
Exam AB-620: Designing and Building Integrated AI Solutions in Copilot Studio leads to the Microsoft Certified: AI Agent Builder Associate certification. It is an intermediate, 120-minute exam on building agents in Microsoft Copilot Studio: planning identity, channels and governance; agent flows with human-in-the-loop steps; topics, generative answers and adaptive cards; enterprise knowledge through connectors and Azure AI Search; tools such as MCP, REST APIs and Computer Use; multi-agent solutions with Foundry agents, Fabric data agents and the A2A protocol; and testing and ALM. It is also one of the associate prerequisites for the AB-100 Expert certification. Power Platform and Microsoft 365 developers in Hyderabad are being asked to build exactly these agents.
This guide follows the official AB-620 study guide and certification page. It covers the exam facts, the three skill areas and their weightings, the Copilot Studio building blocks, an ALM checklist, a 5-week plan and 15 original scenario questions.
Quick facts (Microsoft Learn, November 2026)
| Item | Detail |
|---|---|
| Certification | Microsoft Certified: AI Agent Builder Associate |
| Level | Intermediate |
| Duration | 120 minutes, proctored |
| Passing score | 700 |
| Price | Based on the country where you take the exam (Pearson VUE) |
| Renewal | Every year, with a free online assessment on Microsoft Learn |
Read the official certification page and AB-620 study guide.
Skills measured and weights
| Skill area | Weight | What it covers |
|---|---|---|
| Plan and configure agent solutions | 30–35% | Enterprise integration, identity, channels and deployment; responsible AI, security and governance; reusable components; internal vs external audiences; agent flows with human-in-the-loop, actions, connectors, parameters, error handling and monitoring; topics with tools, custom prompts, custom knowledge, API and Send HTTP requests, the generative answers node, adaptive cards and variables |
| Integrate and extend agents in Copilot Studio | 40–45% | Copilot connectors, Power Platform connectors and Azure AI Search; Computer Use, MCP tools, custom connectors and REST APIs; multi-agent design with Foundry agents, existing agents, Fabric data agents and A2A; Azure AI Search with Foundry, prompts using the Foundry model catalog, monitoring with Application Insights |
| Test and manage agents | 20–25% | Test sets, evaluation methods and results; solutions, environment variables and Power Platform Pipelines |
Integration is the largest area. Spend most of your lab time connecting agents to data, tools and other agents.
Copilot Studio building blocks
| Building block | Use it for |
|---|---|
| Topic | A guided conversation path with triggers, questions, conditions and messages |
| Generative answers node | Answers grounded in knowledge sources, such as SharePoint, websites, Dataverse or Azure AI Search |
| Agent flow | A multi-step automation the agent runs, including approvals (human in the loop) and error handling |
| Tool | An action the agent can call: a connector, custom connector, REST API, MCP server, prompt or Computer Use |
| Computer Use | Automating tasks in apps and websites that have no API, by operating the user interface |
| Connected agents | Delegating to a Foundry agent, a Fabric data agent, another Copilot Studio agent, or an external agent over A2A |
| Adaptive card | Rich, interactive messages such as forms and approval cards |
| Application Insights | Telemetry for conversations, errors and performance |
ALM checklist for agents
- Build in a development environment, inside a solution (not the default solution).
- Add the agent, its flows, connection references and custom connectors to the solution.
- Use environment variables for URLs, site addresses and keys that change between dev, test and prod.
- Deploy with Power Platform Pipelines to test, then production, as a managed solution.
- Run a test set after every change and compare the evaluation results before release.
5-week study plan
| Week | Focus | Hands-on (Copilot Studio trial or developer environment) |
|---|---|---|
| 1 | Planning: identity, channels, audiences, responsible AI, governance; topics, variables, adaptive cards | Build an HR FAQ agent with a topic and an adaptive card form |
| 2 | Agent flows: triggers, actions, connectors, parameters, human in the loop, error handling | Leave-request flow with manager approval |
| 3 | Knowledge: SharePoint, Copilot and Power Platform connectors, Azure AI Search; generative answers; custom prompts and the Foundry model catalog | Ground the agent in an Azure AI Search index |
| 4 | Tools: MCP, REST APIs, custom connectors, Computer Use; multi-agent with Foundry, Fabric data agents and A2A; Application Insights | Add an MCP tool and a connected agent; trace a conversation in Application Insights |
| 5 | Test sets and evaluation; solutions, environment variables and pipelines; review | Move the agent from dev to test with a pipeline; take the practice assessment and book |
15 scenario questions with answers
These are original questions written for this guide, not exam questions.
1. An agent must answer questions from 50,000 policy documents in a search index. Which knowledge source?
Azure AI Search.
2. A purchase over ₹1 lakh needs a manager's approval before the agent continues. What do you build?
A human-in-the-loop agent flow with an approval step.
3. A legacy web portal has no API, but the agent must enter data into it. Which capability?
Computer Use.
4. A partner team exposes its tools through an MCP server. How do you use them?
Add them to the agent as MCP tools.
5. Your agent should hand data questions to an agent that queries Microsoft Fabric. What do you integrate?
A Fabric data agent.
6. An agent built on another vendor's platform must collaborate with yours. Which protocol?
A2A (Agent2Agent).
7. You need to call an internal REST endpoint from a topic without building a connector. Which node?
Send HTTP request.
8. Users should complete a structured form inside the chat. What do you use?
An adaptive card.
9. Where do you see conversation telemetry and errors in detail?
Application Insights, connected to the agent.
10. The SharePoint site URL differs between test and production. How do you handle it?
Use an environment variable in the solution.
11. What packages an agent and its flows for deployment?
A solution.
12. How do you automate deployment from dev to test to prod?
Power Platform Pipelines.
13. How do you check answer quality after changing a prompt?
Run a test set, choose an evaluation method and compare the results.
14. A flow step can fail when a connector times out. What should the flow include?
Error handling, for example a branch that runs on failure and returns a clear message.
15. An agent is for customers on a public website. What changes in planning?
The external audience: authentication options, channel, content safety, data exposure and governance.
Before and after AB-620
- Start with PL-900 and AB-900; learn the protocol with MCPA.
- Go code-first with AI-103, then on to AB-100 (AB-620 counts as its prerequisite).
- Learn it hands-on in our AI, GenAI and Agentic AI course; see live Copilot and AI developer jobs.
Cloud Soft Solutions in Ameerpet, Hyderabad teaches Azure, GenAI, agentic AI and multi-cloud with real-time projects. See the APEX program and all courses.
📞 Call +91 96660 19191 · WhatsApp · Flat No. 513, 5th Floor, Nilagiri Block, Aditya Enclave, Ameerpet, Hyderabad
Frequently Asked Questions
What is exam AB-620?
Designing and Building Integrated AI Solutions in Copilot Studio, the exam for the Microsoft Certified: AI Agent Builder Associate certification.
How long is AB-620?
120 minutes.
What are the AB-620 skill areas?
Plan and configure agent solutions 30–35%, integrate and extend agents in Copilot Studio 40–45%, test and manage agents 20–25%.
Does AB-620 count toward AB-100?
Yes. The AI Agent Builder Associate is one of the prerequisite options for the Agentic AI Business Solutions Architect Expert.
Does AB-620 cover MCP and A2A?
Yes. It includes MCP tools and multi-agent solutions using the A2A protocol.
What is the passing score?
700.



