Build your first agent in Copilot Studio

Terminal objective: After this bite, you can create an agent in Microsoft Copilot Studio, add a knowledge source, test it in the chat panel, and publish it.
Audience: Non-technical adults who use Microsoft 365. No coding background. Assumes they've met Microsoft 365 Copilot (Bite 01).
Duration: 8–10 minutes · 3 lessons
Format note: Build-ready handoff: each block names a Rise block type so an author can place it directly.

Lesson 1 — What Copilot Studio is, and who it's for

Enabling objective

State what Copilot Studio is and one reason you'd build an agent with it.

Statement

Copilot Studio lets you build your own AI agent by describing what you want — in plain language, no code.

Text — the one-line version

Microsoft Copilot Studio is a platform for building, customising, and deploying AI agents through a guided, no-code experience. You don't need data scientists or developers.

Image/Text — sketch-note

A simple agent outline with three inputs: Instructions (how it behaves), Knowledge (what it can reference), Tools (actions it can take, optional).

Accordion — when would you build one?

Answer common questions from your own documents.
Help people self-serve through a product or policy.
Reach people in Teams, Microsoft 365 Copilot, SharePoint, or a website.

Knowledge check — what's the main way you create an agent in Copilot Studio?

Write code in a developer environment
Describe what you want the agent to do in plain language ✓
Buy a finished agent from a store

TakeawayCopilot Studio = build your own agent, by describing it, with no code.

Lesson 2 — Build an agent: describe it, then give it knowledge

Enabling objective

Sequence the first two build steps: describe the agent, then add a knowledge source.

Process — four steps

Describe it — enter a brief description (up to 1,024 characters).
Let it draft — Copilot Studio generates a name, description, instructions, and suggestions.
Add knowledge — a public website, SharePoint, an uploaded file, Dataverse, and more.
Name each source — give it a unique name and a detailed description.

Tabs

Instructions — identity, tone, scope, behaviour.
Knowledge — websites, SharePoint, files, Dataverse, Azure AI Search, connectors.
Tools — actions like calling an API or running a flow.
Model — the AI model that powers the agent's reasoning.

Knowledge check (multiple response) — which can be a knowledge source?

A public website ✓
SharePoint ✓
An uploaded file ✓
A handwritten note on your desk

TakeawayDescribe the agent, accept or adjust what it drafts, then add a knowledge source with a clear name and description.

Lesson 3 — Test it, then publish it

Enabling objective

Use the test chat to check an agent, then publish it to a channel.

Text — test in the chat panel

Use the Test your agent panel while you build. Ask a question, change the instructions or knowledge, ask again. Testing as you go is how the agent gets good.

Process — three steps

Publish — select Publish, then confirm. It updates every connected channel.
Add channels — Teams and Microsoft 365 Copilot, SharePoint, a website.
Share — publish for yourself, test, then make it available to your team or organisation.

Accordion — two things worth knowing

A trial licence can build and test, but not publish.
Authentication is on by default (Microsoft Entra ID). "No authentication" lets anyone with the link chat to it — use with care.

Knowledge check — what must you do before adding channels like Teams?

Nothing — channels are added automatically
Publish the agent at least once ✓
Delete the test chat

Quote

"The best way to improve your agent? Test it. Make some changes. Test it again. Repeat." — Microsoft Learn, Quickstart: Create and deploy an agent.

Bite takeawayDescribe your agent, give it a knowledge source, test it in the chat panel, then publish — and publish again every time you change it.