Overview
Domain 5 is new in the July 24, 2026 revision and carries 20–25% of PL-900. Its first functional group, Describe how to build agents in Copilot Studio, covers agent use cases, conversation paths built with topics, knowledge sources, tools including MCP servers and agent flows, and publishing through channels.
The second group, Describe how to manage agents in Copilot Studio, covers use cases and capabilities for Microsoft Agent 365, monitoring agent usage and adoption, and agent evaluations. Because the whole skill area is new, older PL-900 study material that only mentions chatbots or Power Virtual Agents does not cover it.
How it's tested
Expect scenarios describing what an agent must know, do or reach, followed by a choice of topic, knowledge source, tool or channel. Management items describe an admin or maker goal, such as tracking adoption or checking answer accuracy after an update, and ask which capability provides it. Watch for older terms used as distractors.
-
Knowledge or tool
Answering from documents points to knowledge sources, while performing a task such as creating a record points to a tool.
-
Where to publish
The audience and location of users decide between Teams and Microsoft 365 Copilot, a website or another channel.
-
Govern and measure
Admin oversight across many agents points to Agent 365, while answer quality and usage point to evaluations and analytics.
An HR team wants an agent in Teams that answers policy questions from SharePoint and can submit a leave request into the HR system. Policy answers come from adding the SharePoint site as a knowledge source. Submitting leave is an action, so the agent needs a tool such as an agent flow or connector. Publishing to the Teams and Microsoft 365 Copilot channel puts it where staff work.
Functional groups 2
Official functional groups from the Microsoft outline, in outline order.
5.1 Describe how to build agents in Copilot StudioAgent use cases, topics, knowledge sources, tools including MCP servers and agent flows, and publishing through channels.
- What it means
- This group asks you to describe use cases for agents, how topics build conversation paths, how knowledge sources ground answers, how tools including MCP servers and agent flows let agents act, and how agents are published through channels.
- What to understand
- Agents suit employee self-service, customer support, and task automation that responds to requests or events. Topics are authored conversation paths with nodes for messages, questions, conditions and actions; system topics handle greetings, escalation and fallback. With generative orchestration, the agent picks topics, knowledge and tools from their descriptions. Knowledge sources include public websites, SharePoint, uploaded files, Dataverse and enterprise data through connectors. Tools let agents act: connectors, agent flows built in Copilot Studio, prompts, and MCP servers that expose external tools and data through the Model Context Protocol. Channels include Teams and Microsoft 365 Copilot, websites and other platforms, with authentication settings affecting which apply.
- How it shows up
- Expect scenarios asking whether a need is met by a topic, a knowledge source, a tool or a channel setting.
- Common confusion
- Treating knowledge sources as a way to take actions; knowledge grounds answers, while tools perform tasks.
5.2 Describe how to manage agents in Copilot StudioMicrosoft Agent 365 use cases and capabilities, monitoring usage and adoption, and agent evaluations.
- What it means
- This group covers managing agents after they are built: describing use cases and capabilities for Microsoft Agent 365, monitoring Copilot Studio agent usage and adoption, and describing agent evaluations.
- What to understand
- Microsoft Agent 365 is a control plane for agents across an organization, giving IT a registry of agents, access control tied to agent identities, visibility into agent activity, and security and compliance integration, including for agents built outside Copilot Studio. Within Copilot Studio, the analytics page reports sessions, engagement, resolution and escalation rates, satisfaction, and how often knowledge sources and tools are used, which shows adoption and gaps. Agent evaluations run test sets of questions with expected answers or criteria against an agent and grade the responses, so makers can measure quality and catch regressions before publishing changes.
- How it shows up
- Expect a maker or admin goal and a choice between Agent 365, analytics or evaluations.
- Common confusion
- Confusing analytics, which report real usage, with evaluations, which test quality against prepared test sets.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Agent
An AI assistant built in Copilot Studio that converses and can take actions.
- Remember
- It combines instructions, topics, knowledge and tools.
- Tested as
- The core object of this skill area.
- Trap
- Thinking agents only answer questions and cannot act.
-
Topic
An authored conversation path made of nodes.
- Remember
- Nodes send messages, ask questions, branch on conditions and call tools.
- Tested as
- Chosen when a conversation must follow a defined sequence.
- Trap
- Using a topic to answer every document question instead of knowledge.
-
System topics
Built-in topics such as Greeting, Escalate and Fallback.
- Remember
- They handle common conversation events and can be customized.
- Tested as
- Questions on handing a user to a human or handling unknown input point here.
- Trap
- Deleting them instead of editing behavior.
-
Generative orchestration
The agent choosing topics, knowledge and tools based on their descriptions.
- Remember
- Clear names and descriptions improve selection.
- Tested as
- Explains how an agent decides what to use.
- Trap
- Assuming trigger phrases are the only way topics start.
-
Knowledge source
Content the agent uses to ground generative answers.
- Remember
- Examples: public websites, SharePoint, files, Dataverse and connector-based enterprise data.
- Tested as
- The answer for answering from existing documentation.
- Trap
- Expecting knowledge to update records or send emails.
-
Tool
A capability that lets an agent take an action or fetch data.
- Remember
- Types include connectors, agent flows, prompts and MCP servers.
- Tested as
- Chosen when the agent must do something in another system.
- Trap
- Calling tools knowledge sources because both reach other systems.
-
Agent flow
A flow built in Copilot Studio that an agent can call as a tool.
- Remember
- It uses triggers and actions like Power Automate cloud flows.
- Tested as
- Matches the bullet naming agent flows as tools.
- Trap
- Assuming agent flows are desktop RPA.
-
MCP server
A server using the Model Context Protocol to expose tools and resources to agents.
- Remember
- Adding one gives the agent its tools without building each action separately.
- Tested as
- Matches the bullet naming MCP servers as tools.
- Trap
- Thinking MCP is a publishing channel.
-
Channel
A place where a published agent is available to users.
- Remember
- Examples include Teams and Microsoft 365 Copilot and websites.
- Tested as
- Audience and location questions point to channel choice.
- Trap
- Forgetting to publish again after changes.
-
Microsoft Agent 365
An organization-wide control plane for managing agents.
- Remember
- It offers an agent registry, access control, visibility and security integration.
- Tested as
- The answer for IT governing many agents across the tenant.
- Trap
- Treating it as the tool for authoring agent topics.
-
Analytics
Copilot Studio reporting on how a published agent is used.
- Remember
- Shows sessions, engagement, resolution, escalation and satisfaction.
- Tested as
- Chosen for adoption and usage questions.
- Trap
- Using analytics to test changes before release.
-
Agent evaluations
Automated tests of an agent against test sets.
- Remember
- Test cases hold questions with expected answers or criteria and are graded.
- Tested as
- Chosen for measuring answer quality and preventing regressions.
- Trap
- Confusing evaluations with satisfaction surveys from real users.
-
Instructions
Natural language guidance that shapes an agent's tone and behavior.
- Remember
- They tell the agent how to respond and when to use tools.
- Tested as
- Appears in scenarios about consistent agent behavior.
- Trap
- Expecting instructions alone to grant access to data.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Agent building blocks
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Topic | Authored conversation path. | Predictable, step-by-step dialogue. | Writing topics for every FAQ instead of adding knowledge. |
| Knowledge source | Content for grounded answers. | Answers from documents and sites. | Expecting it to perform transactions. |
| Tool | Action or data call from the agent. | Changes or retrieves data in other systems. | Adding a tool when a document answer is all that is needed. |
| Channel | Where users reach the agent. | Determines audience and experience. | Thinking building blocks are visible before publishing. |
Agent management capabilities
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Microsoft Agent 365 | Tenant-wide agent control plane. | Governs agents at organization scale. | Choosing it for one maker's quality testing. |
| Copilot Studio analytics | Usage and outcome reporting. | Measures real conversations after publishing. | Expecting it to test before release. |
| Agent evaluations | Test sets graded against expected results. | Measures quality in a controlled way. | Treating it as live usage reporting. |
Tool types
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Connector | Prebuilt action for a service. | Quick access to common systems. | Assuming every system has one. |
| Agent flow | Multi-step automation called by the agent. | Chains several actions with logic. | Confusing it with a topic. |
| MCP server | Model Context Protocol endpoint offering tools. | Brings a set of tools from an external server. | Thinking it hosts the agent. |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Topic for every answer
- What candidates pick
- Write a custom topic for each policy question
- Why it looks right
- Topics feel precise and controllable.
- Why it's wrong
- Grounding the agent in the policy documents as a knowledge source answers many questions without authoring each one.
Clue to watch for Large or changing document sets.
-
Knowledge to act
- What candidates pick
- Add the ticketing system as a knowledge source to log tickets
- Why it looks right
- The system holds relevant data.
- Why it's wrong
- Creating a ticket is an action, which needs a tool such as a connector or agent flow.
Clue to watch for Verbs such as create, submit, update or book.
-
MCP as a channel
- What candidates pick
- Publish the agent through an MCP server
- Why it looks right
- MCP sounds like a connection point for users.
- Why it's wrong
- MCP servers give agents tools and data; channels such as Teams or websites reach users.
Clue to watch for The question asks where users talk to the agent.
-
Analytics for testing
- What candidates pick
- Use the analytics page to check answers before launch
- Why it looks right
- Analytics shows answer outcomes.
- Why it's wrong
- Analytics needs real usage; evaluations test prepared questions before or after changes.
Clue to watch for Before publishing or after an update.
-
Agent 365 for authoring
- What candidates pick
- Open Agent 365 to edit a topic
- Why it looks right
- It has agent in the name and manages agents.
- Why it's wrong
- Agent 365 governs and secures agents across the organization; building happens in Copilot Studio.
Clue to watch for Editing conversation content or knowledge.
-
Copilot Studio equals Copilot features in apps
- What candidates pick
- Use Copilot in Power Apps to build the help desk agent
- Why it looks right
- Both carry the Copilot name.
- Why it's wrong
- Copilot inside Power Apps assists makers building apps; agents are built in Copilot Studio.
Clue to watch for End users chatting with an assistant.
Answer framework
Use these steps to map an agent scenario to the right Copilot Studio building block or management capability.
- Confirm the agent use case: answering questions, guiding a process, or taking actions on requests or events in other systems.
- For answers drawn from existing content, add knowledge sources such as SharePoint, websites, files or Dataverse rather than writing many topics.
- For defined conversations that must collect inputs or follow steps, design a topic, and keep system topics for greeting, escalation and fallback.
- For actions, add a tool: a connector for a single operation, an agent flow for multi-step logic, or an MCP server that supplies external tools.
- Publish to the channel where users work, then use analytics for adoption, evaluations for quality, and Agent 365 for organization-wide governance.
Practice question
Want more like this? The PL-900 practice test has questions for every domain.
Checklist
0 of 13
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Because this skill area is new, rely on current Microsoft Learn content and the Copilot Studio documentation rather than older courses. Create a trial or developer agent and work through the pieces in order: write instructions, add a public website and a SharePoint site as knowledge, author one custom topic that asks two questions, and add a tool such as a connector action or a simple agent flow. Read how MCP servers are added so you can explain the idea.
Publish the agent to a test channel, chat with it, and then open the analytics page to see sessions and outcomes appear. Build a small evaluation test set of ten questions and run it after changing a knowledge source, noting how the scores change.
Finally, read the Microsoft Agent 365 overview and summarize its capabilities in a few lines: registry, access control, visibility and security. Plan about a week of short sessions, since this domain carries as much weight as Power Apps.
FAQ
How much of PL-900 does Describe features and capabilities of agents in Microsoft Copilot Studio cover?
Describe features and capabilities of agents in Microsoft Copilot Studio is weighted at 20–25% of PL-900. It was added as a new skill area on July 24, 2026.
Is this the same as Power Virtual Agents?
Copilot Studio replaced Power Virtual Agents, but the exam now covers knowledge sources, tools, MCP servers, generative orchestration, Agent 365 and evaluations, which older material does not.
Do I need to build an MCP server?
No. You need to describe that MCP servers expose tools and data to agents through the Model Context Protocol and that Copilot Studio can add them as tools.
What is Microsoft Agent 365?
It is Microsoft's control plane for agents across an organization, giving IT a registry, access control, visibility and security integration for agents from Microsoft and other builders.
What is the difference between analytics and evaluations?
Analytics reports on real conversations after an agent is published. Evaluations run prepared test sets against the agent to grade answer quality, often before publishing changes.
Which channels should I know?
Know Teams and Microsoft 365 Copilot and websites well, and be aware that other channels exist. Authentication settings affect which channels are available.
Are agent flows the same as Power Automate flows?
They share the trigger and action design, but agent flows are built and managed in Copilot Studio to serve agents as tools.
Other CRM & Business Applications certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
- HubSpot Inbound Marketing Certification Study guide
- Salesforce Certified Platform Developer I Study guide
- Google Ads Search Certification Study guide
- ServiceNow Certified System Administrator Study guide
- Salesforce Certified Platform Administrator Study guide
- ServiceNow Certified Application Developer Study guide
Test your Domain 5 knowledge
Check that you can choose topics, knowledge, tools and channels, and separate Agent 365, analytics and evaluations, before your exam.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the PL-900 AI tutor Get any concept explained your way