Copilot AI Agents: How They Relate to Copilot Studio

The content of this article was checked on Microsoft Learn on October 2, 2026. Microsoft changes product names and licensing structures frequently. Before you adopt anything, check the latest information through the sources at the end.

There are two main ways to use AI agents with Copilot. One is to use ready-made agents or agents built in-house inside Copilot. The other is to build new ones in Copilot Studio. In both cases, what you can use depends on your licenses.

First, note the name changes. According to Microsoft Learn, Microsoft 365 Copilot was renamed Microsoft Copilot, and Microsoft 365 Copilot Chat was renamed Microsoft Copilot Chat. This article uses the new names.

What Agents in Microsoft Copilot Are

Microsoft Learn describes agents as specialized AI assistants tailored to specific areas that extend what Copilot can do. Beyond finding information and summarizing data, they can also take actions such as sending email or updating records. There are two ways to build them: declarative agents and custom engine agents.

Declarative agents use Copilot's AI infrastructure, models and orchestrator as they are. As a result, they run in line with Microsoft 365 security, compliance and responsible AI requirements. The builder defines the agent's purpose, instructions, knowledge and available actions.

The agents, skills, Copilot connectors and MCP servers you build can be bundled into units called plugins for publishing and management. Microsoft Learn explains that publishing alone does not make a plugin available; depending on the distribution channel, steps such as review, assignment and enablement are required.

What You Can Build in Copilot Studio

Copilot Studio is a low-code tool for building and managing AI agents and workflows. Because you assemble things on screen, Microsoft says you can build without deep technical knowledge. You can mainly build the following.

• Agents: carry out work through conversation, based on instructions and connected knowledge. You can start building by describing what you want in plain language
• Workflows: automations you assemble by drag and drop. Each step can call agents or tools
• Agent flows: flows built with operations similar to Power Automate. Run them on their own or use them as tools for agents

Prebuilt or custom connectors link to data and systems inside and outside the company. You can test agents before publishing and evaluate them with test sets. Publishing destinations include Microsoft Teams, Microsoft Copilot, websites and SharePoint.

Steps to Build an Agent

You use Copilot Studio by opening copilotstudio.microsoft.com in a web browser. Based on the steps in Microsoft Learn, an agent that handles a specific job, like an AI employee, is built in this order.

1. Create a new agent and describe the job you want it to do in plain language
2. Write the agent's details and instructions
3. Add knowledge for it to reference (internal documents and sites)
4. Add tools and connectors, and decide the scope of actions it can take
5. Try it in the test chat and evaluate it with test sets
6. Publish it to Teams, Microsoft Copilot and other channels

How to Check the Licenses You Need

Check licenses in two parts: what users need, and what building and running agents needs.

User Licenses

• Microsoft Copilot Chat: users who sign in with a Microsoft Entra work or school account and have an eligible Microsoft 365 subscription can use it at no additional cost. However, access to organizational data and advanced agents depends on licenses and tenant settings
• Microsoft Copilot: an add-on license for a prerequisite plan such as Microsoft 365 Business Standard. Microsoft 365 E7 includes Microsoft Copilot

The Microsoft Copilot license includes Copilot Studio agent features when used inside Microsoft Copilot, Teams and SharePoint. When a person with this license uses an agent to reference internal Microsoft Graph data, that use is not counted toward Copilot Studio billing.

Licenses for Building and Running Agents

Copilot Studio usage is counted in a common unit called Copilot Credits. Each time an agent finds information, answers or takes an action, it uses credits according to the complexity of the work. There are three ways to get them.

• Pay-as-you-go: connect an Azure subscription and pay only for the credits you actually used that month
• Prepaid packs: buy monthly capacity. Unused credits do not carry over to the next month
• One-year prepurchase plan: buy it in the Azure portal and use it across eligible Microsoft products

With a trial license, you can build agents and try them in the test chat. However, you cannot publish them. You can estimate how many credits you will use with the agent usage estimator that Microsoft publishes.

Permissions When Connecting to Internal Data

According to Microsoft Learn, Microsoft Copilot shows only organizational data that the user has at least view permission for. Because it uses the same access model as services such as SharePoint, data from people without permission is not supposed to appear in answers.

Put the other way around, if SharePoint or team sharing settings are too broad, documents you never meant to expose will be used in answers. Microsoft asks you to set up permissions so that only the right people and groups can access the right content.

• Check that no HR or accounting documents sit in sites or folders open to the whole company
• Review access granted to people outside the company as well
• When you connect outside data with Copilot connectors (Graph connectors), only data the user has permission to access appears in answers

For agents, administrators can choose which ones the organization allows. Users can use only agents that the administrator allows and that they installed or were assigned. We also explain how to think about permissions in Managing AI Agent Permissions.

Work Copilot Can Handle and Work Better Suited to Other Tools

Copilot agents suit work whose materials and communication are all inside Microsoft 365. When your main data and tools are outside Microsoft 365, it is better to compare other options too.

For example, summarizing Teams meetings or answering internal questions based on SharePoint documents can be done entirely inside Copilot. On the other hand, for work centered outside Microsoft 365, such as building sales lists or connecting to outside recruiting services, compare other options before you decide. We summarize what to look for in How to Compare AI Agents.

What to Decide with IT Before Adoption

Copilot agents affect settings across the whole tenant. Do not move forward with front-line teams alone; decide the following with your IT department.

• Who gets a Microsoft Copilot license, and who uses only Copilot Chat
• Who is allowed to build agents in Copilot Studio
• How to buy Copilot Credits, and what the monthly limit should be
• Who reviews and approves agents for internal use
• Internal rules on what information may and may not be entered

The 'AI Guidelines for Business (Version 1.2)' from Japan's Ministry of Internal Affairs and Communications (MIC) and Ministry of Economy, Trade and Industry (METI) list, among the steps businesses using AI should take, following the security precautions set by the provider and not entering confidential information inappropriately. Combine Microsoft's documentation with your internal rules and share them with users.

If you want to delegate work outside Microsoft 365, you can also adopt an AI agent that has already been built. Employee Store is a marketplace where companies can adopt AI agents (AI employees) built by developers, through a one-time purchase, a monthly plan, or a setup fee plus monthly. You can see listed AI in the AI employee directory.