Microsoft AB-620: Plan and configure agent solutions Practice Questions

Making the agent solution-aware by adding it to a Dataverse solution (C) is the foundation of ALM in the Power Platform ecosystem — it packages the agent and its dependencies for transport. Using Power Platform pipelines or Azure DevOps (E) is the recommended automated mechanism for promoting solutions across environments, enabling version control and approval gates. Exporting via a manual 'Download' zip (A) is not the supported ALM path for Copilot Studio agents and does not maintain solution integrity or dependencies. Re-creating the agent from scratch in production (B) is error-prone, time-consuming, and not a supported deployment practice. There is no Environment dropdown in the Copilot Studio authoring canvas that directly publishes to a different environment (D); environment changes require proper solution deployment.

Copilot Studio is the correct choice because it is a low-code/no-code platform designed specifically for business analysts and citizen developers to build, configure, and deploy conversational agents quickly — including to a public website — without writing custom code. Azure AI Foundry (B) is a pro-developer platform for building and orchestrating advanced AI models and pipelines, not suited for a non-developer. A Microsoft 365 Copilot declarative agent (C) extends Microsoft 365 Copilot and requires users to have M365 Copilot licenses; it does not deploy to a public website. Azure Bot Service with Direct Line (D) requires significant development effort and coding expertise.

Disabling 'Allow the AI to use its own general knowledge' (E) prevents the language model from drawing on its parametric training data (which includes public internet content) when generating answers — grounding responses only in configured knowledge sources. Removing all public website URLs from knowledge sources (B) ensures the agent does not retrieve and use live public web content. Together these two settings confine the agent strictly to internal configured sources. Setting fallback to human escalation (A) only affects what happens when no topic matches; it does not prevent model-generated answers within matched topics or generative flows. The content moderation filter (D) filters harmful content but does not prevent the model from using general or public knowledge. Turning off generative AI orchestration entirely (C) would prevent knowledge-source-based answers as well as general knowledge answers — overcorrecting and preventing the agent from using the desired internal knowledge sources via generative answers.

When generative AI orchestration is enabled, Copilot Studio uses the combination of trigger phrases AND agent instructions to determine which topic best matches a user's intent. (E) Adding more diverse, specific trigger phrases to 'Vendor Lookup' gives the orchestrator clearer signals to distinguish it from 'Supplier Onboarding,' directly addressing the ambiguity. (A) Refining system instructions to describe the agent's purpose and the distinction between topics guides the AI orchestrator's routing logic, which is one of the key levers available when generative orchestration is on. — (F) is wrong because completely disabling generative AI orchestration is a drastic change that removes beneficial capabilities and is not the recommended approach for a routing accuracy issue; precise trigger phrases and good instructions are preferred. (C) is wrong because adding a redirect node inside the wrong topic is a fragile workaround that only catches one keyword and does not solve the underlying intent-matching problem. (D) is wrong because deleting a legitimate business topic destroys real functionality the agent needs; consolidation is not justified simply to fix a routing issue. (B) is wrong because switching the underlying generative model does not directly address topic trigger phrase coverage or instruction clarity, and model selection is not the appropriate tool for resolving topic disambiguation.

A declarative agent built through Microsoft 365 Copilot extensibility is designed exactly for this scenario: it extends the Microsoft 365 Copilot experience surface (including Teams) with domain-specific knowledge and grounding, without creating a separate standalone bot. Users interact with it through the familiar Copilot interface. A standalone Copilot Studio agent deployed to Teams (A) creates a separate bot experience rather than extending M365 Copilot itself, which is the stated requirement. Azure AI Foundry (C) is a pro-developer platform for model development, not for adding a lightweight CRM knowledge scope to M365 Copilot. A scheduled Power Automate flow (D) does not create a conversational experience and does not extend Copilot.