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 25

Question 240

A model answer is truncated even at low temperature because the API has a small `max_output_tokens` value. Which parameter should be adjusted?

  1. Raise search top-k without changing the completion limit.
  2. Increase temperature while leaving the same tiny output cap.
  3. Increase the output-token limit within model and cost constraints.
  4. Disable source citations because the model will gain unlimited completion capacity.

Explanation

Correct Answer

C:

The completion length cap directly limits emitted tokens, unlike sampling parameters.

Incorrect Answers

A:

Retrieval breadth does not enlarge the response budget.

B:

Sampling variability cannot extend a hard output-token allowance.

D:

Removing references does not eliminate the explicit token cap.

Question 241

An inventory agent needs to look up an item by SKU but must never modify stock counts. Which tool definition best enforces the intended purpose?

  1. Define a tool named change_anything with a free-form string payload.
  2. Expose a read-only get_inventory(sku) function with typed required arguments and scoped access.
  3. Provide a tool that deletes stale products whenever it cannot find a SKU.
  4. Allow the agent to invent stock values when the API is unreachable.

Explanation

Correct Answer

B:

A narrow schema and a non-mutating endpoint permit authorized inventory retrieval without unrelated effects.

Incorrect Answers

A:

An unrestricted write interface is unnecessarily powerful for a lookup-only task.

C:

Failed lookups do not justify destructive changes to stock records.

D:

A missing live result must be reported as uncertainty or a tool failure, not represented as fact.

Question 242

A service agent may create a repair order only after a supervisor approves the specific case. What tool contract should be implemented?

  1. Expose create_order with no user identity or approval fields.
  2. Ask the model in natural language whether it thinks approval probably happened.
  3. Let a search index automatically create the order whenever it retrieves a related policy.
  4. Define a work-order creation tool that requires a valid approval identifier.

Explanation

Correct Answer

D:

The business API can validate a scoped approval and reject unauthorized writes independently of model text.

Incorrect Answers

A:

Without approval evidence, the execution path cannot enforce the business authorization requirement.

B:

Model speculation is not an auditable approval credential for a business-side effect.

C:

Retrieval finds evidence but is not an authorization event for creating a new record.

Question 243

A shared internal agent exposes a payroll-change tool. Ordinary employees can ask general HR questions, but only authorized payroll officers may invoke the change operation. What control is essential?

  1. Trust that the agent’s system prompt will keep unauthorized employees from reaching the tool.
  2. Give the agent organization-wide payroll write credentials for all conversation users.
  3. Enforce caller-specific payroll authorization at tool execution.
  4. Grant every user the payroll writer role, but ask the agent to be cautious.

Explanation

Correct Answer

C:

Authorization at the actual execution boundary prevents unprivileged requests even if the model chooses a tool.

Incorrect Answers

A:

Text instructions alone are not a secure permission check for a privileged external API.

B:

Shared broad credentials allow tool misuse if authorization is not enforced per user and action.

D:

Expanding roles violates the documented separation of duties and increases risk.

Question 244

A customer service agent is told to resolve ticket T-1924 by checking shipping status, retrieving policy, and then deciding whether to request approval for a refund. Which design best distinguishes its role from its actions?

  1. Separate an agent's business goal from the tool steps that achieve it.
  2. Embed live tracking numbers and refund decisions permanently in the system prompt.
  3. Give the agent one unrestricted shell tool and no workflow policy.
  4. Treat the customer’s request text as automatic authorization to issue any refund.

Explanation

Correct Answer

A:

The agent goal guides planning; tools define executable operations with explicit authorization and return values.

Incorrect Answers

B:

Changing business state does not belong in static instructions and may quickly become wrong.

C:

An unconstrained execution surface is not needed for a small approved customer support process.

D:

A user request must still pass enterprise refund rules and any required human approval.

Question 245

A helpdesk agent receives five consecutive messages about the same open ticket. Separate runs lose what was already confirmed. What identity should accompany each turn?

  1. Pass a stable conversation/thread identifier associated with the authorized support session.
  2. Use one global thread ID shared by all customers.
  3. Use the model deployment name as the customer's conversation key.
  4. Generate a fresh conversation identifier on every message rather than reusing the active support session.

Explanation

Correct Answer

A:

Thread identity lets the agent retrieve relevant prior turns for the same conversation.

Incorrect Answers

B:

A shared identifier risks mixing conversations and private data.

C:

A deployment serves many sessions and is not a unique thread identifier.

D:

A new thread key each turn discards context that the user intended to preserve.

Question 246

An agent only needs the last few turns to clarify a single order question, with no consent to store preferences across visits. What memory scope fits?

  1. Create a permanent global profile from every clarification message.
  2. Discard all turns before the next reply, including the unfulfilled order clarification.
  3. Store conversations in model training weights for immediate cross-session access.
  4. Use expiring conversation history for this short-lived session.

Explanation

Correct Answer

D:

One-call continuity does not justify persisting customer details as long-term memory.

Incorrect Answers

A:

Unneeded durable retention exceeds the limited support task.

B:

Losing short-term context can prevent completion of the ongoing request.

C:

Training is not a consented, user-addressable thread store.

Question 247

A user opted in to keeping an explanation-language preference. It works on a laptop, but a new phone session loads another customer’s language selection from shared memory. What identity-scoping change is required?

  1. Store the preferred language in a single global agent setting shared by all users.
  2. Apply the preference only to the current temporary session, with no durable user association.
  3. Persist and retrieve consented preferences under the authenticated user identity across clients.
  4. Infer a language from the newest device IP address and overwrite consented preferences.

Explanation

Correct Answer

C:

Authenticated user-scoped storage and retrieval preserve durable preferences without exposing another customer’s data.

Incorrect Answers

A:

A global preference crosses user boundaries and is the cause of the data mix-up.

B:

A temporary session-only value cannot provide the requested consented cross-device continuity.

D:

IP-based guessing is not identity-bound and must not replace the explicit user preference.

Question 248

A user deletes their agent personalization profile, but old embeddings remain in a memory store and are still retrieved. What should be enforced?

  1. Enforce retention and deletion rules on agent memory stores.
  2. Clear only the latest chat message while preserving durable profile vectors.
  3. Hide the settings screen but continue retrieving all old memories.
  4. Increase model temperature so it is less likely to mention saved details.

Explanation

Correct Answer

A:

Deletion must cover retrievable derived records, not just a visible profile label.

Incorrect Answers

B:

Deleting thread text does not remove separate persistent memories.

C:

UI concealment does not honor the data lifecycle request.

D:

Randomness is not a deletion or access-control mechanism.

Question 249

Employees ask an agent about the current parental leave handbook, which changes often. What source should be connected?

  1. Attach an indexed knowledge source to answer employee handbook questions.
  2. Ask the model to infer company policy from general HR conventions.
  3. Put last year's handbook summary permanently into the agent instructions.
  4. Attach a broad inventory lookup with untyped parameters instead of a ticket-specific validated tool.

Explanation

Correct Answer

A:

A managed knowledge tool supplies current supporting passages without retraining the agent.

Incorrect Answers

B:

General knowledge is not an authoritative private handbook source.

C:

Static summaries may become outdated and omit exceptions.

D:

A generic interface may not enforce the intended input contract and permissions for the actual ticket workflow.

Question 250

A delivery agent needs to report the exact live shipping state for tracking ID T301. It has no access to the logistics database. What tool integration is required?

  1. Use approximate semantic similarity on tracking descriptions as the sole identifier lookup method.
  2. Use model pretraining on historical shipment examples as proof of current status.
  3. Use function calling to retrieve a live shipping status.
  4. Ask the agent to generate a plausible status based on the tracking number format.

Explanation

Correct Answer

C:

Tool calling retrieves the current system-of-record shipment status instead of guessing.

Incorrect Answers

A:

Near-neighbor records can refer to another parcel; tracking requires an exact authorized identifier.

B:

Static examples do not contain later tracking events.

D:

Identifier syntax cannot establish the live parcel position.

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