Exercise - Create a Copilot Chat agent
In this exercise, you create your own custom agent for Microsoft 365 Copilot Chat. Custom agents are tailored AI assistants designed to help with specific workflows, tasks, or domains within your organization. They can also help with a private project that isn't work-related. By designing an agent, you must explore how to guide Copilot's behavior and responses using configuration options such as roles, instructions, data sources, and permissions. The goal of this exercise is to help you understand how to think critically about the purpose and functionality of agents that you create within Microsoft 365 Copilot Chat.
Exercise
In this exercise, you're encouraged to be creative and design an agent that's of significant interest to you. For example, you might want to solve a real-world business problem, improve productivity in a specific area at your company, or address a personal topic that interests you. Because each student creates their own personal agent, the instructions below focus on the structure and order of how to configure an agent rather than specifying actual values to enter in each field. It's up to you to decide what you want to enter in each field as you create your agent.
In Microsoft Edge, open a new tab and enter the following URL: https://M365copilot.com
In Microsoft 365, select Create agent in the navigation pane.
In the Copilot Studio window, the Describe tab is displayed by default. If you want to base your agent on an existing template, you can either select one of the templates that's displayed or select View all templates. If you selected View all templates, then scroll through the list of templates and select the one of your choice. If you selected a template, note the starter prompts that are assigned to your agent from the selected template.
The Type your message field that appears below the templates on the left-side of the page is used for configuring your agent. The Type your message field that appears on the right-side of the page under the starter prompts is used for testing your agent. Since you're configuring the agent, the remaining steps use the Type your message field that appears on the left-side of the page. In this field, write a short summary that describes what your agent is designed to do and then select the Send (forward arrow) icon.
If you didn't provide a name for your agent in the description that you just entered, then Copilot Studio typically provides a suggested name and asks you to either confirm the name or provide a new one. You should reply accordingly.
At this point, Copilot Studio typically asks you to refine the instructions for your agent. It may do this through a series of questions. Respond to each question accordingly until the agent's description is to your liking. Copilot Studio might also ask if you would like to add any publicly available websites as data sources for this agent. While you can respond here, you typically assign your knowledge sources on the Configure tab in the upcoming steps.
Once you finalize the agent's description, select the Configure tab at the top of the form.
On the Configure tab, review the Name, Description, and Instructions fields. Make any changes that you feel are needed. Note that the Instructions were created by Copilot Studio to address the Description that you provided, along with answers to any follow-up questions that you responded to on the Describe tab. Keep in mind that the instructions tend to be much more granular in nature than the description that you entered. Verify the instructions map to your intent in guiding how your agent should behave and respond to users. Make any changes as necessary.
In the Knowledge section, you should attach any relevant data sources, such as SharePoint sites, documents, or third-party integrations, that your agent should reference.
Warning
If you don't add any public websites or work data to your agent, it might still be able to respond to your requests, but only using general capabilities of the underlying large language model (LLM), such as answering common knowledge questions, basic reasoning, or generating generalized text. As such, it's recommended that you ground your agents in specific work data or public websites; otherwise, the agent's responses might lack the context or relevance needed for business-specific tasks.
In the Capabilities section, you can optionally enable the Code interpreter and Image generator tools by selecting their toggle switches. Both tools are turned Off by default.
In the Starter prompts section, Copilot Studio displays the six starter prompts that it automatically created when you entered the agent's description back on the Describe tab. You can edit or delete any of the existing prompts or add new ones.
On the right side of the page, below the starter prompts, you can optionally test the agent by entering a message in the prompt field. You can test your agent as long as you want and refine your configuration if it isn't working or responding as you intended.
Once you feel the agent is configured properly, select the Create button at the top of the page.
Once the agent is created, the Your agent was created successfully window appears. If you want to change the sharing settings, select the Change sharing settings link. Doing so enables you to define who in your organization can use this agent by configuring user access settings. By default, the agent only works for you. Once you've configured the sharing settings (or left it as is just for you), select the Go to agent button to open your agent.
On your agent's window, note how the agent appears in the navigation pane. Feel free to enter any of the starter prompts or enter any other custom prompts to test out the agent.
When you're done using the agent, select All agents in the navigation pane. In your Agent Store window that appears, note how your agent is displayed in the Your agents section. You can access this agent from the Agent Store, or you can select the pin icon that appears next to the agent in the navigation pane if you want the agent to always display under the list of agents in the pane.