Copilot Agents & Copilot Studio

Once you are comfortable with everyday Copilot, the next idea to meet is the agent. Where Copilot is a general assistant that can help with almost anything, an agent is a custom AI helper scoped to a specific set of data, instructions or tasks. Think of the difference between a smart all-rounder and a specialist you have briefed for one job. Agents are how organisations turn Copilot from “handy” into “this knows exactly how we do things.”

What an agent actually is

An agent is Copilot with a narrow brief and a defined knowledge source. Instead of answering from everything you can see, it answers from a chosen slice — a particular SharePoint site, a folder of policies, a product manual — and follows instructions you give it about tone, scope and what to do. You can pin agents in Teams and Copilot Chat and use them like any other chat, except they stay on topic.

Everyday examples make it concrete. An HR Policy agent answers staff questions using only the official HR handbook, so nobody guesses. A Sales Onboarding agent walks new reps through the playbook and points them to the right decks. An IT Helpdesk agent triages common issues using your internal knowledge base. Each one is Copilot, focused.

Declarative and SharePoint agents

Two kinds are worth naming. A declarative agent is built by declaring what it should do — you describe its instructions, the data it is grounded in, and any actions it can take, and Microsoft 365 Copilot’s underlying model does the reasoning. You are not writing code or training a model; you are filling in a clear specification.

A SharePoint agent is a particularly easy starting point: it is an agent grounded in a specific SharePoint site or set of files. If your team already keeps its documents in SharePoint, you can create an agent over that site in a few clicks, and now anyone with access can ask questions and get answers drawn straight from those documents — with the same permissions respected as always.

Building one in Copilot Studio

You do not need to be a developer to make an agent. Copilot Studio is Microsoft’s no-code builder for exactly this. In plain terms, you give the agent a name, write its instructions in ordinary language (“you are a friendly HR assistant; only answer from the handbook; if you’re unsure, say so”), point it at its knowledge source, optionally add an action or two, and test it in a preview pane. When it behaves, you publish it.

For simple cases — an agent over a SharePoint site or a set of files — you can often build it from inside Copilot Chat or SharePoint itself in minutes. Copilot Studio is the fuller workshop when you want more control, multiple data sources, or custom actions that reach into other systems.

Rule of thumb: if you keep answering the same question from the same documents, that is a job for an agent — not for you, again.

Spot it: General Copilot vs. Agent

Read each situation and decide for yourself, then tap a card to flip it and check your answer.

Sort the agent types

Drag each scenario into the bucket it belongs to — or tap an item, then tap a bucket. Hit Check placement when you’re done.

Sharing and governance

An agent you build is not automatically loose in the wild. You choose who to share it with, and it still respects every user’s existing permissions — a person who could not see a file on their own cannot see it through the agent either. Behind the scenes, your IT admins set governance controls: which agents are allowed, which data they may touch, and whether they can be shared beyond your team. Everything stays inside your organisation’s compliance boundary, and the same cautions from everyday Copilot apply — verify the output, and never ground an agent in confidential data through a tool your company has not approved.

How to use it

Look for repetition. The moment you notice yourself answering the same policy question, pointing people to the same files, or copy-pasting the same onboarding steps, you have found a candidate for an agent. Start with the easiest win — a SharePoint agent over a site your team already uses — and try a declarative agent in Copilot Studio when you want tighter instructions. Share it narrowly first, gather feedback, and loop in your admins on anything that touches sensitive data. Useful phrases: “Could we build an agent over the policy site so people stop asking us?” “Let’s make a declarative agent in Copilot Studio and test it before sharing.” “It only answers from the handbook, and it respects everyone’s permissions.” Saying those marks you as someone who scales good answers instead of repeating them.