Brian Alderman

What Is a Microsoft 365 Agent?

A Microsoft 365 agent is an AI-powered assistant designed to perform a specific business function rather than simply provide general-purpose chat.

An agent can:

  • Understand natural-language requests
  • Search organizational knowledge
  • Reason over information
  • Summarize and analyze documents
  • Answer questions
  • Retrieve information from Microsoft 365
  • Execute actions
  • Call business systems and APIs
  • Trigger workflows
  • Create or modify business content
  • Interact with users in Teams and Microsoft 365 Copilot
  • Automate multi-step business processes

Microsoft describes an agent as an AI assistant that handles conversations and completes tasks by combining instructions, knowledge, tools, and actions.

A useful way to think about the difference is:

Copilot helps a user accomplish a task; an agent is specialized to accomplish a particular type of task or business process.

For example:

Microsoft 365 Copilot

“Summarize this document.”

HR Agent

“What is our parental-leave policy?”

IT Help Desk Agent

“My laptop can’t connect to Wi-Fi. Troubleshoot it.”

SharePoint Governance Agent

“Find SharePoint sites containing sensitive documents that haven’t been modified in two years.”

Sales Agent

“Find the latest customer information, summarize the opportunity, and create a follow-up task.”

1. The Basic Architecture of an Agent

A useful model for understanding Microsoft 365 agents is:

User → Agent → Instructions → Knowledge → Reasoning → Tools/Actions → Result

An agent normally contains several important components.

Instructions

Instructions define what the agent is supposed to do.

Example:

You are an internal SharePoint governance assistant. Help administrators identify sites that may require review. Do not modify SharePoint content without explicit authorization.

Instructions establish the agent’s behavior, boundaries, and role.

Knowledge

Knowledge provides the information the agent can use.

Potential sources include:

  • SharePoint sites
  • SharePoint pages
  • SharePoint document libraries
  • Microsoft 365 data
  • Microsoft 365 Copilot connectors
  • Websites
  • Business systems
  • Databases
  • Other enterprise data sources

Agent Builder, for example, can use SharePoint and Microsoft 365 Copilot connector data as knowledge sources.

Tools and Actions

Tools allow an agent to do something, rather than simply answer.

Examples:

  • Send an email
  • Create a task
  • Create a SharePoint item
  • Query a database
  • Call an API
  • Run a Power Automate flow
  • Create a calendar event
  • Retrieve information from another business application

Orchestration

Orchestration determines how the agent interprets a request and decides what information, tools, or actions are needed.

This is one of the major differences between a simple chatbot and an agent.

2. The Two Major Types of Microsoft 365 Agents

Microsoft currently identifies two major categories:

  1. Declarative agents
  2. Custom engine agents

Declarative Agents

Declarative agents extend Microsoft Copilot using Microsoft’s existing AI infrastructure.

You define:

  • Instructions
  • Knowledge
  • Actions
  • Capabilities
  • Behavior

Microsoft provides the underlying model, orchestration, and hosting.

They are therefore generally easier to build and administer.

Typical tools include:

  • Agent Builder
  • SharePoint
  • Microsoft 365 Agents Toolkit
  • Copilot Studio

Declarative agents are particularly appropriate when you want to specialize Microsoft Copilot rather than build an entirely independent AI application.

Example

You could create:

“Contoso SharePoint Administrator Agent”

It could be instructed to:

  • Answer SharePoint administration questions
  • Search SharePoint documentation
  • Explain governance policies
  • Find relevant internal procedures
  • Provide migration guidance

The agent continues to use Microsoft’s Copilot infrastructure.

3. Custom Engine Agents

Custom engine agents provide considerably more control.

The organization can control more of the:

  • AI model
  • Orchestration
  • Application architecture
  • Data sources
  • Business logic
  • Actions
  • Authentication
  • Hosting
  • Automation

Custom engine agents can be built using:

  • Microsoft Copilot Studio
  • Microsoft 365 Agents SDK
  • Microsoft 365 Agents Toolkit
  • Visual Studio
  • Visual Studio Code
  • Azure AI/Fabric-related services
  • .NET
  • JavaScript/TypeScript
  • Python
  • Other development technologies

Microsoft describes custom engine agents as fully customized AI assistants that can integrate with custom data, actions, decision paths, connectors, and automation.

Example

A custom sales agent could:

  1. Receive a customer request.
  2. Search CRM information.
  3. Retrieve recent emails.
  4. Query inventory.
  5. Check pricing.
  6. Generate a proposal.
  7. Create a CRM task.
  8. Send an approval request.

That is substantially more sophisticated than a simple question-answering agent.

4. Microsoft 365 Agent Creation Options

There are four major Microsoft approaches you should understand as an administrator.

Tool Typical user Complexity Primary purpose
SharePoint Agents SharePoint users Low Agents over SharePoint content
Agent Builder Business users Low Quickly customize Microsoft 365 Copilot
Copilot Studio Makers / developers Low–High Advanced low-code agents and automation
Microsoft 365 Agents Toolkit Developers High Pro-code enterprise agents/apps

Microsoft’s current administration guidance identifies SharePoint, Agent Builder, Copilot Studio, and Microsoft 365 Agents Toolkit as the principal ways organizations create agents for Microsoft 365.

5. SharePoint Agents

SharePoint provides one of the easiest ways for users to create an agent.

A SharePoint agent can specialize in information contained within SharePoint.

For example, a company could create:

“Employee Handbook Agent”

using:

  • HR SharePoint site
  • Employee handbook
  • Benefits documents
  • HR policies
  • Procedures

Users can then ask questions without manually searching through documents.

SharePoint agents respect the user’s existing permissions. An agent doesn’t simply give a user access to information they couldn’t otherwise access.

Creating a SharePoint Agent

A typical process is:

  1. Open the appropriate SharePoint site.
  2. Identify the content the agent should use.
  3. Create the agent.
  4. Define its purpose/instructions.
  5. Configure its knowledge.
  6. Test it.
  7. Share it with appropriate users.

SharePoint-created agents can be accessed through SharePoint and Teams Chat, with sharing governed by existing permissions and security settings.

Managing SharePoint Agents

SharePoint administrators need to consider:

  • Who can use the agent
  • What SharePoint content it can access
  • Who can access that content
  • Where the agent is available
  • Licensing/billing
  • Data governance

Microsoft specifically identifies who, what information, and where as three major areas of SharePoint agent access management.

6. Agent Builder in Microsoft 365 Copilot

Agent Builder is designed for quickly creating declarative agents without traditional programming.

It is available within Microsoft 365 Copilot experiences, including:

  • Microsoft365.com
  • Office.com
  • Microsoft Teams desktop/web

A user can describe the desired agent in natural language.

For example:

Create an agent that helps employees understand our company’s travel policy. Use the Travel Policy SharePoint site as the primary knowledge source.

The user can then configure:

  • Name
  • Description
  • Instructions
  • Knowledge
  • Behavior
  • Suggested prompts

The agent can be tested before being shared.

Example Agents

Travel Policy Agent

Answers employee travel questions.

New Employee Agent

Helps new employees understand:

  • Benefits
  • Policies
  • Training
  • Procedures
  • Company resources

Project Assistant

Uses project documentation stored in SharePoint to answer project questions.

Agent Builder is intended for relatively straightforward scenarios. When you need more advanced actions, integrations, or complex workflows, Microsoft recommends moving to Copilot Studio.

7. Microsoft Copilot Studio

Copilot Studio is Microsoft’s primary low-code development environment for sophisticated agents.

It allows makers to create agents and workflows and connect them to organizational data and business systems.

Copilot Studio can provide:

  • Generative AI
  • Knowledge sources
  • Topics
  • Actions
  • Connectors
  • Power Automate integration
  • Authentication
  • APIs
  • Workflows
  • Agent-to-agent interaction
  • Analytics
  • Testing
  • Publishing
  • Multiple channels

It is therefore the natural progression:

Agent Builder → Copilot Studio

when the requirements become more sophisticated.

8. Creating an Agent in Copilot Studio

A typical development process is:

Step 1 — Define the business problem

Don’t start with:

“We need an AI agent.”

Start with:

“We need to reduce the time employees spend finding HR policies.”

Step 2 — Create the agent

Open Copilot Studio and create an agent.

You can describe the agent in natural language.

Step 3 — Define instructions

Example:

You are the company’s HR policy assistant. Answer employee questions using approved HR documentation. If the answer cannot be found in the approved sources, tell the user that you don’t have sufficient information.

Step 4 — Add knowledge

Possible sources include:

  • SharePoint
  • Websites
  • Dataverse
  • Connectors
  • Business applications
  • Custom data sources

Step 5 — Add tools/actions

For example:

“Create HR ticket”

could invoke a Power Automate flow.

Step 6 — Configure authentication

For internal business agents, Microsoft Entra-based authentication is generally important.

Step 7 — Test

Use Copilot Studio’s test experience to test:

  • Normal questions
  • Ambiguous questions
  • Unauthorized requests
  • Incorrect questions
  • Missing information
  • Edge cases

Step 8 — Publish

Publishing creates the version that users can access through configured channels. Changes made after publishing require another publish operation to become live.

9. Microsoft 365 Agents Toolkit

For developers, Microsoft provides the Microsoft 365 Agents Toolkit.

It is the evolution of the former Teams Toolkit and supports development for:

  • Microsoft 365 Copilot
  • Microsoft Teams
  • Outlook
  • Office
  • Web
  • Other channels

It supports development using:

  • Visual Studio Code
  • Visual Studio
  • CLI
  • Microsoft 365 Agents SDK
  • Teams SDK
  • JavaScript/TypeScript
  • C#
  • Python in supported scenarios

This is the approach to consider when the organization needs professional software development rather than primarily low-code development.

10. Agents in Microsoft Teams

Teams is one of the most important agent deployment locations.

Agents can appear:

  • In a user’s Teams experience
  • In Teams chat
  • In team/channel scenarios
  • Through the Teams app store
  • As part of Microsoft 365 Copilot

Copilot Studio can publish an agent to the Teams and Microsoft 365 Copilot channel.

Example

An IT Help Desk agent could allow an employee to ask:

My password is locked.

The agent could:

  1. Identify the user.
  2. Determine the appropriate troubleshooting process.
  3. Provide instructions.
  4. Create a service ticket if necessary.
  5. Escalate to a human technician.

Teams Administration

Teams administrators may control:

  • Which apps are available
  • Which agents can be installed
  • App permission policies
  • App setup policies
  • Organization app catalog
  • User/group availability

Custom-developed agents can also be submitted to the Teams administration experience for organizational approval.

11. Agents and Microsoft 365 Copilot

Microsoft 365 Copilot is increasingly becoming the central experience for agents.

An organization can:

  • Build agents
  • Install agents
  • Share agents
  • Deploy agents
  • Block agents
  • Remove agents
  • Assign agents to users/groups
  • Manage agents through Microsoft 365 administration

Microsoft provides centralized agent controls through the Microsoft 365 admin center.

There are also Microsoft-provided agents, such as Researcher and Analyst, which Microsoft may install for licensed users.

12. Agents in Outlook

Agents can participate in Microsoft 365 experiences involving Outlook and email-related workflows.

Typical scenarios include:

Email Assistant

Summarize the latest customer correspondence.

Sales Agent

Review the latest customer emails and identify outstanding action items.

Meeting Agent

Identify follow-up tasks from this conversation and prepare them for assignment.

HR Agent

Find the relevant HR policy for this employee question.

For developers, Microsoft 365 Agents Toolkit supports building experiences that work across Outlook, Teams, and Microsoft 365 Copilot.

13. Agents and Word, Excel, and PowerPoint

Microsoft 365 agents can participate in Office-based workflows, particularly through Microsoft 365 Copilot and extensibility.

Examples include:

Word

A legal-document agent could:

  • Analyze a document
  • Check it against company standards
  • Identify missing sections
  • Explain clauses
  • Generate suggested changes

Excel

A financial-analysis agent could:

  • Analyze workbook data
  • Explain trends
  • Answer questions
  • Generate analysis
  • Support business decisions

PowerPoint

A presentation agent could:

  • Use corporate presentation standards
  • Find approved content
  • Generate presentation material
  • Help create speaker notes

The important distinction is that an agent does not necessarily have to be “built inside Word” or “built inside Excel.” A Microsoft 365 agent can be designed as an extensibility component and then surfaced through supported Microsoft 365 hosts.

14. Agents and SharePoint

SharePoint is particularly important because it is both:

  1. A Microsoft 365 application
  2. A major enterprise knowledge repository

That makes it an excellent source of agent knowledge.

For example:

Corporate Policy Agent

could use:

  • HR site
  • Legal site
  • Finance site
  • IT site
  • Corporate policies

The agent can retrieve information while respecting the user’s existing permissions.

This is extremely important:

An agent should not be considered a security boundary.

If the underlying Microsoft 365 permissions are incorrect, an agent can potentially expose information to a user who already has access to that information through the connected system.

Therefore, good SharePoint security remains essential.

15. Agents and Power Automate

One of the most powerful combinations is:

Agent + Power Automate

The agent determines what needs to happen, while Power Automate performs the workflow.

Example:

User:

“I need to request vacation from December 22 through December 30.”

Agent:

  1. Understands the request.
  2. Collects missing information.
  3. Calls a Power Automate flow.
  4. Creates the request.
  5. Sends the manager approval.
  6. Reports the result to the employee.

This changes the agent from a question-answering tool into a business-process assistant.

16. Agents and Microsoft Graph

Microsoft Graph provides another important integration mechanism.

Graph can expose Microsoft 365 information such as:

  • Users
  • Groups
  • Teams
  • Outlook
  • Calendar
  • OneDrive
  • SharePoint
  • Files
  • Directory information

Agents can use Microsoft Graph-based capabilities and connectors to work with enterprise information.

This is especially useful when developing sophisticated enterprise agents.

17. Agents and Connectors

Microsoft 365 Copilot connectors can make external organizational information available to Copilot and agents.

For example, an organization might have:

  • Salesforce
  • ServiceNow
  • SAP
  • SQL Server
  • Internal applications
  • Document repositories

An agent could potentially use indexed information from these systems.

This creates a broader architecture:

Microsoft 365

↓

Agent

↓

Enterprise Knowledge

↓

SharePoint + Teams + OneDrive + CRM + ERP + Line-of-Business Systems