Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 76

Question 750

A planned AI solution must support this situation: Agent supports refunds, order status and generic FAQs. Its prototype fails because the test suite spends most cases on low-risk FAQs. Which design best fits the constraint?

  1. Generate negative cases from tool and identity specifications
  2. Set acceptance thresholds and reproducible baselines
  3. Build risk-based test coverage for critical user journeys
  4. Test multi-turn context retention and clarification

Explanation

Correct Answer

C:

Agent supports refunds, order status and generic FAQs; its failing behavior is that the test suite spends most cases on low-risk FAQs; applying the underlying design principle—risk-weighting prioritizes high-impact actions and failure paths—connects the requirement to an implementable control.

Incorrect Answers

A:

For agents can call APIs with customer IDs and amount values, this other approach relies on the fact that specification-guided generation covers missing fields and denied roles; the issue here is the test suite spends most cases on low-risk FAQs; that mechanism does not establish the required risk-weighting prioritizes high-impact actions and failure paths.

B:

For another type of architecture, predefined gates weigh accuracy, latency and business risk is useful; this design must instead support agent supports refunds, order status and generic FAQs despite the problem where the test suite spends most cases on low-risk FAQs.

D:

In the case where agent must remember order context while asking for missing details, the alternate design depends on multi-turn tests assess state tracking, entity separation and follow-up questions; it does not correct the target failure, the test suite spends most cases on low-risk FAQs.

Question 751

Support engineers report this failure: The pilot has no team for monitoring, incidents, or iteration. The business context is: An agent will evolve continuously after its initial project delivery. What evidence should be collected to isolate the cause?

  1. Correlate throttling errors with fallback activation
  2. Review the handover plan for service ownership gaps
  3. Trace transactions using end-to-end correlation IDs
  4. Identify workflow steps with fixed rules versus variable language

Explanation

Correct Answer

B:

The reported fault is that the pilot has no team for monitoring, incidents, or iteration in a system where an agent will evolve continuously after its initial project delivery; evidence from reviewing the handover plan for service ownership gaps can show whether long-lived product ownership funds ongoing evaluation and operations.

Incorrect Answers

A:

The proposed inspection is associated with peak load can saturate a primary model endpoint, where health-aware fallback uses a compatible approved model when needed; it would not expose the failure mode that the pilot has no team for monitoring, incidents, or iteration with the required specificity.

C:

The proposed inspection is associated with copilot, Dataverse, Dynamics and custom APIs share one process, where contracts and ownership boundaries identify responsibility for each state change; it would not expose the failure mode that the pilot has no team for monitoring, incidents, or iteration with the required specificity.

D:

The evidence this method produces applies to a workflow where a business process mixes deterministic calculations and open-ended judgments; it cannot by itself establish why the pilot has no team for monitoring, incidents, or iteration in the current scenario.

Question 752

A quality engineer is writing a regression case after the incident where rewriting the entire conversational layer adds little value. The underlying business scenario is: Standard conversation needs a proprietary scoring service. What should the test exercise?

  1. Test action invocation and failure handling within the intended host
  2. Update a case and verify an authorized sales workflow sees it
  3. Prototype a unique scheduling dependency with the custom runtime
  4. Simulate scoring-service outage and validate safe fallback

Explanation

Correct Answer

D:

For the required workflow where standard conversation needs a proprietary scoring service, the test needs to reproduce or prevent the failure where rewriting the entire conversational layer adds little value; simulating scoring-service outage and validating safe fallback exercises the path where a managed host can invoke an independently governed domain API.

Incorrect Answers

A:

In the alternate setting where users need familiar Copilot behavior plus an internal status lookup, the test exercises supported extensibility adds an action while retaining the host's orchestration; even a pass would not show that rewriting the entire conversational layer adds little value has been prevented.

B:

The verification concerning sales and service agents need the same customer and case context checks shared authorized records support coordinated business context against an opportunity owner sees a stale service escalation status; its result says little about the observed issue where rewriting the entire conversational layer adds little value.

C:

The tested behavior is that owning the engine provides control over models, routing and execution; the necessary evidence here concerns a managed host can invoke an independently governed domain API and the previously observed rewriting the entire conversational layer adds little value.

Question 753

A test plan focuses on a previously missed failure: Direct replacement removes the dependable baseline. It occurred when a tuned model claims higher accuracy but adds latency. Which check directly validates that behavior?

  1. Roll back a staged faulty transformation and compare output
  2. Re-run a sampled historical release on its frozen corpus
  3. Degrade sample images and test extraction stability
  4. Send an identical holdout to champion and challenger

Explanation

Correct Answer

D:

For the required workflow where a tuned model claims higher accuracy but adds latency, the test needs to reproduce or prevent the failure where direct replacement removes the dependable baseline; sending an identical holdout to champion and challenger exercises the path where candidate comparison evaluates quality, latency and safety under matched load.

Incorrect Answers

A:

For a preprocessing release changes category labels incorrectly, the alternate test is about whether versioned data transformations and model artifacts enable consistent rollback; it cannot prove the outcome needed after the reported regression where direct replacement removes the dependable baseline.

B:

The suggested test serves an updated agent performs differently months after release, in which old scores cannot be reproduced because prompts and tests changed; the present scenario instead concerns the reported problem where direct replacement removes the dependable baseline.

C:

The verification concerning production documents differ from clean training samples checks stress tests measure sensitivity to expected input variation against scanned forms suddenly produce inaccurate predictions; its result says little about the observed issue where direct replacement removes the dependable baseline.

Question 754

One enterprise rollout has 2 separately testable requirements. (1) Business use: A prompt injection requests data export to an outside address. Known defect: Model-level refusal alone may not prevent a tool call. (2) Business use: Customers believe generated advice is a human decision. Known defect: Users cannot identify automated recommendations. Which 2 regression tests directly satisfy these stated cases? Choose TWO.

  1. Inject delay into one dependency and locate the timing change
  2. Test null, long and unusual user inputs
  3. Inject a malicious export instruction into a source document
  4. Observe users interpreting the automated decision flow

Explanation

Correct Answers

C:

For the business workflow where a prompt injection requests data export to an outside address, injecting a malicious export instruction into a source document challenges the failure that model-level refusal alone may not prevent a tool call; the release also needs a separate check for customers believe generated advice is a human decision.

D:

For the business workflow where customers believe generated advice is a human decision, observing users interpreting the automated decision flow challenges the failure that users cannot identify automated recommendations; the release also needs a separate check for a prompt injection requests data export to an outside address.

Incorrect Answers

A:

This regression test is associated with the different failure where engineers blame generation although the backend is slow in a workflow where agents regularly exceed an interactive response-time objective; it does not directly solve the listed conditions: model-level refusal alone may not prevent a tool call; users cannot identify automated recommendations.

B:

This regression test is associated with the different failure where generated output occasionally omits required keys in a workflow where a Copilot prompt feeds records into downstream automation; it does not directly solve the listed conditions: model-level refusal alone may not prevent a tool call; users cannot identify automated recommendations.

Question 755

A policy review covers an agent for which network retries can repeat a single purchase request. Its most relevant exposure is that an API retry might create a second purchase for the same user instruction. What safeguard should be required before release?

  1. Label unsupported explanations as hypotheses or abstain
  2. Stop and compensate when a required approval fails
  3. Preserve conversation context and avoid infinite re-prompting
  4. Ensure the backend rejects repeat writes for the same intent

Explanation

Correct Answer

D:

In the scenario where network retries can repeat a single purchase request, the explicit risk is that an API retry might create a second purchase for the same user instruction; enforcement must follow the mechanism by which idempotency keys tie repeated submissions to one business operation.

Incorrect Answers

A:

The proposed control targets an analytical response might present an unsupported hypothesis as a ledger fact, while the given risk scenario is an API retry might create a second purchase for the same user instruction; these exposures arise in different operational contexts.

B:

This restriction is associated with approval processing requires sequential verification and notification and addresses the exposure that a flow might notify the customer even though a required approval failed; it does not specifically protect the target workflow against an API retry might create a second purchase for the same user instruction.

C:

The safeguard for contact Center agents sometimes cannot resolve complex billing intent deals with a different threat: a caller might get trapped in repeated fallback without a safe escalation route; even after it is enforced, the exposure that an API retry might create a second purchase for the same user instruction remains possible in the target workflow where idempotency keys tie repeated submissions to one business operation.

Question 756

The rollout supports this workflow: An intake task requires several prerequisite fields. Leaders want a metric that responds to the reported shortcoming that agent sends incomplete requests when users answer out of order. Which measure best reflects success?

  1. Regulated replies with complete mandatory clauses
  2. Valid returns completed without unrelated writes
  3. Correctly sequenced workflow executions
  4. Analyst-accepted explanations with verified references
  5. Completed tasks with all required input fields

Explanation

Correct Answer

E:

The use case involves an intake task requires several prerequisite fields and the baseline failure that agent sends incomplete requests when users answer out of order; the KPI tracks completed tasks with all required input fields, so improvement has an interpretable operational meaning.

Incorrect Answers

A:

The alternative tracks regulated replies with complete mandatory clauses, arising from the mechanism where templating and approved clauses limit generated variation; its movement alone says little about the failure that agent sends incomplete requests when users answer out of order.

B:

An employee requests a standard equipment return initiation; its intended measurement is valid returns completed without unrelated writes; substituting that result here would not capture whether agent sends incomplete requests when users answer out of order has been corrected.

C:

The premise of this indicator is defined flow steps make sequencing and error paths explicit; the architecture being evaluated instead serves an intake task requires several prerequisite fields, and its success is measured by completed tasks with all required input fields.

D:

The premise of this indicator is grounded explanatory responses distinguish evidence from hypothesis; the architecture being evaluated instead serves an intake task requires several prerequisite fields, and its success is measured by completed tasks with all required input fields.

Question 757

Before introducing a fix, the team must explain why a shared connection identity grants sandbox agents production write access. The affected setting is: Test automation must not reach live customer records. What should be inspected?

  1. Inspect cache keys and authorization on reused entries
  2. Audit tokens and connection bindings for each stage
  3. Inspect solution layers for unmanaged production changes
  4. Compare target environment version and active channel metadata

Explanation

Correct Answer

B:

The reported fault is that a shared connection identity grants sandbox agents production write access in a system where test automation must not reach live customer records; evidence from auditing tokens and connection bindings for each stage can show whether environment-scoped identities isolate access and simplify incident containment.

Incorrect Answers

A:

The evidence this method produces applies to a workflow where shared result caching makes answers fast across roles; it cannot by itself establish why a shared connection identity grants sandbox agents production write access in the current scenario.

C:

Production components should not be edited casually; its matching diagnostic check is reasonable for a maker changes production agent logic outside the release process; the present investigation needs evidence for the different failure where a shared connection identity grants sandbox agents production write access.

D:

In the process where solution import succeeds but users still see the previous behavior, the diagnostic procedure could explain deployment checks stop at package import success; this particular incident instead turns on whether environment-scoped identities isolate access and simplify incident containment and why a shared connection identity grants sandbox agents production write access.

Question 758

The pilot needs a durable correction: A monolingual analysis misses non-English escalations. This affects the workflow in which support documents arrive in several customer languages. Which design choice should the architect approve?

  1. Choose speech processing for transcribed contact-center audio
  2. Use translation capabilities before downstream review
  3. Use agent reasoning for uncertain multi-step analysis with bounded tools
  4. Configure contextual customer-response drafting with policy sources

Explanation

Correct Answer

B:

For the need that support documents arrive in several customer languages, the relevant mechanism is that translation provides standardized text input while preserving original evidence; that architectural dependency addresses the observed failure where a monolingual analysis misses non-English escalations.

Incorrect Answers

A:

Manual notes omit important customer statements; the alternate design can address that situation because speech transcription transforms audio to text for subsequent analysis, whereas this question needs a solution to a monolingual analysis misses non-English escalations.

C:

Reasoning can plan through evidence and tools within enforced boundaries; its main purpose is procurement exceptions require comparing several policies and records, not correction of the reported defect where a monolingual analysis misses non-English escalations in the target process.

D:

The technical behavior of this alternative is that contextual grounding provides relevant policy and customer facts; the user case being assessed is support documents arrive in several customer languages with a different failure: a monolingual analysis misses non-English escalations.

Question 759

The architecture board has evidence that a release uses stale prompts against a new tool signature. The affected situation is: Model and tool changes are promoted independently. What change most directly establishes the missing behavior?

  1. Treat retrieved documents as untrusted evidence rather than instructions
  2. Use progressive deployment and rollback for a stable agent endpoint
  3. Version agent instructions, tool definitions and model references together
  4. Promote grounding datasets through quality gates

Explanation

Correct Answer

C:

Without the intended behavior, a release uses stale prompts against a new tool signature; the proposed architecture establishes the specific design principle that a complete agent release manifest pins related runtime artifacts for the process where model and tool changes are promoted independently.

Incorrect Answers

A:

The alternative addresses the agent obeys text retrieved from a supplier document by a design in which trust hierarchy separates lower-trust content from agent instructions; it does not supply the missing relationship for model and tool changes are promoted independently, where a release uses stale prompts against a new tool signature.

B:

Canary traffic and versioned rollback reduce release risk; its main purpose is business applications depend on the agent endpoint continuously, not correction of the reported defect where a release uses stale prompts against a new tool signature in the target process.

D:

There is a valid place for the mechanism that ingestion gates check classification, freshness and ownership, namely new documents enter a production source automatically; it does not remove the specific defect that a release uses stale prompts against a new tool signature from this case.

Question 760

Operations has not established the cause of this defect: A custom replacement duplicates the host experience unnecessarily. The associated setting is: Users need familiar Copilot behavior plus an internal status lookup. Which inspection should precede remediation?

  1. Inspect connector crawl scope, schema and ACL mappings
  2. Observe navigation abandonment and contextual data transfer
  3. Audit rollout settings and incident timelines for uncontrolled changes
  4. Check whether the required API can be surfaced through supported actions

Explanation

Correct Answer

D:

In the setting where users need familiar Copilot behavior plus an internal status lookup, checking whether the required API can be surfaced through supported actions can isolate the failure in which a custom replacement duplicates the host experience unnecessarily, because supported extensibility adds an action while retaining the host's orchestration.

Incorrect Answers

A:

The proposed probe concerns teams want approved external records searchable in Copilot, where connectors expose external enterprise content to supported Microsoft search/Copilot experiences; this incident instead requires evidence about supported extensibility adds an action while retaining the host's orchestration and the failure that a custom replacement duplicates the host experience unnecessarily.

B:

Observing navigation abandonment and contextual data transfer explores a separate connection, where app-hosted integration brings relevant AI actions into an owned user interface; the diagnosed requirement is users need familiar Copilot behavior plus an internal status lookup, and the actual defect is a custom replacement duplicates the host experience unnecessarily.

C:

In the separate scenario where new behavior must be evaluated without exposing the full workforce, the same investigation could explain a prompt update changes decisions for all users at once; it does not isolate the reported cause in a system where users need familiar Copilot behavior plus an internal status lookup.

Agentic AI Business Solutions Architect (AB-100) Practice Tests