Create an agent in Copilot Chat - Part 1

Completed

To create an agent in Microsoft 365 Copilot Chat, you must select the Create agent option in the navigation pane on the Copilot Chat page. Doing so opens Copilot Studio, which you use to create agents.

Copilot Studio

Copilot Studio provides a simple interface that enables everyday business users with no programming experience to quickly and easily build declarative agents, either by using natural language or manually. Copilot Studio makes it quick and easy to create Copilot agents, which workers can instantly use throughout their organization. In doing so, employees can boost their personal and team productivity without having to learn or access a new app.

Without writing any code, you can use Copilot Studio to create Copilot agents directly from your conversations with workers. Doing so enables you to fulfill specific use cases and scenarios, based on your organization's objectives, guidelines, and branding. Creators can build within the familiar Microsoft 365 environment they know, without needing to learn Power Platform concepts or new apps.

With just a few clicks in Copilot Studio, you can create, share, and customize your Copilot agents. For example, you can connect to other knowledge sources to pinpoint the information Copilot should use to generate responses and take action.

You can use Copilot Studio to author an agent in two ways:

• Through the Describe tab, using natural language. Copilot Studio creates a draft version of your agent, where it interprets your description and instructions and automatically updates the corresponding fields in the agent.
• Through the Configure tab, where you can build an agent manually from scratch, or where you can finish defining the details of the agent if you created the draft version on the Describe tab.

Both tabs work seamlessly together to provide a rich authoring experience.

Warning

The availability of the Describe tab is based on geographic availability and language support. If the Describe tab isn't supported in your region or preferred language, you can manually build your agent through the Configure tab.

The remainder of this training unit focuses on using the Describe tab. The next training unit explores the Configure tab.

Use the Describe tab (if available) to begin the agent creation process

If the Describe tab is available in your region or preferred language, you can use it to create the basic shell of an agent using plain language.

Optionally base your agent on a template

When you begin defining your agent in the Describe tab, Copilot Studio enables you to select from a list of predefined agent templates. If you select a template, its settings, description, and instructions are used as the basis for your agent. While choosing a template is strictly optional, keep in mind that they do serve several important purposes:

• Ease of creation. Templates provide a starting point for building agents, making the process more straightforward and user-friendly. They offer preconfigured settings and structures that users can easily customize to suit their specific needs, significantly reducing the time and effort required to build an agent from scratch.
• Consistency. Using templates ensures that agents are built with a consistent structure and functionality. This consistency helps maintain a standard quality across different agents, making them more reliable and easier to manage.
• Best practices. Templates are often designed based on best practices and common use cases. They incorporate proven methods and configurations that can help users create effective and efficient agents.
• Customization. While templates provide a solid foundation, they're also highly customizable. Users can modify the templates to add specific features, change the agent's behavior, or tailor the agent to meet unique requirements.
• Efficiency. When you start with a template, you can focus on fine-tuning and enhancing the agent rather than building it from the ground up. This efficiency allows for quicker deployment and iteration of agents.

Overall, templates in Copilot Studio are designed to streamline the agent creation process, ensure quality and consistency, and provide flexibility for customization. Templates provide a starting point with preconfigured settings, description, and instruction. Even if you select a template, you can still customize this information to fit your specific needs through the Configure tab.

Define the agent's name, description, and instructions

Within the Describe tab, you provide the agent's name, description, and instructions conversationally with Copilot Studio. In turn, Copilot Studio creates a draft version of your agent, during which it prompts you for updates to the information that you provided. For example, when updating the agent's name:

• If you include the agent's name in your description, Copilot Studio assigns that name to the draft version of that agent that it creates.
• If you don't include an agent name in your description, Copilot Studio recommends a name for you based on your description of the agent. It then allows you to confirm the recommended name or change it to something else.

Based on the description that you provide, Copilot Studio also creates several starter prompts for the agent, which it displays on the right half of the Describe tab. You can later use the Configure tab to edit or remove any of these system-generated starter prompts, or you can manually add new prompts of your own.

After you enter your description, Copilot Studio asks you to enter your instructions for the agent. The instructions guide the agent on how to behave and respond to users. If you earlier selected a template, the system inserts predefined instructions related to the selected template. You can customize these instructions if you wish, or replace them with your own, customized instructions.

When you enter your instructions using natural language, Copilot Studio takes the following steps to create and refine the agent's behavior:

1. Interpretation and parsing. Copilot Studio interprets and parses your natural language instructions to understand the intent and context. It identifies key elements, such as the agent's purpose, tasks, and desired behavior.
2. Progressive updates. After you enter your instructions, Copilot Studio might ask you a series of questions to clarify any of the tasks that were included in the instructions. Typically, the more detail you provide in your instructions, the fewer questions it tends to ask. This conversational experience provides a rich, yet simple natural language way to create an agent. Agent Builder progressively refines the agent in each turn of the conversation, ensuring that changes are saved automatically. Here's a list of some of the more common things Copilot Studio might ask you:
   • Clarification of the agent's purpose. Copilot Studio might ask you to elaborate on the specific tasks or goals you want the agent to achieve. For example, it might ask, "Can you provide more details on the primary function of this agent?"
   • Target audience. It may inquire about the intended users of the agent. Questions like "Who plans to use this agent?" or "What is the skill level of the target audience?" help tailor the agent's responses.
   • Contextual information. The builder might ask for more context or background information that the agent should be aware of. For instance, "Are there any specific scenarios or use cases the agent should be prepared for?"
   • Response style and tone. It may ask about the preferred tone and style of the agent's responses. Questions like "Should the agent's responses be formal or casual?" or "Do you have any specific guidelines for the agent's communication style?" are common.
   • Handling specific queries. The builder might seek clarification on how to handle certain types of queries. For example, "How should the agent respond to questions it can't answer?" or "Are there any specific topics the agent should avoid?"
   • Integration with other systems. If the agent needs to interact with other systems or databases, the builder might ask for details on these integrations. Questions like "Does the agent need to access any external databases or APIs?" are typical.
   • Fallback mechanisms. The builder might ask about fallback mechanisms in case the agent encounters an issue. For instance, "What should the agent do if it encounters an error or can't find the requested information?"

These questions help ensure the agent is well-defined and capable of meeting the specific needs and expectations of its users.

Test your agent on the Describe tab

Once you finalize your conversation with Copilot Studio regarding the agent's instructions, you can optionally test the draft version of your agent through the Describe tab. Even though you have yet to define specific knowledge sources in the Configure tab, the agent can still provide responses based on its built-in capabilities and general knowledge. In doing so, the agent uses its preexisting understanding and any default configurations to generate answers.

If you test this initial draft version of your agent on the Describe tab, keep in mind that agents initially use their built-in general knowledge to respond to prompts. You can enhance the agent's responses by adding specific knowledge sources such as SharePoint sites, documents, and other data sources in the Configure tab.

Tip

Initial testing of the draft version of your agent on the Describe tab is recommended for testing your natural language instruction. However, for more accurate and relevant responses tailored to your specific needs, you should also define and test the knowledge sources in the Configure tab. See the later unit in this training module for detailed information on testing an agent.

The next training unit examines how you can use the Configure tab in Copilot Studio to refine your agent or create an agent from scratch if the Describe tab isn't available.