What did Microsoft actually launch?
An agentic layer inside Microsoft 365 Copilot, called Copilot Cowork. Instead of chatting with an assistant, you delegate an outcome: Cowork plans the multi-step work, reasons across your files and tools in Word, Excel, PowerPoint, Outlook and Teams, runs long tasks in the cloud (they keep going when your laptop is shut), and shows progress you can steer. It was unveiled in March 2026 and reached general availability worldwide on 16 June 2026. Microsoft says more than half of the Fortune 500 used it during the preview - the vendor's own figure, but a fair signal of where the market is going.
The headline for anyone who follows this space is in Microsoft's own words: "Working closely with Anthropic, we have brought the technology that powers Claude Cowork into Microsoft 365 Copilot."
Is it the same as Claude?
Under the hood, largely yes. At general availability, Copilot Cowork runs only on Anthropic models - Claude Opus 4.8 and Sonnet 4.6 - the same models behind Anthropic's own Claude Cowork. Its skills use the same open Agent Skills standard as Claude, and its plugins connect to external systems over MCP - the protocol Anthropic invented and donated. A skill your team builds for Claude is structurally reusable in Cowork, and the other way round.
What Microsoft adds is the wrapper: grounding in your organisation's Microsoft 365 data, tenant-level governance, admin-deployed plugins, and metered billing. What it changes is the choice you face. This isn't "Copilot or Claude" any more - it's the same Claude capability, Claude-native or inside Microsoft 365. And because the skills standard is shared, the capability your team builds travels with you either way. The skills are the asset, not the licence.
How is it priced?
Two layers, and this is where MDs should slow down. First, the gate: Cowork requires the full Microsoft 365 Copilot licence per user - in the UK, roughly £16-£25 per user per month depending on your plan. Then, the meter: Cowork usage itself bills on top in Copilot Credits, priced per task on model use, retrieval, tool calls and runtime - pay-as-you-go at $0.01 per credit, or a discounted advance commitment. Admins get spending limits, alerts and usage reporting, and Microsoft publishes a spend estimator.
Claude's commercial model is the opposite: a flat per-seat subscription, so the bill per head is predictable however hard your team works it. Neither model is automatically better - but a metered agent without governance is how a pilot turns into an unwelcome invoice. Microsoft's own benchmark claims Cowork averaged 30-40% cheaper than Claude Cowork with its Microsoft 365 connector; treat that as vendor positioning - it's Microsoft's internal test, 125 runs, June 2026.
Why is it switched off for UK organisations?
Because of where the data is processed. Anthropic models in Microsoft's stack are currently excluded from the EU Data Boundary and in-country processing commitments, and are disabled by default for UK tenants. Since Cowork at launch runs only on Anthropic models, the practical effect is blunt: a UK organisation cannot use Copilot Cowork at all until a Global or AI Administrator explicitly opts in - Microsoft 365 admin centre, Copilot settings, "AI providers operating as Microsoft subprocessors" - accepting processing outside the UK and EU, scoped per user or group.
That opt-in is a governance decision, not a toggle. Anthropic is onboarded as a Microsoft subprocessor under Microsoft's standard data protection terms, and prompts and outputs flow through your existing Microsoft 365 controls - but note that Data Loss Prevention support was still "coming soon" at launch. If your firm has already thought through where your human checkpoints and accountability sit, this is the same conversation with a new setting attached.
What should an MD do about it?
Three things, in order.
- Decide the platform deliberately. Same Claude capability, two wrappers: Claude-native, with flat per-seat pricing and your own governance; or inside Microsoft 365, with Graph grounding, a licence gate, credits and the UK opt-in. The right answer depends on your stack, your data posture and your team - not on which salesperson called first.
- If it's Copilot Cowork, govern the switch-on. Scope who gets access, set the spending limits before the first task runs, and record the data-residency decision like the board-level call it is.
- Either way, build portable capability. Train your people and build your skills on the open standard both platforms now share. The firms that win this year aren't the ones with the most licences - they're the ones whose teams know how to direct the agents.
That last part is the work I do. The AI Adoption Agency is Scotland's only dedicated Claude adoption specialist - and because Microsoft has now built Copilot Cowork on Claude, that expertise applies whichever wrapper you choose: Claude-native or inside Microsoft 365.
Frequently asked questions
Is Copilot Cowork the same as Claude Cowork?
Same engine, different wrapper. At launch Copilot Cowork runs exclusively on Anthropic's Claude models (Opus 4.8 and Sonnet 4.6) and uses the same open skills standard, so skills transfer between the two. Microsoft adds Microsoft 365 grounding, tenant governance and credit-based metering; Anthropic's version is flat per-seat.
Can UK businesses use Copilot Cowork today?
Not until an admin acts. Anthropic models are disabled by default for UK tenants because they sit outside the EU Data Boundary, and Cowork currently runs only on Anthropic models. A Global or AI Administrator must opt in via the Microsoft 365 admin centre, and can scope access per user or group.
Do skills built for Claude work in Copilot Cowork?
Structurally, yes. Both use the Agent Skills open standard - the same SKILL.md format - and both reach external tools over MCP. Deployment differs: Claude users can install skills and plugins themselves, while Cowork plugins are deployed by admins at tenant level. Build once, and the capability travels.
If you're weighing up which side of this to build on, that's a 30-minute conversation, not a research project - book a discovery call.