Using Copilot agents in Teams

Alongside general-purpose Copilot, Teams increasingly surfaces agents — narrower assistants built for one job: answering questions about a specific policy document, walking through an internal process, or acting as a front door to a particular team's knowledge. As an end user, using one looks almost exactly like using Copilot itself.

Finding an agent

Agents show up in a few places:

• The Teams app store, searchable like any other app, under an Agents or Copilot section.
• Pinned to the left rail by your organisation, so it's always one click away.
• Suggested inside a chat or channel where it's relevant — some agents surface automatically in the context where they're useful.
• Linked directly by a colleague or in a SharePoint page describing a process.

If a specific agent isn't showing up, it usually hasn't been made available to you yet rather than being hidden — ask whoever mentioned it, or check with IT.

Using one

Open the agent like you'd open any chat — it has its own conversation window, separate from general Copilot. Ask it questions the way you would a person who knows that topic well: "what's our expenses policy for international travel?" for an HR-policy agent, or "summarise the open tickets assigned to my team" for one built against a support system.

Agents are typically scoped deliberately — they answer within their subject area and may say so explicitly when a question falls outside it, rather than trying to answer everything the way general Copilot would.

Built-in vs. custom agents

Some agents ship from Microsoft, built into specific apps or workflows. Others are built by your own organisation using Copilot Studio, often by an internal team that's connected the agent to a SharePoint site, a database, or a set of documents specific to your company. From the end-user side the experience is the same — a chat window you type into — regardless of who built it or what it's connected to.

What it needs

Using an agent generally needs the same Microsoft 365 Copilot licensing that covers Copilot elsewhere, though this can vary by agent type, so check current licensing rather than assuming. Which agents you can see and use is controlled centrally through the admin centre, the same way other Teams apps are managed.

Where it falls short

An agent only knows what it's been connected to — a policy agent built against last year's handbook won't know about a policy that changed last week unless someone's updated its source content. Agents can also decline or answer vaguely on questions genuinely outside their scope, which is expected behaviour rather than a fault. For the admin side of publishing and governing agents, see Copilot agents and agent store governance and Copilot Studio: agent design.