prompts, the company put Cowork at 30% to 40% cheaper per prompt than Claude Cowork using its Microsoft 365 connector. That is a vendor benchmark, not an independent, market-wide price claim — treat it as a starting hypothesis to validate against your own usage.
An Admin Checklist Before You Switch It On
If you are weighing Copilot Cowork for your tenant, a short pre-flight check goes a long way:
- Keep it off until you are ready. It is disabled by default — enable it deliberately, for a pilot group first.
- Audit permissions. Review file and mailbox access for pilot users, since the agent inherits exactly those rights.
- Set spending limits and alerts. Usage-based billing rewards guardrails; configure them on day one.
- Decide your model policy. Choose which models are allowed and disable the rest.
- Define approval expectations. Make sure pilot users understand which actions pause for human approval.
Used well, an agent that delivers finished work can absorb genuinely tedious tasks. The job for IT is to make sure it does so without surprise costs or permission leaks.
The Bottom Line
Copilot Cowork marks a real step from assistive AI to agentic AI inside Microsoft 365. The capability is powerful, the permission model is sensible, and the usage-based pricing means cost discipline now sits with administrators.
As Microsoft’s Charles Lamanna put it on launch day, “Copilot Cowork is now generally available.” The smart move for IT teams is not to rush, but to pilot it deliberately — permissions audited, budgets capped, and models chosen — before letting an agent loose on real work.
Sources: Microsoft 365 Roadmap (microsoft.com/microsoft-365/roadmap); WinBuzzer, “Microsoft Opens Copilot Cowork for Delegated Office Work,” June 17, 2026; Computerworld, “Microsoft launches Copilot Cowork with usage-based pricing,” June 17, 2026.