From Asking to Acting: Why Copilot Cowork (and Claude) Redefine the Job of AI

Tuesday 16th June 2026

Microsoft 365 Copilot Cowork moves into general availability today. It has been running in early access through the Frontier programme since late March, with the commercial pieces landing in sequence - Dynamics 365 Sales skills reaching GA on 15 June, and the broader connector set and pricing following close behind.

If you run a Microsoft estate, this is not just another release to note. It changes what Copilot is for.

For the past eighteen months, Copilot has largely been reactive. You ask, it answers. Summarise this thread. Draft that email. Pull those numbers together.

Cowork shifts it from answering to doing. You describe an outcome, and it plans the steps, works across your mail, calendar, Teams and documents, and brings back a finished result - pausing for approval where it matters.

That is a genuine shift, not a feature bump.

From assistant to execution layer

The simplest way to understand it is this:

Copilot = help me do the task

Cowork = go and do the task for me

Under the surface, that change is more structural than it looks. Cowork isn’t just generating content - it is planning and executing multi-step work across your Microsoft 365 environment. It can connect actions across tools: generate a document, trigger an approval, update a system, notify the team.

That turns AI into something closer to an operational layer. Work doesn’t just move faster - it actually moves without you.

And crucially, it does that inside your existing tenant, using your data, permissions and policies.

They’re not the rivals they look like

There is another product with the same word in its name: Claude Cowork.

The instinct is to compare the two and pick a winner. That instinct misses the point.

Microsoft has been explicit that Copilot is now running as a multi‑model system, incorporating external models alongside its own. In practical terms, these tools share underlying capabilities. What differs is where they’re applied.

So the real question is not which one is better.
It’s which surface fits the job.

What each one actually is:

Copilot Cowork lives inside Microsoft 365. It operates across your tenant - Outlook, Teams, SharePoint, Word, Excel - and extends outward through connectors into systems like Dynamics and Power BI. It inherits your identity, permissions, and compliance controls by default, and every action sits inside that boundary.

It also introduces a different commercial model. You still have a per-user Copilot license, but the agentic work draws down on a consumption layer - Copilot Credits - meaning cost scales with how much work you delegate.

Claude Cowork, by contrast, lives on the desktop. It works outward from the user - your files, your connected apps, your workflows. The setup is lighter and the entry point lower, but it sits alongside your enterprise controls rather than inside them.

Read those two descriptions again and the distinction is clear.
One is built around organisational execution.
The other is built around individual productivity.

Where they genuinely differ:

Four things matter when you choose how to use them.

Where the work lives
Copilot Cowork runs where your corporate data already sits - inside the tenant. Claude Cowork runs from the desktop outward. If the work is organisational, governed, shared, Copilot’s position is hard to ignore.

Governance and control
Copilot Cowork inherits your security, compliance and identity model automatically. Every action stays within that boundary and can be monitored centrally. Claude brings enterprise controls, but they sit alongside your Microsoft estate rather than within it. For regulated organisations, that distinction matters.

Cost shape
Copilot introduces a usage curve. Light use is low cost; heavier automation scales with value delivered. That gives flexibility - but it also means you need to manage it. Microsoft is effectively bringing FinOps thinking into AI: monitor usage, set guardrails, treat it like any other consumption service. Claude’s model is simpler, but less elastic.

Reach and adoption
Copilot Cowork benefits from gravity - it is already where your people work. No new tools, no behavioural shift required.

Claude Cowork benefits from independence - it can sit across environments and suit individuals who want capability without platform commitment.

What actually changes now

The headline is general availability.

The more important change is quieter.

For the first time, AI work is not just measurable in “time saved” - it is measurable in output delivered. When Cowork completes a process, that can be tracked and evaluated like any other unit of work.

That opens up a different conversation. Not “does AI help” - but “where should AI own the process?”

It also introduces real operational considerations:

Billing must be configured (via Azure subscription) to continue usage beyond June

Usage needs to be monitored and governed

Early use cases matter - structured, repeatable tasks deliver the first wins

This is no longer experimentation. It is implementation.

The honest read

I work in a Microsoft ecosystem, so the default landing point is obvious: for organisations standardised on Microsoft 365, Copilot Cowork is the natural home for this capability. It places execution where the work already happens and keeps it within the controls you already operate.

But that is not the whole picture.

The fact that these tools share underlying capabilities is the signal. The shift to agentic AI is bigger than any one vendor. The right stance is not platform loyalty - it is clarity about the task.

Most organisations will end up using both:

Copilot Cowork for work that belongs to the organisation

Desktop agents for work that belongs to the individual

That split will become normal.

The moment to pay attention

We have spent two years getting comfortable with AI that answers.

We are now being handed AI that executes - planning steps, taking actions, and completing work on our behalf.

The decisions that matter are no longer about which model is smartest.
They are about where you let AI operate, what you trust it to do, and how closely you watch it while it does it.

If you want a clear view of where Cowork fits in your estate (and what it’s likely to cost and return), let’s map it properly before you switch it on at scale

Want to keep informed? Sign up to our Newsletter