Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 10

Question 90

A test plan focuses on a previously missed failure: A general large model exceeds per-document unit economics. It occurred when high-volume narrow extraction needs predictable latency and cost. Which check directly validates that behavior?

  1. Test the custom action with existing identity and audit policy
  2. Recalculate the case under higher review time and lower adoption
  3. Demonstrate a custom multi-step workflow and failure recovery
  4. Simulate endpoint failure and verify continuity without policy bypass
  5. Run a representative extraction holdout at production concurrency

Explanation

Correct Answer

E:

A general large model exceeds per-document unit economics; a release test must establish that this no longer occurs; running a representative extraction holdout at production concurrency measures the specific behavior for which smaller models can be evaluated for narrow tasks with lower cost.

Incorrect Answers

A:

In the alternate setting where current users need one governed custom action inside their workflow, the test exercises extension retains familiar user experience and adds bounded custom behavior; even a pass would not show that a general large model exceeds per-document unit economics has been prevented.

B:

This check exercises a different system behavior: net benefit compares realized outcomes with implementation and recurring costs; a pass there cannot establish that a general large model exceeds per-document unit economics no longer occurs in the required workflow.

C:

A pass on that verification would speak to the solution demands its own model selection and hosting policies; it would not show that smaller models can be evaluated for narrow tasks with lower cost, which is needed to avoid the observed defect a general large model exceeds per-document unit economics.

D:

The verification concerning peak load can saturate a primary model endpoint checks health-aware fallback uses a compatible approved model when needed against requests fail despite another approved deployment being available; its result says little about the observed issue where a general large model exceeds per-document unit economics.

Question 91

A test plan focuses on a previously missed failure: Long-lived API keys leak from configuration and diagnostics. It occurred when a backend agent must call an approved Azure service without embedded secrets. Which check directly validates that behavior?

  1. Try to send high-impact notices without approval in staging
  2. Attempt inference from an unauthorized identity or network
  3. Rotate away from a test secret and verify identity-based access
  4. Test extraction and secret-disclosure prompts with synthetic fixtures

Explanation

Correct Answer

C:

Rotating away from a test secret and verifying identity-based access is aligned with the failure that long-lived API keys leak from configuration and diagnostics; its causal relevance is that managed identity provides credential-less workload authentication where supported for the situation where a backend agent must call an approved Azure service without embedded secrets.

Incorrect Answers

A:

For service agents draft customer messages but cannot send without review, the alternate test is about whether business approval checks remain independent of agent version; it cannot prove the outcome needed after the reported regression where long-lived API keys leak from configuration and diagnostics.

B:

This check exercises a different system behavior: identity-based access and supported network isolation reduce exposure; a pass there cannot establish that long-lived API keys leak from configuration and diagnostics no longer occurs in the required workflow.

D:

The suggested test serves an internal model sometimes returns confidential text verbatim, in which a client can request arbitrary portions of training examples; the present scenario instead concerns the reported problem where long-lived API keys leak from configuration and diagnostics.

Question 92

Business leaders reject an activity-only metric for this process: Solution import succeeds but users still see the previous behavior. They need to detect the issue where deployment checks stop at package import success. Which performance indicator fits?

  1. Promoted releases passing channel-level smoke tests
  2. Production calls routed to approved production dependencies
  3. Unauthorized cross-environment requests blocked
  4. Customer handoffs free from integration mismatches

Explanation

Correct Answer

A:

Because post-import validation includes publishing and channel behavior, the outcome worth following is promoted releases passing channel-level smoke tests; the metric is tied to the reported issue—deployment checks stop at package import success—for the scenario where solution import succeeds but users still see the previous behavior.

Incorrect Answers

B:

For the situation where a flow moved to production still points to a test endpoint, production calls routed to approved production dependencies is a sensible indicator; the problem that deployment checks stop at package import success requires measurement of promoted releases passing channel-level smoke tests.

C:

A change in unauthorized cross-environment requests blocked is informative for test automation must not reach live customer records; it does not establish whether the current process achieves promoted releases passing channel-level smoke tests.

D:

Although customer handoffs free from integration mismatches is observable, the affected process involves solution import succeeds but users still see the previous behavior; its acceptance outcome is promoted releases passing channel-level smoke tests, not the alternative measurement.

Question 93

Support engineers report this failure: Bot attempts protected changes based only on spoken order numbers. The business context is: Anonymous callers can ask generic questions but not change orders. What evidence should be collected to isolate the cause?

  1. Capture response schemas and returned sensitive columns
  2. Review agent capability manifest against requested functions
  3. Inspect action traces for user identity and authorization results
  4. Compare summaries with mandated case-handoff sections

Explanation

Correct Answer

C:

Bot attempts protected changes based only on spoken order numbers; to attribute the defect rather than guess at a fix, inspecting action traces for user identity and authorization results examines the dependency where authentication and authorization gates separate general help from account control.

Incorrect Answers

A:

The evidence this method produces applies to a workflow where sales copilots should not expose sensitive margin assumptions; it cannot by itself establish why bot attempts protected changes based only on spoken order numbers in the current scenario.

B:

In the process where employees want curated assistance without owning a separate LLM runtime, the diagnostic procedure could explain a custom service would duplicate built-in host orchestration; this particular incident instead turns on whether authentication and authorization gates separate general help from account control and why bot attempts protected changes based only on spoken order numbers.

D:

Comparing summaries with mandated case-handoff sections explores a separate connection, where tailored instructions and contextual fields standardize case handoff content; the diagnosed requirement is anonymous callers can ask generic questions but not change orders, and the actual defect is bot attempts protected changes based only on spoken order numbers.

Question 94

Before production promotion, the team needs evidence for this setting: Independently hosted agents must exchange tasks and results. A prior regression caused the following: Cross-team agents rely on hard-coded, vendor-specific chat formats. Which validation is most persuasive?

  1. Deliver duplicate events and confirm single reversible updates
  2. Verify task handoff and result exchange across two supported runtimes
  3. Pilot the standard agent against a representative task set
  4. Submit paraphrases and multi-intent requests and verify correct tool selection

Explanation

Correct Answer

B:

Verifying task handoff and result exchange across two supported runtimes challenges the path where an agent interoperability protocol supports explicit task delegation between agents for the business setting independently hosted agents must exchange tasks and results; the specific defect to exclude is that cross-team agents rely on hard-coded, vendor-specific chat formats.

Incorrect Answers

A:

This check exercises a different system behavior: event triggers and tool permissions can automate predefined safe steps within explicit limits; a pass there cannot establish that cross-team agents rely on hard-coded, vendor-specific chat formats no longer occurs in the required workflow.

C:

The suggested test serves users need ordinary meeting summaries and follow-up drafts, in which a costly custom runtime replicates a mature standard capability; the present scenario instead concerns the reported problem where cross-team agents rely on hard-coded, vendor-specific chat formats.

D:

In the alternate setting where employees express several intents and need interactive clarification, the test exercises generative orchestration interprets intent and can combine authorized tools and knowledge; even a pass would not show that cross-team agents rely on hard-coded, vendor-specific chat formats has been prevented.

Question 95

The pilot needs a durable correction: Successful chat sessions conceal repeated backend writes. This affects the workflow in which agents modify records through external APIs. Which design choice should the architect approve?

  1. Measure tool success, retries and side-effect duplication
  2. Test concurrent agent runs for duplicate side effects
  3. Measure user friction and escalation separately by channel
  4. Test tool-permission escalation attempts at the execution layer

Explanation

Correct Answer

A:

The defect is that successful chat sessions conceal repeated backend writes; the expected business case is that agents modify records through external APIs; the design fits because tool metrics reveal mutation failures and retry consequences.

Incorrect Answers

B:

There is a valid place for the mechanism that concurrency evaluation tests idempotency and serialization, namely several events refer to one business order at once; it does not remove the specific defect that successful chat sessions conceal repeated backend writes from this case.

C:

Combined metrics hide voice users trapped in fallback loops; the alternate design can address that situation because channel-specific analytics reveal where interactions fail, whereas this question needs a solution to successful chat sessions conceal repeated backend writes.

D:

Authorization evaluations verify server-side deny behavior; its main purpose is a user asks the agent to read an unrelated executive record, not correction of the reported defect where successful chat sessions conceal repeated backend writes in the target process.

Question 96

During integration testing, mean response time looks acceptable. The associated business context is: A small portion of calls time out during peak demand. Which architectural remedy is the strongest match?

  1. Analyze percentile latency instead of averages alone
  2. Tune descriptions and tool constraints before adding more tools
  3. Stage candidate models behind champion-challenger evaluation
  4. Track throughput and throttling to plan capacity

Explanation

Correct Answer

A:

The defect is that mean response time looks acceptable; the expected business case is that a small portion of calls time out during peak demand; the design fits because tail percentiles surface extreme experiences affecting users.

Incorrect Answers

B:

For another type of architecture, clear capabilities and restricted sets improve selection reliability is useful; this design must instead support a small portion of calls time out during peak demand despite the problem where mean response time looks acceptable.

C:

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 mean response time looks acceptable from this case.

D:

For another type of architecture, capacity metrics and throttling errors show saturation risk is useful; this design must instead support a small portion of calls time out during peak demand despite the problem where mean response time looks acceptable.

Question 97

The architect wants evidence before altering the deployment. In a workflow where finance users need current operational data in contextual help, a static knowledge upload gives stale purchase-order states. What should the team examine first?

  1. Inspect plugin capabilities and authentication configuration
  2. Verify virtual entity enablement and row-level access
  3. Inspect outbound payloads and role-based access decisions
  4. Review agent capability manifest against requested functions

Explanation

Correct Answer

B:

For the case where finance users need current operational data in contextual help, the diagnostic path must explain the failure that a static knowledge upload gives stale purchase-order states; verifying virtual entity enablement and row-level access targets the mechanism by which authorized virtual entities can surface current finance and operations data to supported agents.

Incorrect Answers

A:

The proposed probe concerns a Teams user should query and update a supported external service, where a plugin provides authorized callable tools to an agent; this incident instead requires evidence about authorized virtual entities can surface current finance and operations data to supported agents and the failure that a static knowledge upload gives stale purchase-order states.

C:

The proposed inspection is associated with a canvas app passes customer records to a model action, where server-side authorization and minimized inputs protect data flow; it would not expose the failure mode that a static knowledge upload gives stale purchase-order states with the required specificity.

D:

Reviewing agent capability manifest against requested functions explores a separate connection, where declarative agents define instructions, knowledge and actions in the supported host; the diagnosed requirement is finance users need current operational data in contextual help, and the actual defect is a static knowledge upload gives stale purchase-order states.

Question 98

Implementation alternatives must be compared for a workflow where marketing specialists need tailored descriptions from existing briefings. The pilot shows this gap: A broad action agent can modify records unnecessarily. Which alternative should lead the solution?

  1. Use a prompt-and-response agent for read-only document drafting
  2. Control creativity for regulated response templates
  3. Choose bounded task-specific instructions over open-ended autonomy
  4. Define robust tool descriptions and input schemas for planner selection

Explanation

Correct Answer

A:

For the need that marketing specialists need tailored descriptions from existing briefings, the relevant mechanism is that a response-focused agent composes controlled text from supplied context; that architectural dependency addresses the observed failure where a broad action agent can modify records unnecessarily.

Incorrect Answers

B:

When compliance letters must preserve approved language and disclaimers, templating and approved clauses limit generated variation; in the stated workflow, however, a broad action agent can modify records unnecessarily; the relevant architectural behavior is that a response-focused agent composes controlled text from supplied context.

C:

Limited intent and tool scope prevent unnecessary actions; its main purpose is a bot only needs to check warranty and prepare a draft, not correction of the reported defect where a broad action agent can modify records unnecessarily in the target process.

D:

Specific tool metadata helps the planner choose and parameterize actions; its main purpose is the orchestrator chooses between similar invoice and credit tools, not correction of the reported defect where a broad action agent can modify records unnecessarily in the target process.

Question 99

A business stakeholder describes this operating situation: Thousands of transcripts contain recurring misunderstandings. Testing instead shows that analysts overlook a widespread intent failure among unique phrasings. What design approach is most appropriate?

  1. Use identity-aware caching for grounded agent responses
  2. Test audit completeness for AI-assisted customer concessions
  3. Use grounded answer evaluation to separate retrieval and generation faults
  4. Test Finance-to-Supply-Chain exception handling after partial failure
  5. Cluster similar agent failures using embeddings or AI-assisted analysis

Explanation

Correct Answer

E:

For the need that thousands of transcripts contain recurring misunderstandings, the relevant mechanism is that semantic clustering groups related failure patterns for review; that architectural dependency addresses the observed failure where analysts overlook a widespread intent failure among unique phrasings.

Incorrect Answers

A:

For another type of architecture, cache keys and access validation prevent security-context leakage is useful; this design must instead support thousands of transcripts contain recurring misunderstandings despite the problem where analysts overlook a widespread intent failure among unique phrasings.

B:

For service recommends a price adjustment requiring finance approval, this other approach relies on the fact that transaction-path tests confirm recommendation, review and posting records; the issue here is analysts overlook a widespread intent failure among unique phrasings; that mechanism does not establish the required semantic clustering groups related failure patterns for review.

C:

The technical behavior of this alternative is that separate relevance and groundedness tests isolate the failing layer; the user case being assessed is thousands of transcripts contain recurring misunderstandings with a different failure: analysts overlook a widespread intent failure among unique phrasings.

D:

When a blocked invoice triggers a hold on related procurement actions, failure-injection tests verify retries and compensation between systems; in the stated workflow, however, analysts overlook a widespread intent failure among unique phrasings; the relevant architectural behavior is that semantic clustering groups related failure patterns for review.

Question 100

The pilot needs a durable correction: A high aggregate score masks critical field errors. This affects the workflow in which a document model performs well overall but fails on tax identifiers. Which design choice should the architect approve?

  1. Assess robustness under realistic drift and noisy input
  2. Preserve model-compatible preprocessing and postprocessing contracts
  3. Preserve reproducible evaluation evidence for major model changes
  4. Validate custom extraction models by field and document class

Explanation

Correct Answer

D:

Without the intended behavior, a high aggregate score masks critical field errors; the proposed architecture establishes the specific design principle that stratified field metrics protect high-impact values for the process where a document model performs well overall but fails on tax identifiers.

Incorrect Answers

A:

There is a valid place for the mechanism that stress tests measure sensitivity to expected input variation, namely production documents differ from clean training samples; it does not remove the specific defect that a high aggregate score masks critical field errors from this case.

B:

This design would make sense if a retraining release changes the tokenizer and label order, particularly where downstream apps interpret correct predictions as the wrong category; the requested architecture instead needs the dependency that stratified field metrics protect high-impact values because a high aggregate score masks critical field errors.

C:

This design would make sense if an updated agent performs differently months after release, particularly where old scores cannot be reproduced because prompts and tests changed; the requested architecture instead needs the dependency that stratified field metrics protect high-impact values because a high aggregate score masks critical field errors.

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