AB-620
Free AB-620 practice questions
10 questions from our AB-620 practice tests, each with the correct answer and a written explanation. Read them with the answers shown, or switch to quiz mode and answer first.
Question 1 of 10, Plan and configure agent solutions
An organization wants makers to stop adding public website knowledge sources and non-approved connectors to Copilot Studio agents across all production environments. What should the Power Platform administrator configure?
- AA security role in Dataverse that removes the Basic User role from every maker in the environments
- BA sensitivity label in Microsoft Purview applied to every agent in the production environments
- CA Conditional Access policy in Microsoft Entra ID that blocks the Copilot Studio app for all makers in production
- DA data policy in the Power Platform admin center that blocks the relevant connectors for those environments Correct answer
Answer and explanation: Data policies (DLP) in the Power Platform admin center govern Copilot Studio connectors, including knowledge-source connectors such as public websites, and can block them per environment. Conditional Access would stop makers from using Copilot Studio entirely rather than restricting specific capabilities. Sensitivity labels classify content but do not block connector use in agents. Removing Basic User breaks general Dataverse access without targeting knowledge sources.
Question 2 of 10, Plan and configure agent solutions
An expense agent flow must pause when a claim exceeds 5,000 USD and wait for a finance manager to supply a cost-center code before it continues. The manager should answer from Outlook without opening another app. Which human-in-the-loop capability should you add to the agent flow?
- AA Post adaptive card and wait for a response action sent to the manager in a Teams group chat
- BA Request for information action that sends the manager an actionable form for the code Correct answer
- CA Start and wait for an approval action that returns only Approve or Reject outcomes
- DA Delay until action followed by a Get rows action that checks a SharePoint list for the code
Answer and explanation: Request for information is the agent flow human-in-the-loop action designed to collect specific data from a person through an actionable Outlook message and then resume the flow with those values. A Teams adaptive card can collect input but the requirement is to answer from Outlook. A standard approval returns an outcome and comments, not a structured cost-center field. Polling a SharePoint list with a delay is brittle and requires the manager to use another app.
Question 3 of 10, Plan and configure agent solutions
A company plans an employee-facing HR agent in Microsoft Teams that must call Workday and SAP SuccessFactors with each employee's own permissions, so people only see their own records. Which authentication approach should you plan for the agent's tools?
- AConfigure the agent with no authentication and filter the HR results in a Power Fx condition based on the user's display name
- BConfigure every connector tool with maker-provided credentials of a service account that has read access to all employees
- CStore an API key for each system in an environment variable and pass the user's email as a filter parameter in every call
- DUse end-user credentials for the connector tools with Microsoft Entra ID sign-in so each call runs as the signed-in user Correct answer
Answer and explanation: Using end-user credentials with Entra ID sign-in makes each tool call run in the user's security context, so the HR systems enforce their own record-level permissions. A shared service account sees every record and relies on the agent to filter, which is weaker. No authentication cannot identify the user reliably, and display names are not a security boundary. A shared API key with an email filter can be manipulated and still grants full access.
Question 4 of 10, Integrate and extend agents in Copilot Studio
Which type of agent integration in Copilot Studio is designed specifically for natural-language questions over data in Microsoft Fabric lakehouses, warehouses and semantic models?
- AA Copilot connector configured for OneLake files and folders
- BA Fabric data agent added to the Copilot Studio agent Correct answer
- CA Dataverse knowledge source using virtual tables
- DAn Azure AI Search knowledge source over Parquet
Answer and explanation: Fabric data agents answer natural-language questions over Fabric data such as lakehouses, warehouses and semantic models, and Copilot Studio can add them for delegation. Dataverse virtual tables expose external rows but are not a Fabric question-answering agent. A Copilot connector indexes documents for search and does not run analytical queries. Azure AI Search over Parquet files retrieves chunks rather than computing aggregates.
Question 5 of 10, Integrate and extend agents in Copilot Studio. Choose 2.
A tool built from a Power Platform connector sends an email on behalf of the user. You want the agent to show the email details and get the user's approval before sending. Which tool settings should you use? (Choose two.)
- AMove the tool into an agent flow that runs on a recurrence trigger every five minutes
- BSet the tool to use end-user credentials so the email is sent from the user's mailbox Correct answer
- CTurn on the option that asks the user for confirmation before the tool runs Correct answer
- DSet the tool's inputs to custom values so the orchestrator cannot change the body
- ESet the tool's completion to Don't respond so the user never sees the result
Answer and explanation: Requiring user confirmation shows the planned action and inputs for approval before execution, and end-user credentials ensure the email is sent as the user. Fixing all inputs to custom values prevents the agent from composing the email. A recurrence flow detaches the action from the conversation and removes the approval step. Choosing not to respond hides the outcome, which conflicts with the transparency goal.
Question 6 of 10, Integrate and extend agents in Copilot Studio
After connecting Application Insights, the security team insists that user messages must not be stored in telemetry, but node-level execution data is still needed for troubleshooting. Which configuration should you choose?
- ADisable all telemetry options and rely on the test pane activity map in Copilot Studio for all troubleshooting
- BEnable logging of activities and sensitive activity properties, and turn off logging of node actions
- CEnable every logging option and rely on a data retention policy in Log Analytics to delete user messages
- DEnable logging of activities and node actions, and leave logging of sensitive activity properties turned off Correct answer
Answer and explanation: The Application Insights settings let you log activities and node actions separately from sensitive properties such as message text, so keeping that option off meets the privacy requirement while preserving troubleshooting data. Enabling sensitive properties stores user messages. The test pane only shows test conversations, not production. Retention policies delete data later, but the messages would still be stored.
Question 7 of 10, Integrate and extend agents in Copilot Studio. Choose 2.
You plan a multi-agent solution in Copilot Studio. Which statements describe good design practices? (Choose two.)
- APut all specialists in one parent agent's topics so no orchestration across agents is needed
- BGive each sub-agent a focused responsibility and a clear description so the orchestrator can route correctly Correct answer
- CGive every sub-agent identical instructions and tools so any of them can answer any request
- DAvoid descriptions on sub-agents because the orchestrator reads their full instructions to route
- EUse connected agents when a specialist must be reused by several agents and maintained by another team Correct answer
Answer and explanation: Focused sub-agents with clear descriptions let the orchestrator route accurately, and connected agents suit specialists reused across agents and owned by another team. Identical instructions and tools make routing ambiguous. The orchestrator relies on descriptions to select agents, so omitting them hurts routing. Collapsing everything into topics forgoes the separation of concerns that multi-agent design provides.
Question 8 of 10, Integrate and extend agents in Copilot Studio
A data science team built an agent in Microsoft Foundry Agent Service that uses custom code tools to forecast demand. A Copilot Studio agent in Teams must delegate forecasting questions to it. What should you do?
- AAdd the Foundry agent to the Copilot Studio agent as a connected agent by using a Foundry connection Correct answer
- BPublish the Foundry agent to the Teams channel directly and retire the existing Copilot Studio agent completely
- CRebuild the forecasting logic as a prompt tool in Copilot Studio using a model from the catalog
- DExport the Foundry agent's instructions and paste them into a new child agent in Copilot Studio
Answer and explanation: Copilot Studio can integrate a Foundry agent as a connected agent through a Foundry connection, so the orchestrator delegates forecasting while the Foundry team keeps its code tools. Rebuilding as a prompt loses the custom code tools. Copying instructions into a child agent does not bring the tools or the hosting. Removing the Copilot Studio agent abandons the existing Teams solution and its orchestration.
Question 9 of 10, Test and manage agents
A team has an existing spreadsheet of 60 customer questions with approved answers. They want to evaluate a new agent against it with the least effort. What should they do?
- ARecreate each question as a trigger phrase in a new topic and check which topic the agent picks
- BUpload the spreadsheet as a knowledge source so the agent learns the approved answers before the test run
- CAsk the agent each question in the demo website and paste its replies into the spreadsheet by hand
- DImport the questions and expected responses from a CSV file to create a test set in agent evaluation Correct answer
Answer and explanation: Agent evaluation can create a test set by importing questions and expected responses from a file, which reuses the spreadsheet directly. Turning questions into trigger phrases tests routing, not answer quality, and changes the agent. Manual copy and paste is slow and error-prone. Uploading the answers as knowledge contaminates the evaluation because the agent would simply retrieve the expected answers.
Question 10 of 10, Test and manage agents
When deploying with Power Platform Pipelines, you want environment variable values and connection references for production to be set automatically instead of being entered by makers each time. What should you use?
- AA Dataverse plug-in that rewrites the environment variable definitions after every import
- BDeployment settings that supply environment variable values and connection references for the stage Correct answer
- CA Conversation Start topic that sets the production values in global variables on the first use of the agent
- DA second unmanaged solution in production that overrides the environment variables manually
Answer and explanation: Deployment settings provide environment variable values and connection references for a target stage, so deployments run without manual entry. Global variables cannot configure connections or environment variables. An unmanaged override layer adds manual work and can block future updates. Rewriting definitions with a plug-in is unsupported customization.