Question 270 A Copilot Studio prompt must run on an approved Microsoft Foundry model deployment. Which mechanism fits? Change the agent's title to the model name Put the deployment key into the public conversation Configure the Prompt tool with the connected Foundry model deployment and suitable inputs Use an unrelated calendar action to invoke the model Explanation Correct Answer C:Custom prompts can select supported models from the Foundry model catalog. Incorrect Answers A:Naming does not configure model execution. B:Credentials are not safe in chat messages. D:Calendar tools cannot replace the prompt-model connection.
Question 271 An enterprise team wants per-agent diagnostic events visible in Azure Monitor. Which integration should it configure? Publish raw bearer tokens as debug messages Connect the agent's Application Insights telemetry with appropriate resource and permissions Embed logs only in a system prompt Assume model transcripts are the same as live operational traces Explanation Correct Answer B:Application Insights can collect agent diagnostics for monitoring. Incorrect Answers A:This exposes secrets without helping monitoring. C:Prompt text is not telemetry collection. D:Conversation text alone lacks tool and timing telemetry.
Question 272 A tenant wants centralized tracing across many agents rather than separate maker-owned logs. Which telemetry architecture fits where supported? Ask every agent to manually post its logs to a public chat Use environment-level Application Insights telemetry managed by platform administrators Delete per-agent monitoring configuration and collect nothing Collect per-agent maker-owned event logs only and assume they correlate across the tenant Explanation Correct Answer B:Environment-wide traces can correlate agent invocations and tool activity. Incorrect Answers A:Public messages are not a secure monitoring system. C:No logs cannot support centralized visibility. D:Independent logs do not provide the centrally linked cross-agent traces requested.
Question 273 A trace shows an agent response slow after calling Azure AI Search. What should developers compare? A randomly generated delay explanation Compare only the model response duration and ignore search dependency timing Request and dependency timing for retrieval, model and tool steps Only the user's typing speed Explanation Correct Answer C:Distributed traces can isolate where latency accumulates. Incorrect Answers A:Unverified attribution does not diagnose performance. B:A fast generation step cannot rule out slow retrieval or tool operations. D:Client typing does not explain backend retrieval time.
Question 274 A Foundry model used by custom prompts was retired or its deployment renamed. What should be checked? Increase the welcome-message length Suppress model invocation errors Update the model connection to a supported deployment and regression-test prompt outputs Assume the prompt will automatically find any similarly named model Explanation Correct Answer C:Prompt tools depend on a valid model deployment identity. Incorrect Answers A:A longer greeting does not restore a model endpoint. B:Suppressing Foundry model invocation failures prevents operators from distinguishing an invalid model deployment from a temporary backend or authentication problem. D:Deployment names and availability matter.
Question 275 An Azure AI Search query returns relevant but confidential passages. What must be verified? Send all results to every user because the index is trusted Disable audit logging to reduce search latency The retrieval path restricts results according to source and user authorization Filter only displayed text while the model sees all source secrets Explanation Correct Answer C:Search relevance does not override access permissions. Incorrect Answers A:A trusted index may contain restricted documents. B:Removing logs does not fix confidentiality. D:Unnecessary disclosure can occur before presentation.
Question 276 A Copilot Studio agent sends telemetry with full user messages. What privacy consideration is critical? Control conversation-detail logging, access, retention and sensitive-data exposure Enable raw personal-content logging for everyone by default Assume encrypted logs have no retention obligations Publish Application Insights access keys to users Explanation Correct Answer A:Verbose telemetry can contain confidential prompts or tool data. Incorrect Answers B:Unrestricted data collection increases privacy risk. C:Encryption does not eliminate data governance requirements. D:Monitoring credentials should remain protected.
Question 277 A team notices its agent's tool calls fail more often after a deployment. Which telemetry view helps? Disable all detailed traces to simplify dashboards Correlate tool execution failures and dependent agent operations in Application Insights Assume every failure is caused by the model Look only at monthly subscription invoices Explanation Correct Answer B:Tracing can identify failing dependencies after a version change. Incorrect Answers A:Without traces correlation becomes harder. C:API and permission errors may be the true cause. D:Billing data does not identify failed tool operations.
Question 278 A custom prompt should extract three contractual fields from a document using a Foundry model. What output design helps? Specify a structured three-field schema and validate each field before downstream use Ask for unconstrained natural-language text and extract three fields with loose substring patterns Treat every generated field as verified legal data Use the document title as a substitute for missing fields Explanation Correct Answer A:Model outputs need shape and evidence checks before business actions. Incorrect Answers B:Free text can omit fields or confuse contract values and labels. C:Extraction can be inaccurate and needs validation. D:Titles do not necessarily contain the requested contract data.
Question 279 A Copilot Studio agent invokes an Azure AI Search index through a generative answers node. What is the essential distinction from a REST write tool? Any search query automatically changes the source records A write tool can always return document citations without a retrieval design Both must be anonymous on private enterprise data Search retrieves grounding passages; a write tool performs an authenticated operation Explanation Correct Answer D:Retrieval and transaction APIs serve different purposes. Incorrect Answers A:A query normally does not commit business writes. B:Tool output does not imply knowledge grounding. C:Protected systems require proper authentication.
Question 280 An Application Insights dashboard has success metrics but no link between user requests and tool latency. What should be added? Correlation identifiers and appropriate dependency tracing at the supported telemetry scope An alphabetical list of agent names Use only daily aggregated dashboard counts without a per-request trace ID Remove timing fields from telemetry Explanation Correct Answer A:Linked traces associate a turn with its downstream operations. Incorrect Answers B:Names alone are not per-request correlation. C:Daily totals cannot correlate an individual turn to its tool latency. D:Missing durations make latency diagnosis impossible.