Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 29

Question 280

Implementation alternatives must be compared for a workflow where operations must distinguish healthy traffic from quality incidents. The pilot shows this gap: Teams discover failures through complaints several days later. Which alternative should lead the solution?

  1. Inspect retrieval spans when cited answers omit evidence
  2. Test negative and adversarial cases alongside happy paths
  3. Create actionable SLOs and incident alerts for agent flows
  4. Connect Foundry agent tracing to Application Insights

Explanation

Correct Answer

C:

The mechanism that service-level objectives and alert thresholds turn telemetry into response signals fits the scenario where operations must distinguish healthy traffic from quality incidents; it addresses the defect that teams discover failures through complaints several days later.

Incorrect Answers

A:

For an assistant responds fluently but cites unrelated support articles, this other approach relies on the fact that retrieval telemetry distinguishes evidence search from answer generation; the issue here is teams discover failures through complaints several days later; that mechanism does not establish the required service-level objectives and alert thresholds turn telemetry into response signals.

B:

The alternative addresses a malicious instruction triggers a privileged tool by a design in which negative tests expose policy and validation boundaries; it does not supply the missing relationship for operations must distinguish healthy traffic from quality incidents, where teams discover failures through complaints several days later.

D:

Agent trace spans provide component-level execution evidence through connected telemetry; its main purpose is operations need to correlate prompts, model calls and tools in a Foundry workflow, not correction of the reported defect where teams discover failures through complaints several days later in the target process.

Question 281

A business stakeholder describes this operating situation: A rewritten prompt improves a single demonstration. Testing instead shows that slightly different wording breaks correct entity selection. What design approach is most appropriate?

  1. Test grounded prompts for citation and abstention behavior
  2. Test prompt changes on paraphrases and edge conditions
  3. Use production feedback to propose targeted regression tests
  4. Assess robustness under realistic drift and noisy input

Explanation

Correct Answer

B:

Because regression sets cover lexical variation and task boundaries, this approach corrects the reported breakdown where slightly different wording breaks correct entity selection; the affected situation is that a rewritten prompt improves a single demonstration.

Incorrect Answers

A:

There is a valid place for the mechanism that evidence and refusal evaluations enforce source-supported claims, namely policy answers must be tied to a current source; it does not remove the specific defect that slightly different wording breaks correct entity selection from this case.

C:

For support tickets reveal an unusual multistep conversation failure, this other approach relies on the fact that feedback-derived examples convert incidents into lasting safeguards; the issue here is slightly different wording breaks correct entity selection; that mechanism does not establish the required regression sets cover lexical variation and task boundaries.

D:

For production documents differ from clean training samples, this other approach relies on the fact that stress tests measure sensitivity to expected input variation; the issue here is slightly different wording breaks correct entity selection; that mechanism does not establish the required regression sets cover lexical variation and task boundaries.

Question 282

In a business situation where agents need structured case summaries at handoff, a previous build showed this defect: Summaries omit customer impact and current ownership. Which reproducible test belongs in the release gate?

  1. Change a permitted credit field and verify updated authorized retrieval
  2. Test duplicate legal names with different external IDs
  3. Query as multiple roles and confirm only authorized fields appear
  4. Test escalated and reopened cases for required fields

Explanation

Correct Answer

D:

Testing escalated and reopened cases for required fields challenges the path where tailored instructions and contextual fields standardize case handoff content for the business setting agents need structured case summaries at handoff; the specific defect to exclude is that summaries omit customer impact and current ownership.

Incorrect Answers

A:

In a solution where sellers need current credit exposure held outside CRM, changing a permitted credit field and verifying updated authorized retrieval could catch the defect that Copilot relies on last quarter's manually imported account report; the current release gate must challenge the failure where summaries omit customer impact and current ownership.

B:

The verification concerning two systems share company names but not primary keys checks a mapping service resolves stable identities before enrichment against Copilot attaches one customer's risk note to another account; its result says little about the observed issue where summaries omit customer impact and current ownership.

C:

Querying as multiple roles and confirming only authorized fields appear verifies an adjacent objective where sales copilots should not expose sensitive margin assumptions; it leaves untested the causal path in which tailored instructions and contextual fields standardize case handoff content for this case.

Question 283

A quality engineer is writing a regression case after the incident where few-shot prompts require excessive tokens and yield inconsistent formats. The underlying business scenario is: A stable specialized classification style persists despite prompt examples. What should the test exercise?

  1. Demonstrate a custom multi-step workflow and failure recovery
  2. Evaluate held-out examples against base-model baselines
  3. Test adversarial lookalike cases not present in prompt examples
  4. Submit edge cases and validate response shape

Explanation

Correct Answer

B:

For the required workflow where a stable specialized classification style persists despite prompt examples, the test needs to reproduce or prevent the failure where few-shot prompts require excessive tokens and yield inconsistent formats; evaluating held-out examples against base-model baselines exercises the path where fine-tuning can adapt recurring behavior using vetted examples.

Incorrect Answers

A:

In the alternate setting where the solution demands its own model selection and hosting policies, the test exercises custom-engine architecture controls runtime, model calls and integrations; even a pass would not show that few-shot prompts require excessive tokens and yield inconsistent formats has been prevented.

C:

Although the exercise might catch a generic instruction cannot express subtle category boundaries, it would not directly validate the mechanism that fine-tuning can adapt recurring behavior using vetted examples in the target implementation.

D:

In a solution where agents return inconsistent fields for downstream integrations, submitting edge cases and validating response shape could catch the defect that narrative answers cannot be parsed by consuming workflows; the current release gate must challenge the failure where few-shot prompts require excessive tokens and yield inconsistent formats.

Question 284

A proposed deployment supports this business case: A regulated workload must keep data in permitted locations. Reviewers also warn that prompts or telemetry might leave an explicitly permitted processing geography. Which restriction mitigates that concrete risk?

  1. Prohibit unsupported cross-border processing and connector destinations
  2. Keep enough prior approved artifacts for recovery
  3. Restrict modifications and protect audit storage from tampering
  4. Prevent training with unapproved or expired sensitive data

Explanation

Correct Answer

A:

When prompts or telemetry might leave an explicitly permitted processing geography, the use case a regulated workload must keep data in permitted locations requires a guardrail aligned with the behavior that data-flow architecture identifies processing, storage and telemetry locations.

Incorrect Answers

B:

Where a preprocessing release changes category labels incorrectly, the guard can address a rollback might restore a model but not its required preprocessing artifacts; the exposure that prompts or telemetry might leave an explicitly permitted processing geography remains relevant to the present assessment.

C:

A model is replaced without a traceable change ticket; its guardrail addresses the possibility that audit logs might be altered to obscure who deployed a model; the present architecture must constrain the different threat that prompts or telemetry might leave an explicitly permitted processing geography.

D:

The alternate guard addresses training could ingest data whose use rights expired or were never approved for model retraining consumes periodically revised business records; the present threat, prompts or telemetry might leave an explicitly permitted processing geography, needs a different enforcement boundary.

Question 285

A planned AI solution must support this situation: An agent needs to update a governed ticket status. Its prototype fails because its knowledge source can explain but cannot commit a database change. Which design best fits the constraint?

  1. Use connector actions for authenticated data mutations
  2. Define robust tool descriptions and input schemas for planner selection
  3. Restrict Sales connector fields to approved data classifications
  4. Use a prompt-and-response agent for read-only document drafting

Explanation

Correct Answer

A:

The defect is that its knowledge source can explain but cannot commit a database change; the expected business case is that an agent needs to update a governed ticket status; the design fits because an authorized connector action invokes an operational service.

Incorrect Answers

B:

The alternative addresses it passes customer IDs into supplier inputs by a design in which specific tool metadata helps the planner choose and parameterize actions; it does not supply the missing relationship for an agent needs to update a governed ticket status, where its knowledge source can explain but cannot commit a database change.

C:

There is a valid place for the mechanism that field projection and endpoint authorization minimize exposed data, namely sales copilots should not expose sensitive margin assumptions; it does not remove the specific defect that its knowledge source can explain but cannot commit a database change from this case.

D:

There is a valid place for the mechanism that a response-focused agent composes controlled text from supplied context, namely marketing specialists need tailored descriptions from existing briefings; it does not remove the specific defect that its knowledge source can explain but cannot commit a database change from this case.

Question 286

Operations has not established the cause of this defect: The conversation alone authorizes a permanent business change. The associated setting is: The agent may refund large amounts or deactivate accounts. Which inspection should precede remediation?

  1. Inspect approvals against the approved decision table
  2. Review action audit events for missing approval IDs
  3. Document specific requirements unmet by available platforms
  4. Trace triggering events and idempotency keys for repeated actions

Explanation

Correct Answer

B:

For the situation where the agent may refund large amounts or deactivate accounts, the defect is that the conversation alone authorizes a permanent business change; reviewing action audit events for missing approval IDs helps verify the cause-and-effect path: confirmation and human approval separate advice from execution authority.

Incorrect Answers

A:

The evidence this method produces applies to a workflow where the task follows fixed approval branches with no interpretation needed; it cannot by itself establish why the conversation alone authorizes a permanent business change in the current scenario.

C:

The proposed probe concerns competitive requirements depend on unique multi-model control, where custom engineering trades more control for more operating responsibility; this incident instead requires evidence about confirmation and human approval separate advice from execution authority and the failure that the conversation alone authorizes a permanent business change.

D:

Tracing triggering events and idempotency keys for repeated actions explores a separate connection, where event triggers and tool permissions can automate predefined safe steps within explicit limits; the diagnosed requirement is the agent may refund large amounts or deactivate accounts, and the actual defect is the conversation alone authorizes a permanent business change.

Question 287

A planned AI solution must support this situation: Staff need a contextual workspace assembled around one customer case. Its prototype fails because generic chat loses visibility into related records and actions. Which design best fits the constraint?

  1. Use an AI hub-governed reusable prompt for Power Platform makers
  2. Build a code-first generative page with explicitly bound data
  3. Design clear provenance and confidence cues in generated interfaces
  4. Apply security and privacy controls to app-held AI context

Explanation

Correct Answer

B:

Staff need a contextual workspace assembled around one customer case; its failing behavior is that generic chat loses visibility into related records and actions; applying the underlying design principle—generated interface components can be grounded in governed app data and defined actions—connects the requirement to an implementable control.

Incorrect Answers

A:

In the case where multiple low-code apps need a common extraction prompt, the alternate design depends on centralized reusable AI capabilities promote consistency and controlled updates; it does not correct the target failure, generic chat loses visibility into related records and actions.

C:

There is a valid place for the mechanism that visual provenance separates authoritative records from generated advice, namely users need to distinguish verified fields from AI suggestions; it does not remove the specific defect that generic chat loses visibility into related records and actions from this case.

D:

Server-side authorization and minimized inputs protect data flow; its main purpose is a canvas app passes customer records to a model action, not correction of the reported defect where generic chat loses visibility into related records and actions in the target process.

Question 288

The architect wants evidence before altering the deployment. In a workflow where a tool deletes a record following an ambiguous conversation, logs show deletion but not the requesting principal or approval. What should the team examine first?

  1. Inspect artifacts for training data hashes and evaluation reports
  2. Match release manifest to transformation and model checksums
  3. Correlate conversation request IDs with business API audit entries
  4. Compare backend access decisions for different users
  5. Trace source exception through target state and error queue

Explanation

Correct Answer

C:

Logs show deletion but not the requesting principal or approval; to attribute the defect rather than guess at a fix, correlating conversation request IDs with business API audit entries examines the dependency where end-to-end audit binds user, tool, affected record and authorization event.

Incorrect Answers

A:

Models move to production without an audit of training inputs; examining that other path does not determine whether end-to-end audit binds user, tool, affected record and authorization event, which is what distinguishes the issue in the stated use case.

B:

In the separate scenario where a preprocessing release changes category labels incorrectly, the same investigation could explain restoring code alone does not restore previous predictions; it does not isolate the reported cause in a system where a tool deletes a record following an ambiguous conversation.

D:

In the process where agent looks up records on behalf of employees with different rights, the diagnostic procedure could explain shared admin credentials return restricted cases to ordinary users; this particular incident instead turns on whether end-to-end audit binds user, tool, affected record and authorization event and why logs show deletion but not the requesting principal or approval.

E:

A blocked invoice triggers a hold on related procurement actions; its matching diagnostic check is reasonable for a network fault leaves the two apps inconsistent; the present investigation needs evidence for the different failure where logs show deletion but not the requesting principal or approval.

Question 289

Business leaders reject an activity-only metric for this process: AI assistance reads operational records through Dataverse. They need to detect the issue where an update changes a field mapping and grounded answers become stale. Which performance indicator fits?

  1. Mapped operational fields reconciled across releases
  2. Cross-app exceptions ending in consistent state
  3. AI-related incidents during financial-close windows
  4. Model deployments passing artifact-integrity controls

Explanation

Correct Answer

A:

For the business use where AI assistance reads operational records through Dataverse, the result to quantify is mapped operational fields reconciled across releases; it connects measurement to the mechanism where integration tests protect virtual-entity schema and source semantics, not merely to agent traffic.

Incorrect Answers

B:

The premise of this indicator is failure-injection tests verify retries and compensation between systems; the architecture being evaluated instead serves AI assistance reads operational records through Dataverse, and its success is measured by mapped operational fields reconciled across releases.

C:

Invoice posting agents affect critical close processes; its intended measurement is aI-related incidents during financial-close windows; substituting that result here would not capture whether an update changes a field mapping and grounded answers become stale has been corrected.

D:

Although model deployments passing artifact-integrity controls is observable, the affected process involves AI assistance reads operational records through Dataverse; its acceptance outcome is mapped operational fields reconciled across releases, not the alternative measurement.

Question 290

One enterprise rollout has 2 separately testable requirements. (1) Business use: A regulated workload must keep data in permitted locations. Known defect: An external model call exports sensitive prompts to an unapproved region. (2) Business use: A Copilot Studio agent can read corporate data and use public messaging. Known defect: A permitted connector combination enables uncontrolled disclosure. Which 2 architectural choices directly satisfy these stated cases? Choose TWO.

  1. Map all agent data flows to approved geographic boundaries
  2. Deploy actions with compatible rollback behavior
  3. Parameterize environment-specific connection references and variables
  4. Apply data loss prevention policies to connector combinations

Explanation

Correct Answers

A:

For the independent requirement that a regulated workload must keep data in permitted locations, data-flow architecture identifies processing, storage and telemetry locations; it addresses the defect an external model call exports sensitive prompts to an unapproved region, while a Copilot Studio agent can read corporate data and use public messaging needs its own architecture.

D:

For the independent requirement that a Copilot Studio agent can read corporate data and use public messaging, Power Platform data policies constrain connector use and movement pathways; it addresses the defect a permitted connector combination enables uncontrolled disclosure, while a regulated workload must keep data in permitted locations needs its own architecture.

Incorrect Answers

B:

This design is associated with the different failure where rolling back the agent leaves an incompatible backend action in a workflow where a new action version changes how cases are updated; it does not directly solve the listed conditions: an external model call exports sensitive prompts to an unapproved region; a permitted connector combination enables uncontrolled disclosure.

C:

This design is associated with the different failure where agent writes real customer data into the sandbox in a workflow where a flow moved to production still points to a test endpoint; it does not directly solve the listed conditions: an external model call exports sensitive prompts to an unapproved region; a permitted connector combination enables uncontrolled disclosure.

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