Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 99

Question 980

Security and business owners want to keep this capability: Standard capabilities draft customer communications but policy requires approval. They must also prevent the following: A drafting assistant could dispatch customer notices without a reviewer. Which guardrail should the design incorporate?

  1. Set enforceable monetary and scope thresholds outside prompts
  2. Quarantine unclassified or ownerless content pending review
  3. Honor cross-app role permissions and data minimization
  4. Separate draft permission from final dispatch permission

Explanation

Correct Answer

D:

When a drafting assistant could dispatch customer notices without a reviewer, the use case standard capabilities draft customer communications but policy requires approval requires a guardrail aligned with the behavior that a review step preserves accountability without discarding assisted drafting.

Incorrect Answers

A:

The proposed control targets a model could override a monetary limit through conversational instructions, while the given risk scenario is a drafting assistant could dispatch customer notices without a reviewer; these exposures arise in different operational contexts.

B:

For the use case where search spans internal manuals with different owners and permissions, the control addresses the threat that ownerless confidential drafts could become searchable before classification; the specified agent must be protected against a drafting assistant could dispatch customer notices without a reviewer.

C:

The alternate guard addresses cross-application aggregation might expose records outside the user's Dynamics permissions for sales and service agents need the same customer and case context; the present threat, a drafting assistant could dispatch customer notices without a reviewer, needs a different enforcement boundary.

Question 981

A proposed deployment supports this business case: Business units create agents with no central visibility. Reviewers also warn that production agents might have no recorded accountable owner or risk classification. Which restriction mitigates that concrete risk?

  1. Require finance process-owner approval for production changes
  2. Block publication without accountable owner and classification
  3. Reject broad admin consent not required by the workload
  4. Enforce owner sign-off for high-risk grounding collections
  5. Mask sensitive input/output fields and control log access

Explanation

Correct Answer

B:

The business setting is business units create agents with no central visibility, with the exposure that production agents might have no recorded accountable owner or risk classification; the enforceable protection must correspond to the path where inventory metadata records owner, risk tier, tools and lifecycle state.

Incorrect Answers

A:

The alternate guard addresses a finance Copilot prompt might change during a protected close window for invoice posting agents affect critical close processes; the present threat, production agents might have no recorded accountable owner or risk classification, needs a different enforcement boundary.

C:

For the use case where an imported agent can authenticate in dev but not production, the control addresses the threat that production consent might grant a wider tenant scope than necessary; the specified agent must be protected against production agents might have no recorded accountable owner or risk classification.

D:

The alternate guard addresses sensitive ownerless documents might enter production grounding for new documents enter a production source automatically; the present threat, production agents might have no recorded accountable owner or risk classification, needs a different enforcement boundary.

E:

Another process could need this guardrail because tracing might store secrets and personal content without adequate access controls; the design being assessed supports business units create agents with no central visibility and has the explicit exposure that production agents might have no recorded accountable owner or risk classification.

Question 982

An agent rollout is intended for this situation: A Copilot Studio agent needs a reusable interface to governed operations. The known problem is that every agent reimplements the same proprietary tool signatures. Which success measure connects to that specific outcome?

  1. Interoperable tasks completed without custom point-to-point parsing
  2. Authorized tool calls completed without excess privileges
  3. Custom workflow completion against the required execution contract
  4. Cross-application entity consistency rate

Explanation

Correct Answer

B:

Because MCP presents supported tools and resources through a common integration contract, the outcome worth following is authorized tool calls completed without excess privileges; the metric is tied to the reported issue—every agent reimplements the same proprietary tool signatures—for the scenario where a Copilot Studio agent needs a reusable interface to governed operations.

Incorrect Answers

A:

Although interoperable tasks completed without custom point-to-point parsing is observable, the affected process involves a Copilot Studio agent needs a reusable interface to governed operations; its acceptance outcome is authorized tool calls completed without excess privileges, not the alternative measurement.

C:

The alternative tracks custom workflow completion against the required execution contract, arising from the mechanism where custom-engine architecture controls runtime, model calls and integrations; its movement alone says little about the failure that every agent reimplements the same proprietary tool signatures.

D:

That scorecard concerns the activity in which agents and apps need shared definitions of customers and work orders and its result of cross-application entity consistency rate; the reported shortcoming here is every agent reimplements the same proprietary tool signatures.

Question 983

A planned AI solution must support this situation: Contact Center agents sometimes cannot resolve complex billing intent. Its prototype fails because repeated fallback messages trap callers in a loop. Which design best fits the constraint?

  1. Extend service assistance with an approved escalation action
  2. Maintain context when transferring voice to human support
  3. Use a declarative agent for role-tailored Microsoft 365 work
  4. Design graceful agent-to-agent and human escalation on low confidence

Explanation

Correct Answer

D:

Without the intended behavior, repeated fallback messages trap callers in a loop; the proposed architecture establishes the specific design principle that confidence, retry ceilings and escalation routing enable resolution for the process where contact Center agents sometimes cannot resolve complex billing intent.

Incorrect Answers

A:

For another type of architecture, an authorized action creates a governed workflow item from the current case is useful; this design must instead support contact Center agents sometimes cannot resolve complex billing intent despite the problem where repeated fallback messages trap callers in a loop.

B:

Structured handoff passes allowed intent, transcript summary and case context; its main purpose is a voice bot cannot resolve a complex eligibility dispute, not correction of the reported defect where repeated fallback messages trap callers in a loop in the target process.

C:

Declarative agents define instructions, knowledge and actions in the supported host; its main purpose is employees want curated assistance without owning a separate LLM runtime, not correction of the reported defect where repeated fallback messages trap callers in a loop in the target process.

Question 984

An agent-assisted service does not behave as specified: LLM-generated arithmetic changes across otherwise identical prompts. It is used in this situation: A page must show a reproducible service-level metric. Which observable diagnostic procedure is best?

  1. Review agent capability manifest against requested functions
  2. Compare tool inventory with legitimate workflow steps
  3. Inspect target UI changes and automation failure traces
  4. Compare displayed metrics with independently computed results

Explanation

Correct Answer

D:

Where lLM-generated arithmetic changes across otherwise identical prompts, the useful evidence comes from comparing displayed metrics with independently computed results; its focus matches the architectural mechanism that deterministic application code computes numeric outputs from authorized data in the process where a page must show a reproducible service-level metric.

Incorrect Answers

A:

The evidence this method produces applies to a workflow where employees want curated assistance without owning a separate LLM runtime; it cannot by itself establish why lLM-generated arithmetic changes across otherwise identical prompts in the current scenario.

B:

The proposed probe concerns a bot only needs to check warranty and prepare a draft, where limited intent and tool scope prevent unnecessary actions; this incident instead requires evidence about deterministic application code computes numeric outputs from authorized data and the failure that lLM-generated arithmetic changes across otherwise identical prompts.

C:

For the failure that teams propose browser clicks despite an available stable API, inspecting target UI changes and automation failure traces is relevant; this question instead reports that lLM-generated arithmetic changes across otherwise identical prompts, so different evidence is needed.

Question 985

A solution architecture is being reviewed for this business situation: An agent scales from a trial to enterprise usage. The observed defect is that token and action costs exceed the pilot estimate. Which architectural decision addresses the underlying requirement?

  1. Compute ROI from incremental net benefits and full lifecycle costs
  2. Route simple classification requests to a validated inexpensive model
  3. Use staged rollout with feature flags and pilot cohorts
  4. Account for capacity, consumption and monitoring in TCO

Explanation

Correct Answer

D:

An agent scales from a trial to enterprise usage; its failing behavior is that token and action costs exceed the pilot estimate; applying the underlying design principle—workload forecasts include model usage, connectors, storage and observability—connects the requirement to an implementable control.

Incorrect Answers

A:

Net benefit compares realized outcomes with implementation and recurring costs; its main purpose is a pilot reports gross labor savings but omits operating expenses, not correction of the reported defect where token and action costs exceed the pilot estimate in the target process.

B:

In the case where many calls require only a small fixed-label output, the alternate design depends on task-aware routing chooses economical models after quality gates; it does not correct the target failure, token and action costs exceed the pilot estimate.

C:

The technical behavior of this alternative is that controlled release gates restrict blast radius and support rollback; the user case being assessed is an agent scales from a trial to enterprise usage with a different failure: token and action costs exceed the pilot estimate.

Question 986

One enterprise rollout has 2 separately testable requirements. (1) Business use: Users report stale answers after a policy refresh. Known defect: Feedback tickets lack the evidence needed to reproduce them. (2) Business use: Fixes are marked complete when a prompt is deployed. Known defect: Previously reported errors recur in the next release. Which 2 regression tests directly satisfy these stated cases? Choose TWO.

  1. Inject a malicious export instruction into a source document
  2. Re-run the original complaint and neighboring cases
  3. Introduce a known distribution shift and verify review triggers
  4. Reproduce a reported answer from recorded non-sensitive context

Explanation

Correct Answers

B:

For the business workflow where fixes are marked complete when a prompt is deployed, re-runing the original complaint and neighboring cases challenges the failure that previously reported errors recur in the next release; the release also needs a separate check for users report stale answers after a policy refresh.

D:

For the business workflow where users report stale answers after a policy refresh, reproducing a reported answer from recorded non-sensitive context challenges the failure that feedback tickets lack the evidence needed to reproduce them; the release also needs a separate check for fixes are marked complete when a prompt is deployed.

Incorrect Answers

A:

This regression test is associated with the different failure where model-level refusal alone may not prevent a tool call in a workflow where a prompt injection requests data export to an outside address; it does not directly solve the listed conditions: feedback tickets lack the evidence needed to reproduce them; previously reported errors recur in the next release.

C:

This regression test is associated with the different failure where team retrains monthly despite unchanged model behavior in a workflow where customer terminology changes over time; it does not directly solve the listed conditions: feedback tickets lack the evidence needed to reproduce them; previously reported errors recur in the next release.

Question 987

Before introducing a fix, the team must explain why an update changes a field mapping and grounded answers become stale. The affected setting is: AI assistance reads operational records through Dataverse. What should be inspected?

  1. Review virtual-entity metadata and source-field changes
  2. Audit dataset readers and movement into training jobs
  3. Review endpoint access logs and allowed network paths
  4. Trace source exception through target state and error queue

Explanation

Correct Answer

A:

For the case where AI assistance reads operational records through Dataverse, the diagnostic path must explain the failure that an update changes a field mapping and grounded answers become stale; reviewing virtual-entity metadata and source-field changes targets the mechanism by which integration tests protect virtual-entity schema and source semantics.

Incorrect Answers

B:

The evidence this method produces applies to a workflow where proprietary records are copied to an open test workspace; it cannot by itself establish why an update changes a field mapping and grounded answers become stale in the current scenario.

C:

That probe evaluates identity-based access and supported network isolation reduce exposure in a workflow where a sensitive model endpoint is reachable using a broadly shared key; the reported failure concerns the path for which integration tests protect virtual-entity schema and source semantics instead.

D:

For the failure that a network fault leaves the two apps inconsistent, tracing source exception through target state and error queue is relevant; this question instead reports that an update changes a field mapping and grounded answers become stale, so different evidence is needed.

Question 988

A product owner must measure whether an AI-enabled change actually improves this process: A helpful assistant can still produce unsafe financial or legal guidance. The earlier problem was: Pilot satisfaction hides severe low-frequency failures. Which outcome indicator is most relevant?

  1. Root-cause clusters converted to verified corrections
  2. Severe unsafe outputs in adversarial review
  3. High-severity agent defects remediated within policy target
  4. High-impact decisions with substantive human review

Explanation

Correct Answer

B:

The use case involves a helpful assistant can still produce unsafe financial or legal guidance and the baseline failure that pilot satisfaction hides severe low-frequency failures; the KPI tracks severe unsafe outputs in adversarial review, so improvement has an interpretable operational meaning.

Incorrect Answers

A:

That scorecard concerns the activity in which thousands of transcripts contain recurring misunderstandings and its result of root-cause clusters converted to verified corrections; the reported shortcoming here is pilot satisfaction hides severe low-frequency failures.

C:

The proposed metric concerns product owners favor easy fixes over high-consequence failures, whereas the failure in this case is that pilot satisfaction hides severe low-frequency failures; the indicator needed here is severe unsafe outputs in adversarial review.

D:

The proposed metric concerns AI proposes eligibility decisions affecting customers, whereas the failure in this case is that pilot satisfaction hides severe low-frequency failures; the indicator needed here is severe unsafe outputs in adversarial review.

Question 989

The rollout supports this workflow: Copilot, Dataverse, Dynamics and custom APIs share one process. Leaders want a metric that responds to the reported shortcoming that failures cannot be attributed to any single integration owner. Which measure best reflects success?

  1. Unsupported regulated answers reaching users
  2. Invalid tool-call requests rejected before mutation
  3. Error-rate delta between canary and control cohorts
  4. Share of failed workflows with an identified responsible component

Explanation

Correct Answer

D:

The use case involves copilot, Dataverse, Dynamics and custom APIs share one process and the baseline failure that failures cannot be attributed to any single integration owner; the KPI tracks share of failed workflows with an identified responsible component, so improvement has an interpretable operational meaning.

Incorrect Answers

A:

For the workflow where the system cannot always find evidence for regulated guidance, unsupported regulated answers reaching users is an actionable measure; it does not establish whether failures cannot be attributed to any single integration owner in the situation under review.

B:

Although invalid tool-call requests rejected before mutation is observable, the affected process involves copilot, Dataverse, Dynamics and custom APIs share one process; its acceptance outcome is share of failed workflows with an identified responsible component, not the alternative measurement.

C:

Error-rate delta between canary and control cohorts; that indicator describes a distinct workflow where new behavior must be evaluated without exposing the full workforce; it does not quantify share of failed workflows with an identified responsible component, which is needed for the stated case.

Question 990

A quality engineer is writing a regression case after the incident where it passes customer IDs into supplier inputs. The underlying business scenario is: The orchestrator chooses between similar invoice and credit tools. What should the test exercise?

  1. Test calls with background noise and varied accents
  2. Test ambiguous requests with incomplete required inputs
  3. Evaluate multilingual short texts and domain-specific names
  4. Try the action with a user lacking record access
  5. Request drafts using multiple brief formats

Explanation

Correct Answer

B:

It passes customer IDs into supplier inputs; a release test must establish that this no longer occurs; testing ambiguous requests with incomplete required inputs measures the specific behavior for which specific tool metadata helps the planner choose and parameterize actions.

Incorrect Answers

A:

In a solution where quality teams need searchable insights from recorded calls, testing calls with background noise and varied accents could catch the defect that manual notes omit important customer statements; the current release gate must challenge the failure where it passes customer IDs into supplier inputs.

C:

A pass on that verification would speak to the system must detect customer sentiment and named entities at scale; it would not show that specific tool metadata helps the planner choose and parameterize actions, which is needed to avoid the observed defect it passes customer IDs into supplier inputs.

D:

The verification concerning a canvas app passes customer records to a model action checks server-side authorization and minimized inputs protect data flow against hidden fields from the UI appear in the prompt payload; its result says little about the observed issue where it passes customer IDs into supplier inputs.

E:

This check exercises a different system behavior: a response-focused agent composes controlled text from supplied context; a pass there cannot establish that it passes customer IDs into supplier inputs no longer occurs in the required workflow.

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