Cowork
Cowork is collaborative work with agents in the loop — humans still own the outcome. Treat it as an operating model (and an evolving Copilot capability), not a mystery SKU.
What it is
In Microsoft’s agent narrative, Cowork (also referenced as Copilot Cowork in roadmap and feature materials) describes multi-step outcomes where agents draft, fetch and act across Microsoft 365 — mail, meetings, messages, files and related data — with people approving sensitive steps. Microsoft release notes and feature matrices describe Planner capabilities in Cowork and government-cloud variants as they roll out. Confirm current GA vs preview status on Learn / Roadmap before you promise dates.
Where it sits in the stack
Cowork sits in the Copilot / agent experience layer — above Chat, alongside Microsoft 365 Copilot, fed by Graph grounding. Studio and Foundry can supply specialised agents that participate in longer workflows.
Key features / capabilities
- Delegate multi-step tasks with grounding in email, meetings, messages and files (as licensed).
- Approval loops before sensitive actions.
- Integration with Planner and related work surfaces where released.
- Same permission and compliance boundaries as other Copilot experiences.
- Pairs with Agent 365 observability on E7-class programmes.
Who it is for / when to use
Knowledge workers with repeatable multi-step jobs and a clear human owner. Not for unsupervised customer-facing automation on day one.
Operating model before branding
Whether your tenant sees Cowork branding this month or next, the operating model is stable: human-led, agent-operated, approval before sensitive side effects. Write that into RACI charts. Map each Cowork-style workflow to a production-grade owner, measure and kill switch. Suite context (especially E7 + Agent 365) changes how you observe agents — it does not remove the need for human accountability.
Cost layer
Usually requires Microsoft 365 Copilot (or E7). Agent actions may still consume Studio or other meters — map them. See Managing cost and the feature matrix.
Risk and governance
Define what the agent may do without asking, what needs approval, and the kill switch. Log actions. More: Managing risk.
Adoption tip
Start with one workflow: “turn this thread into tasks and a draft update” with mandatory human send. Habits: Adoption.
Related Ideas
Scout / research · M365 Copilot · Agent Builder · E3 vs E5 vs E7
Source note: check Microsoft Learn and Product Terms before you brief finance or legal. Products, names and meters change. This site is independent commentary, not Microsoft official documentation.