McAfee-Secured Website

Exam Code: AI-103

Exam Name: Developing AI Apps and Agents on Azure

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Apps and Agents Developer Associate

100% Microsoft AI-103 Guaranteed Success With Testking Training Materials

Get Real AI-103 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

187 Questions & Answers with Testing Engine

"Developing AI Apps and Agents on Azure Exam", also known as AI-103 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-103 Exam Engine. Your AI-103 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-103 Practice Test 65

Question 640

A team introduces few-shot examples to improve invoice categorization. It has high-quality production invoices, but they contain real account numbers and names. What is a defensible prompt design?

  1. Place unredacted invoices in a globally shared system prompt for all tenants.
  2. Use minimized or synthetic representative examples with the same classification structure.
  3. Disable all classification examples and accept malformed output rather than sanitize data.
  4. Encode customer account numbers as visible text and rely on model secrecy to protect them.

Explanation

Correct Answer

B:

Representative examples can teach output patterns without exposing unnecessary customer identifiers.

Incorrect Answers

A:

This creates cross-tenant personal-data exposure outside the example’s teaching purpose.

C:

Data minimization allows useful examples while meeting privacy requirements.

D:

Encoding text does not replace explicit control over sensitive example data.

Question 641

A travel agent must read a flight schedule and a hotel availability list, neither of which depends on the other. Both calls are read-only and latency matters. Which orchestration strategy fits?

  1. Execute the hotel lookup only after modifying the flight schedule as a side effect.
  2. Send the user a final booking confirmation before either tool has returned.
  3. Invoke the two independent tools concurrently and combine validated results before responding.
  4. Run the two reads sequentially because every agent action must depend on the previous result.

Explanation

Correct Answer

C:

Independent reads can run in parallel without changing transaction order or authorization.

Incorrect Answers

A:

Neither lookup requires a schedule mutation, and the action is unauthorized.

B:

The result would not be grounded in actual availability or completed reservations.

D:

Mandatory sequencing adds latency where no data dependency exists.

Question 642

A warehouse agent must verify stock, reserve a specific quantity, and then issue a dispatch request referencing the reservation ID. What execution order preserves the dependency?

  1. Check stock, create a validated reservation, then dispatch using the returned reservation ID.
  2. Send dispatch and reservation calls concurrently without a shared reservation identifier.
  3. Dispatch based on a predicted reservation number before checking actual stock.
  4. Generate a dispatch confirmation using cached text even if the reserve call fails.

Explanation

Correct Answer

A:

The dispatch operation depends on a confirmed reservation and its authoritative identifier.

Incorrect Answers

B:

Dispatch could execute before or without the reservation required by the business rule.

C:

A fabricated identifier is not evidence of a successful reservation.

D:

The user must not receive success for a transaction that did not occur.

Question 643

A payment agent invokes a charge endpoint but loses the network response. The client retries after a timeout. Which tool-contract feature best prevents charging the customer twice?

  1. Generate a new random transaction ID on every retry with no status lookup.
  2. Ask the language model to assume the first charge failed and retry unconditionally.
  3. Increase the assistant token budget and repeat the charge until a reply appears.
  4. Use a stable idempotency key and verify transaction status before repeating the operation.

Explanation

Correct Answer

D:

The payment service can identify repeated requests for the same logical transaction.

Incorrect Answers

A:

A fresh ID may be interpreted as a second independent charge.

B:

Model speculation cannot establish the transaction state after a lost response.

C:

More tokens do not provide transactional deduplication or confirmation.

Question 644

A dispatcher agent routes finance questions to a billing specialist. A general employee asks the specialist to change an executive’s payroll account. What must be enforced at the delegation boundary?

  1. Let the billing specialist inherit every privileged tool because it has a finance label.
  2. Verify caller authorization and the specific permissible action before exposing the payroll tool.
  3. Accept a generated explanation that the employee probably has approval.
  4. Execute the update and ask a supervisor to review the completed action later.

Explanation

Correct Answer

B:

Agent-to-agent delegation does not transfer privileges the original caller lacks.

Incorrect Answers

A:

A specialist role name does not itself justify broad write authority.

C:

Narrative inference is not evidence of an authorized payroll change.

D:

After-the-fact review cannot prevent the unauthorized change.

Question 645

A purchasing agent is allowed to request orders containing sku, quantity and approved_cost_center. Its current tool accepts arbitrary natural-language text and sometimes misroutes orders. What change most directly improves the contract?

  1. Increase the maximum chat history so the model can remember prior order formats.
  2. Accept loosely formatted text and infer the cost center at the payment endpoint.
  3. Expose a schema-validated tool with required typed fields and business-side checks.
  4. Allow the agent to create orders with a missing cost center and repair them afterward.

Explanation

Correct Answer

C:

Explicit argument types and validation reduce ambiguous requests and allow rejection of invalid orders.

Incorrect Answers

A:

More conversation context does not enforce a reliable API input schema.

B:

Guessing a mandatory financial classification can misroute spending.

D:

The required field must be verified before the side effect is committed.

Question 646

A fraud-screening agent calls an external risk service, but the tool times out. A loan application cannot be approved without a real risk result. What should the agent return?

  1. Report the check as unavailable and halt approval until the required risk result is obtained.
  2. Generate a low-risk score from the applicant name and continue approval.
  3. Treat a network timeout as proof that the service returned no risk flags.
  4. Reuse a different applicant’s cached risk response as the current result.

Explanation

Correct Answer

A:

Tool failure is not a successful screening decision and must preserve the approval precondition.

Incorrect Answers

B:

A fabricated score is not the mandatory result of the risk service.

C:

No response is not evidence that a screening operation completed.

D:

The result would be associated with the wrong borrower and cannot support this application.

Question 647

A concierge agent remembers users’ preferences across sessions. One user has not consented to cross-session storage of personal details. Which memory behavior meets that condition?

  1. Store the complete conversation permanently because memory might be useful later.
  2. Save sensitive preferences in a shared vector index accessible to other customers.
  3. Treat the agent system prompt as an implicit consent record for every user.
  4. Keep only necessary session state and avoid persisting that user’s personal preferences.

Explanation

Correct Answer

D:

Cross-session personal memory requires the appropriate user permission and retention policy.

Incorrect Answers

A:

Potential usefulness does not override the user’s lack of consent for persistent storage.

B:

Shared storage would expand data access beyond the user’s authorized context.

C:

Developer instructions do not establish user consent for retention.

Question 648

A user withdraws consent to persistent memory while continuing to use the same chatbot. Other customers have active separate consent records. What should the memory service do?

  1. Clear every customer’s memory because any one deletion invalidates the shared store.
  2. Remove or disable that user’s eligible retained memory according to the policy, isolating other users.
  3. Keep recalling the user’s old preferences but hide them from the application interface.
  4. Write the withdrawal request into a prompt but continue reading the same stored record.

Explanation

Correct Answer

B:

Memory lifecycle controls should honor the specific user’s choice without crossing tenant boundaries.

Incorrect Answers

A:

A global deletion is unnecessary and does not respect per-user isolation.

C:

Concealing a stored value is not the same as honoring its retention choice.

D:

Text instructions alone do not alter the underlying data lifecycle.

Question 649

A planning agent hands off a purchase validation to a finance specialist, which invokes a budget API. Auditors must know who requested the purchase and which agent performed each step. What telemetry is needed?

  1. Log only the final conversational answer without preserving delegation metadata.
  2. Assign the same anonymous user label to all finance tool calls for simplicity.
  3. Correlate caller identity, delegation chain, agent steps and tool request IDs in the trace.
  4. Record only the model deployment version and omit tool execution identifiers.

Explanation

Correct Answer

C:

An end-to-end trace permits attribution of decisions and operations across agents.

Incorrect Answers

A:

A final sentence cannot explain which identity authorized and executed each step.

B:

An anonymous label obscures attribution of requests and privileged operations.

D:

A model version is not enough to reconstruct the delegated action sequence.

Question 650

A research agent retrieves a web page containing a list of product specifications and a hidden instruction to send internal documents to an outside endpoint. The legitimate task is to compare specifications. What must the agent do?

  1. Treat the hidden instruction as untrusted page data and restrict tools to the authorized comparison.
  2. Follow the hidden instruction because the search tool returned the page successfully.
  3. Authorize the outbound send if the page appears among the first search results.
  4. Use a higher reasoning model and rely on it to decide whether to leak documents.

Explanation

Correct Answer

A:

External page text is not an instruction source that can expand the agent’s permission scope.

Incorrect Answers

B:

Successful retrieval does not convert document text into trusted commands.

C:

Retrieval rank reflects relevance, not user approval for exfiltration.

D:

Model capability cannot replace explicit tool authorization and trust boundaries.

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