Exercise: Build an agent using Copilot Chat
Microsoft 365 Copilot Chat lets you create custom agents to support your work—from answering questions to helping with specific workflows. In this unit, you’ll build your first agent using Copilot Studio, a tool integrated into Copilot Chat. You’ll guide the creation using natural language or by configuring details manually—no code required.
Tip
You can build agents in two ways:
- Describe tab: Use natural language to describe what the agent should do, and Copilot builds the draft.
- Configure tab: Manually define details like knowledge sources, starter prompts, and capabilities.
Both options are synced, so you can start in one and finish in the other.
Exercise
In this exercise, you’ll create and test a custom Copilot agent designed to support a specific task, team, or initiative. You’ll use either the Describe or Configure tab to define its purpose, then test and publish the agent for use.
Steps
Start in Copilot Studio
Open your browser and navigate to m365.cloud.microsoft/chat.
Sign in using your Microsoft work account.
Select Create an agent in the right-hand rail to launch Copilot Studio.
Define your Agent (Describe tab or Configure tab)
Choose the Describe tab and use this sample prompt (or write your own):
You’re a virtual assistant for our [project/team name]. Your role is to help with [key tasks]. Be concise, stay on-brand, and reference our shared resources when possible.
Note
You can start from scratch or base your agent on a template, which prepopulates settings and instructions you can later customize.
If Describe isn’t available, switch to the Configure tab and enter the same details manually: name, description, and agent instructions.
Customize your Agent
In the Configure tab, explore these options:
Add knowledge sources (e.g., SharePoint sites, documents)
Tip
Use the Word or PowerPoint files you created in earlier exercises—like your Copilot Research.docx or the presentation draft—as knowledge sources for your agent. Upload them to SharePoint or OneDrive first, then link them as part of your agent’s knowledge base.
Define starter prompts to help others get started with your agent
Tip
Starter prompts help guide users on how to interact with your agent—especially useful for new or infrequent users.
Optionally enable capabilities like the code interpreter or image generator
Test and Create
Use the Test feature (available in the right pane throughout the agent-building process) to try out your draft agent and refine any issues.
Once satisfied, select Create to publish the agent.
Share your agent with others or open it for immediate use.
Important
If you want your agent to pull from specific knowledge, be sure to add and configure those sources before publishing.
Key takeaways
- Copilot Studio lets you create custom agents using natural language or a manual configuration experience.
- You can use predefined templates or build from scratch.
- Published agents can reference internal resources and complete tasks, enhancing productivity and collaboration.