Free AB-620 sample questions with answers
Try these before reading the answers. The mix follows the real weighting, so two come from the integration domain.
Question 1. An agent must retrieve records from a legacy internal application that has no API, no connector and no supported integration surface. Users interact with it only through its web interface. Which Copilot Studio capability fits?
- A. A custom connector
- B. A REST API tool
- C. Computer use
- D. Azure AI Search as a knowledge source
Show answer
Answer: C
Computer use lets an agent operate an application interface directly, which is the mechanism for systems with no API or connector. A custom connector and a REST API tool both require an API to call, and Azure AI Search would index content rather than drive the application.
Question 2. A refund agent can issue refunds up to a limit. Above that limit a manager must approve before the refund is issued. What should you build?
- A. A topic instruction telling the agent to be careful above the limit
- B. A human-in-the-loop agent flow requiring manager approval before the action runs
- C. Remove the refund action from the agent entirely
- D. Log all refunds to Application Insights and review them weekly
Show answer
Answer: B
A human-in-the-loop agent flow pauses for a person to approve before the action executes, which is exactly the requirement. Instructing the agent in its topic does not enforce anything, removing the action makes the agent unable to do its job, and logging afterwards does not prevent the refund.
Question 3. You need a Copilot Studio agent to hand a specialised task to an agent built and hosted by another team on a different platform, using an open protocol for agent-to-agent communication. What should you use?
- A. Adaptive cards
- B. An environment variable
- C. A Dataverse table
- D. The A2A protocol
Show answer
Answer: D
The Agent2Agent (A2A) protocol exists for agent-to-agent collaboration across implementations, which is what handing work to another team’s agent on another platform requires. Adaptive cards render UI, environment variables hold configuration, and a Dataverse table stores data.
Question 4. Your team must move a Copilot Studio agent from development to test to production, with the service endpoint differing in each environment. Which combination is correct?
- A. A solution containing the agent, environment variables for the endpoint, and Power Platform Pipelines
- B. Rebuild the agent by hand in each environment
- C. Export the chat transcripts and import them into production
- D. Hard-code the production endpoint and copy the agent
Show answer
Answer: A
Packaging the agent in a solution with environment variables for the endpoint, promoted through Power Platform Pipelines, is the supported application lifecycle management approach. Rebuilding by hand invites drift, exporting chat transcripts is unrelated, and hard-coding the endpoint defeats the purpose of having environments.
Question 5. Before releasing an updated agent you must show that its answers did not get worse than the previous version. What should you do first?
- A. Ask a few colleagues to try it and report back
- B. Release it and watch Application Insights for complaints
- C. Create a test set and run an evaluation, then compare results with the previous version
- D. Increase the agent’s capacity before release
Show answer
Answer: C
Creating a test set and choosing an evaluation method gives you a repeatable measurement you can compare between versions. Asking colleagues is not repeatable, monitoring after release detects problems too late, and raising capacity affects throughput rather than answer quality.
How did you do?
Notice the pattern: each question names a requirement and asks which mechanism meets it. That is how the integration domain is examined. If you were unsure which mechanism to reach for, the fix is configuring each one once rather than reading about them again — see the study plan.