McAfee Secured

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 19

Question 180

A coordinator runs independent agents against a rate-limited research API. Which TWO orchestration controls matter?

  1. Shared concurrency and retry budgets across delegated workers
  2. Treat 429 failures as successful empty search responses
  3. Correlation IDs with result-set reconciliation
  4. Assume each worker owns the full API quota
  5. Assign results solely by arrival order

Explanation

Correct Answers

A:

Aggregate traffic must respect the backend's quota.

C:

Parallel completion needs explicit mapping to assigned subtasks.

Incorrect Answers

B:

Throttling does not prove that documents are absent.

D:

Uncoordinated workers can overwhelm the provider.

E:

Fast completion does not identify the original task.

Question 181

A photo captioning system should avoid identifying objects hidden behind a closed door. What prompt rule helps?

  1. Use the user's desired conclusion as visual truth
  2. Replace the photo with a generated room scene
  3. Describe only visible evidence and express uncertainty about occluded content
  4. List every object normally found in similar rooms

Explanation

Correct Answer

C:

Captions should not invent objects absent from the camera view.

Incorrect Answers

A:

Requested wording does not alter the actual pixels.

B:

Synthetic imagery is not evidence about the original view.

D:

Typical layouts do not prove hidden contents.

Question 182

A model tends to repeat the same phrase in marketing drafts. What prompt strategy might help without changing facts?

  1. Clear style constraints and varied evaluated examples appropriate to the task
  2. Disable validation for all content
  3. Remove the source facts to allow improvisation
  4. Increase output length indefinitely

Explanation

Correct Answer

A:

Prompt design can reduce repetition while preserving required factual content.

Incorrect Answers

B:

Stylistic repetition does not justify abandoning factual checks.

C:

Unbounded creativity can introduce unsupported claims.

D:

Longer text may amplify recurring wording.

Question 183

A tool integration leaks secrets into traces through raw HTTP headers. What telemetry change is required?

  1. Collect every Authorization header for full fidelity
  2. Redact or exclude credentials and sensitive payload fields before export
  3. Use a larger model to summarize leaked headers
  4. Disable all service authentication

Explanation

Correct Answer

B:

Observability should preserve diagnostic metadata without disclosing authentication material.

Incorrect Answers

A:

Raw bearer tokens would become accessible to log readers.

C:

Summarization does not remove the original secret from logs.

D:

Anonymous access is not a remedy for unsafe logging.

Question 184

A call assistant receives audio containing both English and Urdu in one sentence. What should be evaluated?

  1. Use a text-only benchmark as proof of audio quality
  2. Code-switching recognition and multilingual language handling
  3. Translate only the first recognized word
  4. Force one unrelated language label for the entire call

Explanation

Correct Answer

B:

Mixed-language speech may require models or settings supporting transitions.

Incorrect Answers

A:

Speech input introduces acoustic and recognition errors absent from text.

C:

Partial translation loses the speaker's full intent.

D:

Unsupported settings can corrupt terms from the other language.

Question 185

An entity extractor finds 'Paris' in a sentence about Paris Hilton and treats it as a city. What improves classification?

  1. Use only capitalization for entity classification
  2. Use surrounding context to assign entity type rather than surface string alone
  3. Remove all names so there can be no error
  4. Hardcode every occurrence of Paris as a city

Explanation

Correct Answer

B:

The same text token can refer to a person or a location.

Incorrect Answers

A:

Both persons and places can be capitalized.

C:

Deletion discards useful named-entity evidence.

D:

Name ambiguity requires contextual disambiguation.

Question 186

An AI agent occasionally performs a tool call against the wrong customer tenant. What control should be strengthened first?

  1. Increase model temperature to encourage better judgment
  2. Disable audit logs to prevent exposure
  3. Enforce tenant authorization at each tool and retrieval boundary
  4. Trust the tenant name printed in the user's message

Explanation

Correct Answer

C:

Deterministic resource checks prevent a model routing mistake from crossing isolation.

Incorrect Answers

A:

Sampling randomness cannot enforce cross-tenant authorization.

B:

Removing evidence does not stop unauthorized operations.

D:

Untrusted text cannot prove the caller's allowed tenant.

Question 187

A mobile insurance assistant must classify short claim descriptions on a constrained latency budget and needs no long reasoning. Which model choice should the architect benchmark first?

  1. A task-capable small language model deployed with measured accuracy and latency
  2. The largest reasoning model for every incoming message
  3. An image embedding model that returns vectors only
  4. A video-generation model trained for motion scenes

Explanation

Correct Answer

A:

A smaller model may satisfy classification requirements at lower cost and response time.

Incorrect Answers

B:

Maximum model scale is not justified without measured gains for the narrow classification task.

C:

Embeddings alone do not deliver the required final class label.

D:

Visual synthesis does not match text label classification requirements.

Question 188

A user asks a billing agent only to explain a charge. What tools should be available?

  1. A general database drop-table operation
  2. Read-only billing lookup tools, not refund or charge mutation without request
  3. An unrestricted production payment key in the prompt
  4. Refund tools exposed automatically on any billing keyword

Explanation

Correct Answer

B:

An informational task does not authorize financial side effects.

Incorrect Answers

A:

Destructive infrastructure access is unrelated to explanation.

C:

Broad credentials create unnecessary exposure.

D:

Tool presence can induce unintended monetary actions.

Question 189

A subagent reads confidential contracts but its task is only to return risk categories to its parent. What should be in the handoff?

  1. Every full contract copied into all parent messages
  2. A new production contract amendment without approval
  3. Only 'looks safe' with no source or reasoning
  4. Bounded conclusions with cited clauses and uncertainty, excluding unrelated raw contract content

Explanation

Correct Answer

D:

A narrow handoff supports review without expanding sensitive context.

Incorrect Answers

A:

Unnecessary raw disclosure increases context and privacy risk.

B:

The subagent was delegated analysis, not legal execution.

C:

An unsupported summary is not auditable contract analysis.

Question 190

A model self-critique identifies a likely error but proposes an untested correction. What should the release system do?

  1. Treat self-critique as independent human approval
  2. Discard the passing previous tests after one suggestion
  3. Apply the targeted change and rerun affected deterministic tests and evaluations
  4. Release the critique text as a substitute for a fixed product

Explanation

Correct Answer

C:

A critic suggestion is a hypothesis until revised behavior is checked.

Incorrect Answers

A:

Model review cannot be misrepresented as external SME signoff.

B:

Unnecessary evidence loss does not strengthen confidence.

D:

Observations alone do not correct application behavior.

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