Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 19

Question 180

An investment review asks what measurable result would justify continuing an AI solution where users need to distinguish verified fields from AI suggestions. The original issue is: Staff mistake speculative recommendations for recorded facts. Which indicator should be tracked?

  1. Drafts meeting structure and source-accuracy criteria
  2. Complex exception tasks resolved with supported reasoning traces
  3. Numeric calculations matching the governed reference formula
  4. Compound requests completed with correct tool sequence
  5. User understanding of verified versus suggested content

Explanation

Correct Answer

E:

This scorecard needs an outcome related to users need to distinguish verified fields from AI suggestions; user understanding of verified versus suggested content. That indicator can be checked against the earlier symptom that staff mistake speculative recommendations for recorded facts.

Incorrect Answers

A:

A change in drafts meeting structure and source-accuracy criteria is informative for marketing specialists need tailored descriptions from existing briefings; it does not establish whether the current process achieves user understanding of verified versus suggested content.

B:

Although complex exception tasks resolved with supported reasoning traces is observable, the affected process involves users need to distinguish verified fields from AI suggestions; its acceptance outcome is user understanding of verified versus suggested content, not the alternative measurement.

C:

Although numeric calculations matching the governed reference formula is observable, the affected process involves users need to distinguish verified fields from AI suggestions; its acceptance outcome is user understanding of verified versus suggested content, not the alternative measurement.

D:

That scorecard concerns the activity in which users request a status check and an authorized follow-up action and its result of compound requests completed with correct tool sequence; the reported shortcoming here is staff mistake speculative recommendations for recorded facts.

Question 181

Governance approves the business purpose but requires control of this exposure: Long-lived credentials might grant an agent unnecessary Azure resource privileges. The agent will be used when a backend agent must call an approved Azure service without embedded secrets. Which boundary is most defensible?

  1. Block release if privileged writes bypass approvals
  2. Separate draft and send permissions in service integrations
  3. Rotate compromised keys and enforce least-privilege callers
  4. Scope the identity to only required resource permissions

Explanation

Correct Answer

D:

Because long-lived credentials might grant an agent unnecessary Azure resource privileges, a guardrail must constrain the actual system behavior: managed identity provides credential-less workload authentication where supported; protecting a response alone cannot enforce that boundary.

Incorrect Answers

A:

Another process could need this guardrail because an updated connector might gain a direct privileged posting route; the design being assessed supports a backend agent must call an approved Azure service without embedded secrets and has the explicit exposure that long-lived credentials might grant an agent unnecessary Azure resource privileges.

B:

This restriction is associated with service agents draft customer messages but cannot send without review and addresses the exposure that a service drafting assistant might gain permission to send unapproved notices; it does not specifically protect the target workflow against long-lived credentials might grant an agent unnecessary Azure resource privileges.

C:

Another process could need this guardrail because a leaked endpoint credential might permit inference from an untrusted client; the design being assessed supports a backend agent must call an approved Azure service without embedded secrets and has the explicit exposure that long-lived credentials might grant an agent unnecessary Azure resource privileges.

Question 182

A release must prove that lLM-generated arithmetic changes across otherwise identical prompts will not recur. The application is used in this situation: A page must show a reproducible service-level metric. Which acceptance test most directly checks that failure mode?

  1. Install the agent to a pilot Microsoft 365 audience
  2. Refresh the same records and verify stable calculations
  3. Ask unrelated questions and verify no supplier action is triggered
  4. Search for an external document as permitted and denied users

Explanation

Correct Answer

B:

The historical failure is that lLM-generated arithmetic changes across otherwise identical prompts; for the setting where a page must show a reproducible service-level metric, refreshing the same records and verifying stable calculations verifies the dependency where deterministic application code computes numeric outputs from authorized data.

Incorrect Answers

A:

In a solution where employees want curated assistance without owning a separate LLM runtime, installing the agent to a pilot Microsoft 365 audience could catch the defect that a custom service would duplicate built-in host orchestration; the current release gate must challenge the failure where lLM-generated arithmetic changes across otherwise identical prompts.

C:

Although the exercise might catch it autonomously contacts a supplier without approval, it would not directly validate the mechanism that deterministic application code computes numeric outputs from authorized data in the target implementation.

D:

This check exercises a different system behavior: supported connectors expose external content with discoverability and security metadata; a pass there cannot establish that lLM-generated arithmetic changes across otherwise identical prompts no longer occurs in the required workflow.

Question 183

A solution architecture is being reviewed for this business situation: Data scientists cannot explain why a newer classifier behaves differently. The observed defect is that models move to production without an audit of training inputs. Which architectural decision addresses the underlying requirement?

  1. Register each trained model with dataset and evaluation provenance
  2. Validate custom extraction models by field and document class
  3. Assess robustness under realistic drift and noisy input
  4. Retain rollback snapshots for models and data transformations
  5. Set acceptance thresholds and reproducible baselines
  6. Stage candidate models behind champion-challenger evaluation

Explanation

Correct Answer

A:

Without the intended behavior, models move to production without an audit of training inputs; the proposed architecture establishes the specific design principle that a model registry ties versions to lineage, evaluations and approval for the process where data scientists cannot explain why a newer classifier behaves differently.

Incorrect Answers

B:

For a document model performs well overall but fails on tax identifiers, this other approach relies on the fact that stratified field metrics protect high-impact values; the issue here is models move to production without an audit of training inputs; that mechanism does not establish the required a model registry ties versions to lineage, evaluations and approval.

C:

For another type of architecture, stress tests measure sensitivity to expected input variation is useful; this design must instead support data scientists cannot explain why a newer classifier behaves differently despite the problem where models move to production without an audit of training inputs.

D:

In the case where a preprocessing release changes category labels incorrectly, the alternate design depends on versioned data transformations and model artifacts enable consistent rollback; it does not correct the target failure, models move to production without an audit of training inputs.

E:

The technical behavior of this alternative is that predefined gates weigh accuracy, latency and business risk; the user case being assessed is data scientists cannot explain why a newer classifier behaves differently with a different failure: models move to production without an audit of training inputs.

F:

There is a valid place for the mechanism that candidate comparison evaluates quality, latency and safety under matched load, namely a tuned model claims higher accuracy but adds latency; it does not remove the specific defect that models move to production without an audit of training inputs from this case.

Question 184

A solution architecture is being reviewed for this business situation: Downstream research agents run longer than the interactive request. The observed defect is that orchestrator returns a partial result as though the workflow finished. Which architectural decision addresses the underlying requirement?

  1. Provide a unified approval handoff for AI-identified account risks
  2. Use explicit handoff contracts for asynchronous specialist work
  3. Create a custom-engine agent for bespoke model orchestration
  4. Use a supervisor agent that routes to specialized child agents

Explanation

Correct Answer

B:

The mechanism that task IDs, status transitions, deadlines and durable handoff state support coordination fits the scenario where downstream research agents run longer than the interactive request; it addresses the defect that orchestrator returns a partial result as though the workflow finished.

Incorrect Answers

A:

There is a valid place for the mechanism that cross-app workflow routes the recommendation to the accountable approver, namely service, sales and finance jointly decide high-value concessions; it does not remove the specific defect that orchestrator returns a partial result as though the workflow finished from this case.

C:

When the solution demands its own model selection and hosting policies, custom-engine architecture controls runtime, model calls and integrations; in the stated workflow, however, orchestrator returns a partial result as though the workflow finished; the relevant architectural behavior is that task IDs, status transitions, deadlines and durable handoff state support coordination.

D:

This design would make sense if a request spans procurement, finance and policy expertise, particularly where one general agent conflates expertise and may call the wrong tools; the requested architecture instead needs the dependency that task IDs, status transitions, deadlines and durable handoff state support coordination because orchestrator returns a partial result as though the workflow finished.

Question 185

In a business situation where retraining uses revised labels and a different source snapshot, a previous build showed this defect: Model behavior changes without reproducible training provenance. Which reproducible test belongs in the release gate?

  1. Ask questions whose evidence spans a heading and adjacent paragraph
  2. Reproduce a candidate evaluation from the recorded dataset
  3. Feed outages of different scope and inspect suggested severity
  4. Evaluate difficult queries against annotated source passages

Explanation

Correct Answer

B:

For the required workflow where retraining uses revised labels and a different source snapshot, the test needs to reproduce or prevent the failure where model behavior changes without reproducible training provenance; reproducing a candidate evaluation from the recorded dataset exercises the path where dataset lineage and model versioning support repeatable assessment.

Incorrect Answers

A:

A pass on that verification would speak to retrieved policy fragments lose context when documents are split; it would not show that dataset lineage and model versioning support repeatable assessment, which is needed to avoid the observed defect model behavior changes without reproducible training provenance.

C:

A pass on that verification would speak to a specialized support team uses a nonstandard severity matrix; it would not show that dataset lineage and model versioning support repeatable assessment, which is needed to avoid the observed defect model behavior changes without reproducible training provenance.

D:

In the alternate setting where stakeholders report hallucinations but cannot isolate search errors, the test exercises labeled queries and expected evidence distinguish retrieval from synthesis; even a pass would not show that model behavior changes without reproducible training provenance has been prevented.

Question 186

An investment review asks what measurable result would justify continuing an AI solution where an internal model sometimes returns confidential text verbatim. The original issue is: A client can request arbitrary portions of training examples. Which indicator should be tracked?

  1. Model releases containing unapproved training records
  2. Privileged mutations without verified principal authorization
  3. Unauthorized accesses to model-training datasets
  4. Sensitive output disclosures in model misuse evaluations

Explanation

Correct Answer

D:

This scorecard needs an outcome related to an internal model sometimes returns confidential text verbatim; sensitive output disclosures in model misuse evaluations. That indicator can be checked against the earlier symptom that a client can request arbitrary portions of training examples.

Incorrect Answers

A:

Although model releases containing unapproved training records is observable, the affected process involves an internal model sometimes returns confidential text verbatim; its acceptance outcome is sensitive output disclosures in model misuse evaluations, not the alternative measurement.

B:

The premise of this indicator is server-side access checks enforce real actor entitlements; the architecture being evaluated instead serves an internal model sometimes returns confidential text verbatim, and its success is measured by sensitive output disclosures in model misuse evaluations.

C:

The proposed metric concerns proprietary records are copied to an open test workspace, whereas the failure in this case is that a client can request arbitrary portions of training examples; the indicator needed here is sensitive output disclosures in model misuse evaluations.

Question 187

A planned AI solution must support this situation: A new topic description changes which agent is invoked. Its prototype fails because regression appears only for compound user requests. Which design best fits the constraint?

  1. Connect Foundry agent tracing to Application Insights
  2. Inspect retrieval spans when cited answers omit evidence
  3. Compare traces before and after orchestration configuration changes
  4. Use managed identities for service-to-service agent components

Explanation

Correct Answer

C:

Because trace comparisons reveal shifts in plan selection and step order, this approach corrects the reported breakdown where regression appears only for compound user requests; the affected situation is that a new topic description changes which agent is invoked.

Incorrect Answers

A:

A failing agent shows only a generic HTTP error; the alternate design can address that situation because agent trace spans provide component-level execution evidence through connected telemetry, whereas this question needs a solution to regression appears only for compound user requests.

B:

Surface quality scores appear healthy despite wrong documents; the alternate design can address that situation because retrieval telemetry distinguishes evidence search from answer generation, whereas this question needs a solution to regression appears only for compound user requests.

D:

This design would make sense if a backend agent must call an approved Azure service without embedded secrets, particularly where long-lived API keys leak from configuration and diagnostics; the requested architecture instead needs the dependency that trace comparisons reveal shifts in plan selection and step order because regression appears only for compound user requests.

Question 188

A proposed deployment supports this business case: A research agent repeatedly queries sources on difficult cases. Reviewers also warn that an autonomous run might continue consuming resources after losing usable evidence. Which restriction mitigates that concrete risk?

  1. Separate draft generation from external communication authority
  2. Escalate unresolved tasks with a partial evidence trail
  3. Require confirmation when intent remains ambiguous
  4. Block release when critical evidence is consistently omitted

Explanation

Correct Answer

B:

When an autonomous run might continue consuming resources after losing usable evidence, the use case a research agent repeatedly queries sources on difficult cases requires a guardrail aligned with the behavior that time, step and spend budgets terminate unproductive loops.

Incorrect Answers

A:

A bot only needs to check warranty and prepare a draft; its guardrail addresses the possibility that a drafting tool might send an external supplier message without authority; the present architecture must constrain the different threat that an autonomous run might continue consuming resources after losing usable evidence.

C:

The alternate guard addresses overlapping scheduling intents might invoke a change before confirmation for two topics claim nearly identical scheduling intents; the present threat, an autonomous run might continue consuming resources after losing usable evidence, needs a different enforcement boundary.

D:

For the use case where stakeholders report hallucinations but cannot isolate search errors, the control addresses the threat that a release might proceed despite consistently missing the required evidence passage; the specified agent must be protected against an autonomous run might continue consuming resources after losing usable evidence.

Question 189

A release must prove that automatically generated notices are sent without required review will not recur. The application is used in this situation: Standard capabilities draft customer communications but policy requires approval. Which acceptance test most directly checks that failure mode?

  1. Request a high-value refund and verify approval blocks execution
  2. Add a confidential draft and verify it is excluded for unauthorized users
  3. Attempt publication with and without reviewer sign-off
  4. Update a case and verify an authorized sales workflow sees it
  5. Remove required sources and check that the agent declines

Explanation

Correct Answer

C:

Attempting publication with and without reviewer sign-off challenges the path where a review step preserves accountability without discarding assisted drafting for the business setting standard capabilities draft customer communications but policy requires approval; the specific defect to exclude is that automatically generated notices are sent without required review.

Incorrect Answers

A:

The tested behavior is that confirmation and human approval separate advice from execution authority; the necessary evidence here concerns a review step preserves accountability without discarding assisted drafting and the previously observed automatically generated notices are sent without required review.

B:

A pass on that verification would speak to search spans internal manuals with different owners and permissions; it would not show that a review step preserves accountability without discarding assisted drafting, which is needed to avoid the observed defect automatically generated notices are sent without required review.

D:

For sales and service agents need the same customer and case context, the alternate test is about whether shared authorized records support coordinated business context; it cannot prove the outcome needed after the reported regression where automatically generated notices are sent without required review.

E:

This check exercises a different system behavior: confidence and evidence checks route insufficiently supported cases to humans; a pass there cannot establish that automatically generated notices are sent without required review no longer occurs in the required workflow.

Question 190

The rollout supports this workflow: Service representatives draft responses from authorized knowledge. Leaders want a metric that responds to the reported shortcoming that replies cite generic guidance instead of contract-specific policy. Which measure best reflects success?

  1. Handoffs with complete approved case context
  2. Successful authorized external service actions
  3. Approved drafts requiring no policy correction
  4. Sales task completion under external API impairment

Explanation

Correct Answer

C:

Approved drafts requiring no policy correction; this outcome measure belongs to the workflow where service representatives draft responses from authorized knowledge; a change in it can help detect whether the defect that replies cite generic guidance instead of contract-specific policy is resolved.

Incorrect Answers

A:

For the situation where agents need structured case summaries at handoff, handoffs with complete approved case context is a sensible indicator; the problem that replies cite generic guidance instead of contract-specific policy requires measurement of approved drafts requiring no policy correction.

B:

For the workflow where a Teams user should query and update a supported external service, successful authorized external service actions is an actionable measure; it does not establish whether replies cite generic guidance instead of contract-specific policy in the situation under review.

D:

That scorecard concerns the activity in which external credit service responds slowly during regional outages and its result of sales task completion under external API impairment; the reported shortcoming here is replies cite generic guidance instead of contract-specific policy.

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