Microsoft AI-103 (Developing AI Apps and Agents on Azure) Practice Test 25

Question 240

After implementing a workflow to publish a versioned agent endpoint for stable clients, the team needs a robust release gate. Which check should it choose?

  1. Verify contract compatibility and rollback on representative client calls
  2. Monitor calls to the retired version and test that the old tool endpoint is inaccessible
  3. Attempt cross-user requests and confirm access denials in both retrieval and tools
  4. Measure field-level extraction accuracy on labeled scans and low-quality images

Explanation

Correct Answer

A:

The accepted change is to publish a versioned agent endpoint for stable clients; this evaluation checks whether it can let several applications consume an agent while its implementation changes under controlled release, since versioned published endpoints decouple client contracts from internals and support staged promotion when the platform supports it.

Incorrect Answers

B:

This test concerns the ability to remove old versions that retain deprecated tools without disrupting active clients, for which a controlled retirement checks clients, revokes unused credentials, and archives relevant audit records before deletion; it does not establish whether the change to publish a versioned agent endpoint for stable clients works for the current release.

C:

This test concerns the ability to enforce what both the service and the delegated user are authorized to do, for which a workload identity authenticates the agent service, while user-aware authorization limits operations on behalf of a caller; it does not establish whether the change to publish a versioned agent endpoint for stable clients works for the current release.

D:

This test concerns the ability to enable an agent to read key fields from photographed or scanned paperwork, for which document extraction tools convert visual text and layout into structured data suitable for subsequent agent reasoning; it does not establish whether the change to publish a versioned agent endpoint for stable clients works for the current release.

Question 241

Production telemetry reveals this failure: A workflow fails after several tool calls and developers cannot identify the first invalid intermediate state. Which change most directly corrects it?

  1. Score chosen tool calls against permitted expected action traces
  2. Categorize traces by root-cause stage and track regressions by version
  3. Enable correlated tracing across decision and tool spans with appropriate redaction
  4. Use stable operation identifiers and enforce idempotent processing server-side

Explanation

Correct Answer

C:

The fault signal is that a workflow fails after several tool calls and developers cannot identify the first invalid intermediate state; the targeted fix works because per-step tracing captures model requests, tool parameters, statuses and transitions; a final response log hides the path taken, restoring the ability to find where a lengthy agent plan diverges from intended workflow.

Incorrect Answers

A:

If the failing function were to detect unnecessary or dangerous tool calls even when its final prose looks plausible, this change could be relevant because tool-selection evaluation examines intermediate actions and authorization fit, not only the final natural-language answer; the actual symptom is that a workflow fails after several tool calls and developers cannot identify the first invalid intermediate state, which calls for a different repair.

B:

If the failing function were to prioritize corrections across retrieval misses, malformed tools, policy blocks, and model planning errors, this change could be relevant because failure classification links observed symptoms to the component that can be improved; a single aggregate error rate obscures causes; the actual symptom is that a workflow fails after several tool calls and developers cannot identify the first invalid intermediate state, which calls for a different repair.

D:

If the failing function were to prevent double charging when an agent retries after a network timeout, this change could be relevant because an idempotency key ties a logical transaction to one effect even if the same request is retried; the actual symptom is that a workflow fails after several tool calls and developers cannot identify the first invalid intermediate state, which calls for a different repair.

Question 242

An operator can reproduce this error: One agent has every privileged tool even though most requests require only one department's data. Which targeted repair is justified?

  1. Use stable operation identifiers and enforce idempotent processing server-side
  2. Classify request intent and dispatch to an authorized specialist with scoped capabilities
  3. Enforce an extraction-to-analysis handoff with schema validation at the boundary
  4. Evaluate external side effects and final artifacts against labeled target states
  5. Add a supervisor that gathers evidence, resolves disagreements, and approves final synthesis

Explanation

Correct Answer

B:

The fault signal is that one agent has every privileged tool even though most requests require only one department's data; the targeted fix works because a router delegates to an appropriate specialist with a narrow tool set; a single all-powerful agent is harder to govern, restoring the ability to route finance, operations, and security questions to different bounded expert agents.

Incorrect Answers

A:

If the failing function were to prevent double charging when an agent retries after a network timeout, this change could be relevant because an idempotency key ties a logical transaction to one effect even if the same request is retried; the actual symptom is that one agent has every privileged tool even though most requests require only one department's data, which calls for a different repair.

C:

If the failing function were to require one specialist's validated output before another starts its dependent work, this change could be relevant because a sequential handoff makes intermediate contracts explicit; parallel execution would violate a data dependency; the actual symptom is that one agent has every privileged tool even though most requests require only one department's data, which calls for a different repair.

D:

If the failing function were to determine whether the agent actually achieved a multi-step business goal, this change could be relevant because task-success tests compare final system state with expected outcomes; prose fluency does not prove work was completed; the actual symptom is that one agent has every privileged tool even though most requests require only one department's data, which calls for a different repair.

E:

If the failing function were to combine independent research subtasks into a coherent final response with oversight, this change could be relevant because a supervisor can assign independent tasks, collect outputs, and apply cross-checks before presenting an answer; the actual symptom is that one agent has every privileged tool even though most requests require only one department's data, which calls for a different repair.

Question 243

An operator can reproduce this error: A support agent recalls a one-time code from an earlier completed session. Which targeted repair is justified?

  1. Carry relevant prior tool results and user selections in session-scoped conversation state
  2. Summarize older turns, keep current facts verbatim, and validate the retained state
  3. Use an approved shared tool integration with centralized access policy and secret handling
  4. Exclude transient secrets from memory and enforce state expiration policies

Explanation

Correct Answer

D:

The fault signal is that a support agent recalls a one-time code from an earlier completed session; the targeted fix works because short-lived state and purpose-based retention protect secrets and sensitive data; summarization is not a substitute for removal, restoring the ability to avoid retaining authentication tokens or temporary PII inside agent memory longer than needed.

Incorrect Answers

A:

If the failing function were to resolve 'those two records' by retaining the preceding selection context, this change could be relevant because conversation tracking gives referential continuity across turns, but the context should be bounded to the correct user session; the actual symptom is that a support agent recalls a one-time code from an earlier completed session, which calls for a different repair.

B:

If the failing function were to retain key decisions after a long conversation becomes too large to replay verbatim, this change could be relevant because a validated rolling summary compresses history while preserving commitments and identifiers; careless compression can discard safety constraints; the actual symptom is that a support agent recalls a one-time code from an earlier completed session, which calls for a different repair.

C:

If the failing function were to avoid giving every agent separate uncontrolled credentials for the same external service, this change could be relevant because centralized governed tool access can standardize authentication and logging while each agent remains least privileged; the actual symptom is that a support agent recalls a one-time code from an earlier completed session, which calls for a different repair.

Question 244

Production telemetry reveals this failure: An old published agent still exposes a risky tool even though a replacement has been launched. Which change most directly corrects it?

  1. Set tool-call ceilings and route exhausted searches to a controlled no-result response
  2. Propagate the caller's security context and authorize each sensitive query appropriately
  3. Migrate consumers, disable the stale agent endpoint, and revoke its credentials
  4. Publish a stable agent interface and manage revision selection in deployment configuration

Explanation

Correct Answer

C:

The fault signal is that an old published agent still exposes a risky tool even though a replacement has been launched; the targeted fix works because a controlled retirement checks clients, revokes unused credentials, and archives relevant audit records before deletion, restoring the ability to remove old versions that retain deprecated tools without disrupting active clients.

Incorrect Answers

A:

If the failing function were to stop self-triggering tool loops from consuming unlimited tokens or external calls, this change could be relevant because step, time, and spending limits create enforceable failure boundaries regardless of the prompt plan; the actual symptom is that an old published agent still exposes a risky tool even though a replacement has been launched, which calls for a different repair.

B:

If the failing function were to enforce what both the service and the delegated user are authorized to do, this change could be relevant because a workload identity authenticates the agent service, while user-aware authorization limits operations on behalf of a caller; the actual symptom is that an old published agent still exposes a risky tool even though a replacement has been launched, which calls for a different repair.

D:

If the failing function were to let several applications consume an agent while its implementation changes under controlled release, this change could be relevant because versioned published endpoints decouple client contracts from internals and support staged promotion when the platform supports it; the actual symptom is that an old published agent still exposes a risky tool even though a replacement has been launched, which calls for a different repair.

Question 245

After implementing a workflow to use a dispatcher agent to select a specialist, the team needs a robust release gate. Which check should it choose?

  1. Inspect inter-agent messages for unauthorized fields and verify output correctness
  2. Measure unresolved conflicts and evidence attribution across worker outputs
  3. Test correct routing and reject cross-department privileged actions
  4. Inspect tool-call schema adherence and argument-validation failure rate

Explanation

Correct Answer

C:

The accepted change is to use a dispatcher agent to select a specialist; this evaluation checks whether it can route finance, operations, and security questions to different bounded expert agents, since a router delegates to an appropriate specialist with a narrow tool set; a single all-powerful agent is harder to govern.

Incorrect Answers

A:

This test concerns the ability to stop one agent from receiving private data that another agent needed only for its subtask, for which scoped messages and redacted handoff contracts minimize sensitive context propagation across multiple agents; it does not establish whether the change to use a dispatcher agent to select a specialist works for the current release.

B:

This test concerns the ability to combine independent research subtasks into a coherent final response with oversight, for which a supervisor can assign independent tasks, collect outputs, and apply cross-checks before presenting an answer; it does not establish whether the change to use a dispatcher agent to select a specialist works for the current release.

D:

This test concerns the ability to make the parameters of an account-lookup function explicit to the model and executor, for which a typed tool contract constrains argument names and shapes and simplifies validation; authorization still occurs on execution; it does not establish whether the change to use a dispatcher agent to select a specialist works for the current release.

Question 246

A system now includes a component to maintain task state apart from transient model output. Which evaluation should accompany the rollout?

  1. Inject mid-workflow failures and audit the final reconciled business state
  2. Inspect tool-call schema adherence and argument-validation failure rate
  3. Red-team embedded document instructions and inspect unauthorized tool-call attempts
  4. Replay interrupted workflows and verify each external side effect occurs once

Explanation

Correct Answer

D:

The accepted change is to maintain task state apart from transient model output; this evaluation checks whether it can resume a long-running workflow safely after an inference retry, since durable workflow state records completed steps and pending approvals; relying only on the model's last text risks duplicate side effects.

Incorrect Answers

A:

This test concerns the ability to recover from a partially completed workflow when a later operation fails, for which a compensating action undoes or offsets previously committed steps where a transaction cannot be atomic across services; it does not establish whether the change to maintain task state apart from transient model output works for the current release.

B:

This test concerns the ability to make the parameters of an account-lookup function explicit to the model and executor, for which a typed tool contract constrains argument names and shapes and simplifies validation; authorization still occurs on execution; it does not establish whether the change to maintain task state apart from transient model output works for the current release.

C:

This test concerns the ability to keep untrusted customer-uploaded text from redefining the agent's permissions, for which trusted control instructions must have higher precedence than external content; retrieved pages are data, not agent policy; it does not establish whether the change to maintain task state apart from transient model output works for the current release.

Question 247

Engineers have implemented a change to limit information passed between specialist agents. What validation would expose a remaining defect?

  1. Test correct routing and reject cross-department privileged actions
  2. Compare computed totals against independently verified reference calculations
  3. Measure unresolved conflicts and evidence attribution across worker outputs
  4. Inspect inter-agent messages for unauthorized fields and verify output correctness

Explanation

Correct Answer

D:

The accepted change is to limit information passed between specialist agents; this evaluation checks whether it can stop one agent from receiving private data that another agent needed only for its subtask, since scoped messages and redacted handoff contracts minimize sensitive context propagation across multiple agents.

Incorrect Answers

A:

This test concerns the ability to route finance, operations, and security questions to different bounded expert agents, for which a router delegates to an appropriate specialist with a narrow tool set; a single all-powerful agent is harder to govern; it does not establish whether the change to limit information passed between specialist agents works for the current release.

B:

This test concerns the ability to let an agent aggregate a supplied numeric CSV where exact computation is important, for which a sandboxed execution tool performs calculations deterministically; executable code needs strict isolation and resource limits; it does not establish whether the change to limit information passed between specialist agents works for the current release.

C:

This test concerns the ability to combine independent research subtasks into a coherent final response with oversight, for which a supervisor can assign independent tasks, collect outputs, and apply cross-checks before presenting an answer; it does not establish whether the change to limit information passed between specialist agents works for the current release.

Question 248

The engineering change request is to use a supervisor to coordinate parallel worker agents. What is the most direct acceptance criterion?

  1. Inspect memory contents and verify secrets are absent after session termination
  2. Inspect inter-agent messages for unauthorized fields and verify output correctness
  3. Measure unresolved conflicts and evidence attribution across worker outputs
  4. Test correct routing and reject cross-department privileged actions

Explanation

Correct Answer

C:

The accepted change is to use a supervisor to coordinate parallel worker agents; this evaluation checks whether it can combine independent research subtasks into a coherent final response with oversight, since a supervisor can assign independent tasks, collect outputs, and apply cross-checks before presenting an answer.

Incorrect Answers

A:

This test concerns the ability to avoid retaining authentication tokens or temporary PII inside agent memory longer than needed, for which short-lived state and purpose-based retention protect secrets and sensitive data; summarization is not a substitute for removal; it does not establish whether the change to use a supervisor to coordinate parallel worker agents works for the current release.

B:

This test concerns the ability to stop one agent from receiving private data that another agent needed only for its subtask, for which scoped messages and redacted handoff contracts minimize sensitive context propagation across multiple agents; it does not establish whether the change to use a supervisor to coordinate parallel worker agents works for the current release.

D:

This test concerns the ability to route finance, operations, and security questions to different bounded expert agents, for which a router delegates to an appropriate specialist with a narrow tool set; a single all-powerful agent is harder to govern; it does not establish whether the change to use a supervisor to coordinate parallel worker agents works for the current release.

Question 249

A project needs to complete all of these distinct tasks: (1) prevent an autonomous agent from executing account closure without explicit review; and (2) stop self-triggering tool loops from consuming unlimited tokens or external calls. Select one suitable capability for each task. Choose TWO.

  1. Give agents bounded execution budgets
  2. Implement compensation for multi-step side effects
  3. Apply idempotency to write-capable tools
  4. Design a typed tool schema for actions
  5. Require approval before irreversible business actions

Explanation

Correct Answers

A:

For the stated requirement to stop self-triggering tool loops from consuming unlimited tokens or external calls, step, time, and spending limits create enforceable failure boundaries regardless of the prompt plan; this is the mechanism that the system must implement. This choice satisfies the specific deliverable to stop self-triggering tool loops from consuming unlimited tokens or external calls, while the other deliverables require separate capabilities.

E:

For the stated requirement to prevent an autonomous agent from executing account closure without explicit review, a human approval checkpoint separates proposed actions from committed side effects and provides an auditable decision; this is the mechanism that the system must implement. This choice satisfies the specific deliverable to prevent an autonomous agent from executing account closure without explicit review, while the other deliverables require separate capabilities.

Incorrect Answers

B:

For a system whose objective is to recover from a partially completed workflow when a later operation fails, a compensating action undoes or offsets previously committed steps where a transaction cannot be atomic across services; this does not directly meet the required outcome to prevent an autonomous agent from executing account closure without explicit review.

C:

For a system whose objective is to prevent double charging when an agent retries after a network timeout, an idempotency key ties a logical transaction to one effect even if the same request is retried; this does not directly meet the required outcome to prevent an autonomous agent from executing account closure without explicit review.

D:

For a system whose objective is to make the parameters of an account-lookup function explicit to the model and executor, a typed tool contract constrains argument names and shapes and simplifies validation; authorization still occurs on execution; this does not directly meet the required outcome to prevent an autonomous agent from executing account closure without explicit review.

Question 250

The implementation does not meet its business requirement: The model invents API argument names and fails account-lookup requests repeatedly. Which corrective action is appropriate?

  1. Delimit external evidence and apply instruction-hierarchy and injection controls
  2. Specify the support role and allowed actions, and constrain available tools to that role
  3. Place a mandatory approval gate on the account-closure tool
  4. Provide a precise function description and schema with required parameters

Explanation

Correct Answer

D:

The fault signal is that the model invents API argument names and fails account-lookup requests repeatedly; the targeted fix works because a typed tool contract constrains argument names and shapes and simplifies validation; authorization still occurs on execution, restoring the ability to make the parameters of an account-lookup function explicit to the model and executor.

Incorrect Answers

A:

If the failing function were to keep untrusted customer-uploaded text from redefining the agent's permissions, this change could be relevant because trusted control instructions must have higher precedence than external content; retrieved pages are data, not agent policy; the actual symptom is that the model invents API argument names and fails account-lookup requests repeatedly, which calls for a different repair.

B:

If the failing function were to keep a support agent focused on diagnosing product issues rather than taking unrelated administrative action, this change could be relevant because a clear role and allowed-goal scope guide planning; enforceable permissions remain necessary at tool boundaries; the actual symptom is that the model invents API argument names and fails account-lookup requests repeatedly, which calls for a different repair.

C:

If the failing function were to prevent an autonomous agent from executing account closure without explicit review, this change could be relevant because a human approval checkpoint separates proposed actions from committed side effects and provides an auditable decision; the actual symptom is that the model invents API argument names and fails account-lookup requests repeatedly, which calls for a different repair.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests