Microsoft AI-102 (Designing and Implementing a Microsoft Azure AI Solution) Practice Test 27

Question 260

An agent has a file-search tool connected, but it still retrieves an archived warranty policy after the team approves a replacement PDF. What data-attachment maintenance is required?

  1. Validate generated tool arguments against the declared parameter schema
  2. Define durable role, policy, and tool-use boundaries in the agent's instructions
  3. Pause execution and request explicit approval before invoking the transfer tool
  4. Refresh the file-search knowledge store with the approved policy and verify retrieval of the current version

Explanation

Correct Answer

D:

Attaching a tool is not sufficient when its backing content is stale; updating the approved knowledge source and confirming retrieval distinguishes current policy evidence from archived material.

Incorrect Answers

A:

That change addresses another failure mode: function-calling structure reduces ambiguity but does not guarantee every proposed value is valid for the backend; it does not address this case, for which the relevant intervention is to refresh the file-search knowledge store with the approved policy and verify retrieval of the current version.

B:

Although useful elsewhere in the Foundry customer-service agent, stable agent instructions guide behavior across turns; ad hoc user messages should not set corporate policy; the present case is different and calls for the team to refresh the file-search knowledge store with the approved policy and verify retrieval of the current version.

C:

Human approval gates high-impact side effects; an agent's plan is not sufficient authorization in its own workflow, but that is a different correction from the present requirement to refresh the file-search knowledge store with the approved policy and verify retrieval of the current version. The reported issue would remain.

Question 261

A specialist can solve part of a request but needs another agent to continue with the accumulated context. What pattern fits?

  1. Use a controlled handoff that transfers relevant task state to the next agent
  2. Use a routing or triage step based on the user's intent and allowed capabilities
  3. Use sequential orchestration with explicit dependencies between each stage
  4. Run independent analysis branches in parallel and join their results

Explanation

Correct Answer

A:

Handoff orchestration moves responsibility and context, unlike a simple parallel fan-out. In the multi-step operations assistant, confirm the fix by retesting the affected behavior after using a controlled handoff that transfers relevant task state to the next agent.

Incorrect Answers

B:

That change addresses another failure mode: triage maps a request to the responsible worker rather than running every specialist indiscriminately; it does not address this case, for which the relevant intervention is to use a controlled handoff that transfers relevant task state to the next agent.

C:

Downstream work depends on earlier validated outputs, so parallel execution risks using incomplete data in its own workflow, but that is a different correction from the present requirement to use a controlled handoff that transfers relevant task state to the next agent. The reported issue would remain.

D:

Although useful elsewhere in the multi-step operations assistant, independent tasks can execute concurrently before a synchronization step produces the combined report; the present case is different and calls for the team to use a controlled handoff that transfers relevant task state to the next agent.

Question 262

The router repeatedly sends unknown requests to the payroll agent. What fallback should be defined?

  1. Checkpoint workflow progress and make side-effecting steps idempotent
  2. Pass the necessary validated context and authorization scope with the handoff
  3. Wait for required branches or apply an explicit timeout and partial-result policy
  4. Provide an explicit unknown-intent route to clarification or human triage

Explanation

Correct Answer

D:

Uncertain routing should not silently select a high-impact specialist without sufficient evidence. In the multi-step operations assistant, confirm the fix by retesting the affected behavior after providing an explicit unknown-intent route to clarification or human triage.

Incorrect Answers

A:

Although useful elsewhere in the multi-step operations assistant, recoverable orchestration must track completed actions to avoid duplicate writes on retries; the present case is different and calls for the team to provide an explicit unknown-intent route to clarification or human triage.

B:

That change addresses another failure mode: a handoff should transfer task-relevant facts without widening the receiving agent's permissions; it does not address this case, for which the relevant intervention is to provide an explicit unknown-intent route to clarification or human triage.

C:

Fan-out/fan-in behavior needs a defined join condition; arbitrary early merge can omit required evidence in its own workflow, but that is a different correction from the present requirement to provide an explicit unknown-intent route to clarification or human triage. The reported issue would remain.

Question 263

A benchmark for a multi-agent system changes tool fixtures between runs. What would enable a fair comparison?

  1. Gate the downstream stage on successful completion of its prerequisite
  2. Isolate branch state and merge outputs deterministically after completion
  3. Enforce handoff limits and an explicit escalation or terminal condition
  4. Use versioned test fixtures and controlled external-tool responses

Explanation

Correct Answer

D:

Stable test dependencies distinguish orchestration regressions from variations in outside systems. In the multi-step operations assistant, confirm the fix by retesting the affected behavior after using versioned test fixtures and controlled external-tool responses.

Incorrect Answers

A:

Although useful elsewhere in the multi-step operations assistant, explicit completion dependencies keep sequential workflows from consuming unvalidated intermediate results; the present case is different and calls for the team to use versioned test fixtures and controlled external-tool responses.

B:

Independent branch outputs avoid races; a controlled merge preserves all required findings in its own workflow, but that is a different correction from the present requirement to use versioned test fixtures and controlled external-tool responses. The reported issue would remain.

C:

That change addresses another failure mode: handoff cycles need bounded control flow to avoid infinite delegation and unnecessary model/tool calls; it does not address this case, for which the relevant intervention is to use versioned test fixtures and controlled external-tool responses.

Question 264

Agents bounce the same request between each other indefinitely. What guard should be added?

  1. Gate the downstream stage on successful completion of its prerequisite
  2. Isolate branch state and merge outputs deterministically after completion
  3. Regression-test the router against representative existing and new intents
  4. Enforce handoff limits and an explicit escalation or terminal condition

Explanation

Correct Answer

D:

Handoff cycles need bounded control flow to avoid infinite delegation and unnecessary model/tool calls. In the multi-step operations assistant, confirm the fix by retesting the affected behavior after enforcing handoff limits and an explicit escalation or terminal condition.

Incorrect Answers

A:

Although useful elsewhere in the multi-step operations assistant, explicit completion dependencies keep sequential workflows from consuming unvalidated intermediate results; the present case is different and calls for the team to enforce handoff limits and an explicit escalation or terminal condition.

B:

Independent branch outputs avoid races; a controlled merge preserves all required findings in its own workflow, but that is a different correction from the present requirement to enforce handoff limits and an explicit escalation or terminal condition. The reported issue would remain.

C:

That change addresses another failure mode: routing classifiers and rules can introduce cross-category mistakes as new paths are introduced; it does not address this case, for which the relevant intervention is to enforce handoff limits and an explicit escalation or terminal condition.

Question 265

One assistant handles payroll requests, another handles IT incidents, and requests arrive in a common inbox. What component should select the specialist?

  1. Use sequential orchestration with explicit dependencies between each stage
  2. Use a controlled handoff that transfers relevant task state to the next agent
  3. Use a routing or triage step based on the user's intent and allowed capabilities
  4. Capture step-level traces, tool inputs, outputs, and terminal outcomes

Explanation

Correct Answer

C:

Triage maps a request to the responsible worker rather than running every specialist indiscriminately. In the multi-step operations assistant, confirm the fix by retesting the affected behavior after using a routing or triage step based on the user's intent and allowed capabilities.

Incorrect Answers

A:

Although useful elsewhere in the multi-step operations assistant, downstream work depends on earlier validated outputs, so parallel execution risks using incomplete data; the present case is different and calls for the team to use a routing or triage step based on the user's intent and allowed capabilities.

B:

Handoff orchestration moves responsibility and context, unlike a simple parallel fan-out in its own workflow, but that is a different correction from the present requirement to use a routing or triage step based on the user's intent and allowed capabilities. The reported issue would remain.

D:

That change addresses another failure mode: workflow correctness depends on actions taken during the run, not just the fluent final message; it does not address this case, for which the relevant intervention is to use a routing or triage step based on the user's intent and allowed capabilities.

Question 266

After a handoff, the new agent asks the user to repeat account details already verified by the first. What needs preservation?

  1. Pass the necessary validated context and authorization scope with the handoff
  2. Score tool authorization and side effects in addition to final-answer quality
  3. Provide an explicit unknown-intent route to clarification or human triage
  4. Checkpoint workflow progress and make side-effecting steps idempotent

Explanation

Correct Answer

A:

A handoff should transfer task-relevant facts without widening the receiving agent's permissions. In the multi-step operations assistant, confirm the fix by retesting the affected behavior after passing the necessary validated context and authorization scope with the handoff.

Incorrect Answers

B:

A good final answer cannot erase an unsafe intermediate action in its own workflow, but that is a different correction from the present requirement to pass the necessary validated context and authorization scope with the handoff. The reported issue would remain.

C:

Although useful elsewhere in the multi-step operations assistant, uncertain routing should not silently select a high-impact specialist without sufficient evidence; the present case is different and calls for the team to pass the necessary validated context and authorization scope with the handoff.

D:

That change addresses another failure mode: recoverable orchestration must track completed actions to avoid duplicate writes on retries; it does not address this case, for which the relevant intervention is to pass the necessary validated context and authorization scope with the handoff.

Question 267

A release for a multi-step operations assistant is blocked until the application can direct a request to the appropriate specialist based on the detected task. The tester will verify that a billing inquiry reaches the billing specialist and not the hardware specialist. What should be implemented?

  1. Use Microsoft Agent Framework parallel workflow branches
  2. Use Microsoft Agent Framework routing or triage
  3. Evaluate multi-agent traces and outcomes
  4. Use Microsoft Agent Framework sequential workflow steps

Explanation

Correct Answer

B:

Routing uses task interpretation to choose a domain agent instead of sending every request through every specialist. For this multi-step operations assistant, the intended evidence is that a billing inquiry reaches the billing specialist and not the hardware specialist. A fixed sequential chain is unnecessarily rigid when only one branch should run.

Incorrect Answers

A:

Parallel orchestration permits independent tasks to execute together and joins outputs for a final synthesis; it may be used to have independent specialists analyze separate evidence sets concurrently before results are merged in a multi-step operations assistant, but the current requirement is instead to direct a request to the appropriate specialist based on the detected task. That separate capability does not prove that a billing inquiry reaches the billing specialist and not the hardware specialist; It is unsuitable when later steps require data not produced until an earlier step completes, so selecting that alternative would not meet the stated acceptance check.

C:

Workflow evaluation combines step-level traces with outcome-based measures to test orchestration and reliability; it may be used to identify incorrect delegation loops, unsafe actions, and poor end-to-end task success in a multi-step operations assistant, but the current requirement is instead to direct a request to the appropriate specialist based on the detected task. That separate capability does not prove that a billing inquiry reaches the billing specialist and not the hardware specialist; Merely increasing the number of agents does not measure whether they accomplish the assigned task, so selecting that alternative would not meet the stated acceptance check.

D:

Sequential orchestration enforces dependencies such as research preceding drafting and review preceding publication; it may be used to pass a validated intermediate result from one specialized agent to the next in a required order in a multi-step operations assistant, but the current requirement is instead to direct a request to the appropriate specialist based on the detected task. That separate capability does not prove that a billing inquiry reaches the billing specialist and not the hardware specialist; Running dependent stages in parallel can consume incomplete inputs, so selecting that alternative would not meet the stated acceptance check.

Question 268

An agent fails on one of ten tool steps, but evaluation records only the final response. What should the test framework preserve?

  1. Capture step-level traces, tool inputs, outputs, and terminal outcomes
  2. Use a controlled handoff that transfers relevant task state to the next agent
  3. Run independent analysis branches in parallel and join their results
  4. Use sequential orchestration with explicit dependencies between each stage

Explanation

Correct Answer

A:

Workflow correctness depends on actions taken during the run, not just the fluent final message. In the multi-step operations assistant, confirm the fix by retesting the affected behavior after capturing step-level traces, tool inputs, outputs, and terminal outcomes.

Incorrect Answers

B:

Although useful elsewhere in the multi-step operations assistant, handoff orchestration moves responsibility and context, unlike a simple parallel fan-out; the present case is different and calls for the team to capture step-level traces, tool inputs, outputs, and terminal outcomes.

C:

That change addresses another failure mode: independent tasks can execute concurrently before a synchronization step produces the combined report; it does not address this case, for which the relevant intervention is to capture step-level traces, tool inputs, outputs, and terminal outcomes.

D:

Downstream work depends on earlier validated outputs, so parallel execution risks using incomplete data in its own workflow, but that is a different correction from the present requirement to capture step-level traces, tool inputs, outputs, and terminal outcomes. The reported issue would remain.

Question 269

In a multi-step operations assistant, engineers must transfer an active task and its relevant context to a specialist that takes over the next interaction. The expected evidence is that the receiving agent continues with appropriate task context. Which implementation is the best match?

  1. Use Microsoft Agent Framework sequential workflow steps
  2. Use Microsoft Agent Framework handoff orchestration
  3. Use Microsoft Agent Framework routing or triage
  4. Use Microsoft Agent Framework parallel workflow branches

Explanation

Correct Answer

B:

Handoff patterns change which agent owns the next step while preserving needed state and authority boundaries. For this multi-step operations assistant, the intended evidence is that the receiving agent continues with appropriate task context. A parallel result merge is not the same as transferring conversational responsibility.

Incorrect Answers

A:

Sequential orchestration enforces dependencies such as research preceding drafting and review preceding publication; it may be used to pass a validated intermediate result from one specialized agent to the next in a required order in a multi-step operations assistant, but the current requirement is instead to transfer an active task and its relevant context to a specialist that takes over the next interaction. That separate capability does not prove that the receiving agent continues with appropriate task context; Running dependent stages in parallel can consume incomplete inputs, so selecting that alternative would not meet the stated acceptance check.

C:

Routing uses task interpretation to choose a domain agent instead of sending every request through every specialist; it may be used to direct a request to the appropriate specialist based on the detected task in a multi-step operations assistant, but the current requirement is instead to transfer an active task and its relevant context to a specialist that takes over the next interaction. That separate capability does not prove that the receiving agent continues with appropriate task context; A fixed sequential chain is unnecessarily rigid when only one branch should run, so selecting that alternative would not meet the stated acceptance check.

D:

Parallel orchestration permits independent tasks to execute together and joins outputs for a final synthesis; it may be used to have independent specialists analyze separate evidence sets concurrently before results are merged in a multi-step operations assistant, but the current requirement is instead to transfer an active task and its relevant context to a specialist that takes over the next interaction. That separate capability does not prove that the receiving agent continues with appropriate task context; It is unsuitable when later steps require data not produced until an earlier step completes, so selecting that alternative would not meet the stated acceptance check.

Question 270

Two parallel agents edit the same shared variable and overwrite each other's findings. What architecture change helps?

  1. Regression-test the router against representative existing and new intents
  2. Enforce handoff limits and an explicit escalation or terminal condition
  3. Gate the downstream stage on successful completion of its prerequisite
  4. Isolate branch state and merge outputs deterministically after completion

Explanation

Correct Answer

D:

Independent branch outputs avoid races; a controlled merge preserves all required findings. In the multi-step operations assistant, confirm the fix by retesting the affected behavior after isolating branch state and merge outputs deterministically after completion.

Incorrect Answers

A:

That change addresses another failure mode: routing classifiers and rules can introduce cross-category mistakes as new paths are introduced; it does not address this case, for which the relevant intervention is to isolate branch state and merge outputs deterministically after completion.

B:

Handoff cycles need bounded control flow to avoid infinite delegation and unnecessary model/tool calls in its own workflow, but that is a different correction from the present requirement to isolate branch state and merge outputs deterministically after completion. The reported issue would remain.

C:

Although useful elsewhere in the multi-step operations assistant, explicit completion dependencies keep sequential workflows from consuming unvalidated intermediate results; the present case is different and calls for the team to isolate branch state and merge outputs deterministically after completion.

Designing and Implementing a Microsoft Azure AI Solution (AI-102) Practice Tests