Create a M365 Copilot Chat agent - Part 1

Friday, January 2, 2026 5:07 PM

Create a Microsoft 365 Copilot Chat agent - Part 1

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 two experiences for creating agents:

• Copilot Studio "lite" experience. This experience is designed for business users with no
programming background. It provides a simple interface that makes creating agents easy. You can
either build an agent by selecting options in the UI and filling in fields step-by-step, or simply
typing what you want in plain language, and Copilot Studio interprets your instructions to build
the agent for you. This flexibility allows employees to quickly build agents and start using them
across the organization, boosting productivity without the need for technical skills or new tools.

• Copilot Studio "full" experience. This experience is intended for software developers who require
advanced customization and governance. It supports building enterprise-grade agents with
capabilities such as multi-step workflows, automated tasks, custom connectors, lifecycle
management, and granular security controls. This experience is ideal for agents serving entire
departments, organizations, or external audiences, and for scenarios requiring integration with
external systems or complex automation.

In summary, the lite Copilot Studio experience empowers nontechnical users to innovate quickly and
safely, while the full Copilot Studio experience gives technical professionals the tools to build, manage,
and govern sophisticated AI solutions at scale.

The focus of this training is on the lite experience, where everyday business users with no programming
experience can use Copilot Studio to create Copilot agents directly from their conversations with
workers. Doing so enables users to fulfill specific use cases and scenarios, based on their 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.

In the lite experience, you can create, share, and customize your Copilot agents with just a few clicks in
Copilot Studio. For example, you can connect to other knowledge sources to pinpoint the information
Copilot should use to generate responses and take action. The lite experience in Copilot Studio enables
you to author an agent in two ways:

• Through the Describe tab. In this tab, you use plain language to describe what you want your
agent to do. Copilot Studio interprets your instructions and creates a draft version of your
requested agent, automatically filling in the relevant fields for you. Here are some examples of
natural language instructions to create various types of agents:

 Create an agent that answers common questions about our company's product warranty
and return policy. It should provide clear steps for returns and link to the warranty page.

 Build an agent that helps employees find information about vacation days, sick leave, and
parental leave. It should give short answers and link to the HR portal.

 Make an agent that helps users reset their passwords and troubleshoot sign-in issues. It
should guide them through steps and suggest contacting IT if the problem continues.

 Create an agent that recommends training courses for new hires. It should suggest beginner
courses and provide links to the learning platform.

 Build an agent that answers questions about upcoming company events, such as dates,

AB-900 Page 1
  Build an agent that answers questions about upcoming company events, such as dates,
locations, and registration details.

• Through the Configure tab. In this tab, you manually build your agent step-by-step by choosing
options and entering details. You can start from scratch or refine the draft agent you created in
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.

AB-900 Page 2
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

• 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,

AB-900 Page 3
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 might 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 might 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

AB-900 Page 4
 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.

END.

AB-900 Page 5