Create agents in Microsoft Copilot Studio

In the previous unit, you created an agent using a natural language description. In this unit, you create an agent manually and add instructions to the agent.

Create an agent manually

1. In Copilot Studio, select Agents in the left-hand menu.

   

2. Select the drop-down next to Create blank agent and select Advanced create.

3. Select the Language and, Solution, and enter the Schema name of your agent.

4. Select Confirm and create.

5. The agent is created named Agent with blank description and instructions.

   

Configure basic features

Once your agent is created, you can change its name, description, and instructions. You can also make it unique with a distinctive icon.

Select the Edit icon in the Details section and enter the Name and add a Description, and select Save.

Add an icon

Select the agent icon in the top bar, select Change icon, and browse and select an image from your device, and select Save.

Agent instructions

Instructions are important. Without clear instructions, your agent won’t know what to do. Instructions are the central directions and parameters an agent follows. Agents depend on instructions to:

• Decide what resources (tools, knowledge, topics, or other agents) to call to address a user query or autonomous trigger.
• Fill inputs for any tool based on the available context.
• Generate a response to the end user.

Select the Edit icon in the Instructions section, enter your instructions, and select Save.

Craft instructions

There's an art to crafting agent instructions. Unclear instructions for an agent result in poor responses. If you don't tell your agent exactly what to do, how to do it, and what to avoid, you get unpredictable or even useless results. That's why effective agent instructions need to cover five key areas: Role, Scope, Context, Tone, and Error Handling. Let's break these down one by one:

• Role - What’s the purpose of the agent?
• Scope - What should it do? What should it avoid?
• Context - What does the agent need to know?
• Tone - How should the agent respond?
• Error handling -What does it do if it’s unsure?

To learn how to write instructions, see Instruction guidance for generative orchestration.

Suggested prompts

You can configure up to six suggested prompts that customers can choose from to start a conversation with your agent in Microsoft Teams or Microsoft 365 Copilot Chat.

In Teams and in Copilot Chat, suggested prompts appear on the agent's welcome page, before you start a new chat.

Enhance an agent created in Microsoft 365 Copilot

Creating agents in Microsoft 365 Copilot is similar to creating custom agents in Copilot Studio, but the starting point is different; you start from the Microsoft 365 Copilot agent page.

Once the agent is created, you can interact with it in the test chat, or add more capabilities and knowledge to it. Its only source of information is what you provided in the instructions. The agent doesn't have access to enterprise knowledge or any other tools at this time, so any responses come from the generative AI's own knowledge.

You can add knowledge from SharePoint and Copilot connectors.

You can see the newly created agent under the Agents tab in the Microsoft 365 Copilot app. Because the declarative agent isn't a custom agent, it doesn't appear in the Agents list in Copilot Studio.

A declarative agent can be published to Microsoft 365 Copilot and Microsoft Teams.

To discover more about creating agents in Microsoft 365 Copilot, see Extend Microsoft 365 Copilot with agents.

Copy a declarative agent to Copilot Studio

If you start building an agent in Microsoft 365 Copilot and want to add more capabilities that are only available in Copilot Studio, use the Copy to Copilot Studio button in the More options (…) menu. This option copies your agent to Copilot Studio so you don't need to recreate it.

To Learn more about enhancing declarative agents in Copilot Studio, see Copy an agent to Copilot Studio.