Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 84

Question 830

Security and business owners want to keep this capability: Deployment automation pulls model artifacts from unverified origins. They must also prevent the following: An attacker might swap a model artifact in the deployment pipeline. Which guardrail should the design incorporate?

  1. Keep enough prior approved artifacts for recovery
  2. Restrict production to trusted vetted model versions
  3. Restrict modifications and protect audit storage from tampering
  4. Reject unregistered or unapproved models from production

Explanation

Correct Answer

B:

Avoiding the consequence where an attacker might swap a model artifact in the deployment pipeline requires a control for deployment automation pulls model artifacts from unverified origins; in this architecture, artifact provenance and integrity checks guard model supply chain, which is the point at which the boundary can be applied.

Incorrect Answers

A:

The alternate guard addresses a rollback might restore a model but not its required preprocessing artifacts for a preprocessing release changes category labels incorrectly; the present threat, an attacker might swap a model artifact in the deployment pipeline, needs a different enforcement boundary.

C:

The alternate boundary mitigates audit logs might be altered to obscure who deployed a model, which is relevant to a model is replaced without a traceable change ticket; it leaves the current workflow vulnerable to an attacker might swap a model artifact in the deployment pipeline.

D:

This restriction is associated with data scientists cannot explain why a newer classifier behaves differently and addresses the exposure that a model outside the approved registry might reach production; it does not specifically protect the target workflow against an attacker might swap a model artifact in the deployment pipeline.

Question 831

A staging test has to detect the specific defect that the on-call team cannot explain failed tool calls. In the workflow where a newly deployed Foundry agent has no detailed diagnostics, which exercise verifies the corrective behavior?

  1. Invoke an agent after deployment and locate its trace
  2. Submit urgent and ambiguous cases to the candidate agent
  3. Test a conversation that ends politely without resolving the case
  4. Inject timeout after commit and verify a single write

Explanation

Correct Answer

A:

The scenario is that a newly deployed Foundry agent has no detailed diagnostics; the defect to rule out is that the on-call team cannot explain failed tool calls; testing via invoking an agent after deployment and locating its trace challenges the design dependency where connected observability supplies traces and operational metrics.

Incorrect Answers

B:

In a solution where a support agent changes how it classifies priority intents, submitting urgent and ambiguous cases to the candidate agent could catch the defect that high-severity tickets begin routing to general queues; the current release gate must challenge the failure where the on-call team cannot explain failed tool calls.

C:

In a solution where a service bot handles many chats but few cases are resolved, testing a conversation that ends politely without resolving the case could catch the defect that a dashboard calls the bot successful because messages increased; the current release gate must challenge the failure where the on-call team cannot explain failed tool calls.

D:

Although the exercise might catch successful chat sessions conceal repeated backend writes, it would not directly validate the mechanism that connected observability supplies traces and operational metrics in the target implementation.

Question 832

A planned AI solution must support this situation: A regional launch multiplies concurrent agent requests. Its prototype fails because rate limiting causes intermittent failures. Which design best fits the constraint?

  1. Track throughput and throttling to plan capacity
  2. Measure user friction and escalation separately by channel
  3. Use cost-versus-quality dashboards to evaluate model changes
  4. Tune model routing based on segmented evaluation outcomes

Explanation

Correct Answer

A:

The architecture must reconcile two facts: a regional launch multiplies concurrent agent requests, yet rate limiting causes intermittent failures; the selected pattern does so through the mechanism that capacity metrics and throttling errors show saturation risk.

Incorrect Answers

B:

For chat and voice agents have different interaction patterns, this other approach relies on the fact that channel-specific analytics reveal where interactions fail; the issue here is rate limiting causes intermittent failures; that mechanism does not establish the required capacity metrics and throttling errors show saturation risk.

C:

When a new model reduces price but also degrades exact extractions, joint quality and consumption measures capture real efficiency; in the stated workflow, however, rate limiting causes intermittent failures; the relevant architectural behavior is that capacity metrics and throttling errors show saturation risk.

D:

Lower-cost routing degrades a business-critical subgroup; the alternate design can address that situation because stratified evaluation guides thresholds for difficult requests, whereas this question needs a solution to rate limiting causes intermittent failures.

Question 833

A solution architecture is being reviewed for this business situation: Procurement exceptions require comparing several policies and records. The observed defect is that a simple scripted topic cannot handle legitimate variants. Which architectural decision addresses the underlying requirement?

  1. Use translation capabilities before downstream review
  2. Use MCP tools for a governed reusable integration surface
  3. Use agent reasoning for uncertain multi-step analysis with bounded tools
  4. Add a plugin action to a supported declarative agent

Explanation

Correct Answer

C:

Procurement exceptions require comparing several policies and records; its failing behavior is that a simple scripted topic cannot handle legitimate variants; applying the underlying design principle—reasoning can plan through evidence and tools within enforced boundaries—connects the requirement to an implementable control.

Incorrect Answers

A:

When support documents arrive in several customer languages, translation provides standardized text input while preserving original evidence; in the stated workflow, however, a simple scripted topic cannot handle legitimate variants; the relevant architectural behavior is that reasoning can plan through evidence and tools within enforced boundaries.

B:

A supported MCP server exposes typed tool capabilities to clients; its main purpose is several agents need the same inventory query operation, not correction of the reported defect where a simple scripted topic cannot handle legitimate variants in the target process.

D:

In the case where a Teams user should query and update a supported external service, the alternate design depends on a plugin provides authorized callable tools to an agent; it does not correct the target failure, a simple scripted topic cannot handle legitimate variants.

Question 834

A policy review covers an agent for which thousands of transcripts contain recurring misunderstandings. Its most relevant exposure is that a semantic clustering exercise might expose transcript data or invent misleading themes. What safeguard should be required before release?

  1. Forbid direct mutation before recorded delegated approval
  2. Require reviewed datasets and independent promotion sign-off
  3. Do not approve procurement while financial hold status is unknown
  4. Use redacted samples and check for misleading cluster labels

Explanation

Correct Answer

D:

In the scenario where thousands of transcripts contain recurring misunderstandings, the explicit risk is that a semantic clustering exercise might expose transcript data or invent misleading themes; enforcement must follow the mechanism by which semantic clustering groups related failure patterns for review.

Incorrect Answers

A:

The alternate boundary mitigates a suggested customer concession might post before delegated finance approval, which is relevant to service recommends a price adjustment requiring finance approval; it leaves the current workflow vulnerable to a semantic clustering exercise might expose transcript data or invent misleading themes.

B:

Business analysts can submit examples but not change production behavior; its guardrail addresses the possibility that unreviewed training examples might poison a promoted model; the present architecture must constrain the different threat that a semantic clustering exercise might expose transcript data or invent misleading themes.

C:

Another process could need this guardrail because a procurement approval might proceed before a finance hold is resolved; the design being assessed supports thousands of transcripts contain recurring misunderstandings and has the explicit exposure that a semantic clustering exercise might expose transcript data or invent misleading themes.

Question 835

After deployment, stakeholders need an interpretable measure for this use case: A new model reduces price but also degrades exact extractions. The pilot showed that finance sees savings without the downstream rework cost. Which metric captures the needed result?

  1. Ninety-fifth percentile end-to-end response latency
  2. Contract-valid predictions after custom-model updates
  3. Total processing cost per correct verified extraction
  4. Plan selection agreement on a frozen regression set

Explanation

Correct Answer

C:

Total processing cost per correct verified extraction; this outcome measure belongs to the workflow where a new model reduces price but also degrades exact extractions; a change in it can help detect whether the defect that finance sees savings without the downstream rework cost is resolved.

Incorrect Answers

A:

The proposed metric concerns a small portion of calls time out during peak demand, whereas the failure in this case is that finance sees savings without the downstream rework cost; the indicator needed here is total processing cost per correct verified extraction.

B:

Contract-valid predictions after custom-model updates; that indicator describes a distinct workflow where a retraining release changes the tokenizer and label order; it does not quantify total processing cost per correct verified extraction, which is needed for the stated case.

D:

That scorecard concerns the activity in which a new topic description changes which agent is invoked and its result of plan selection agreement on a frozen regression set; the reported shortcoming here is finance sees savings without the downstream rework cost.

Question 836

The architect wants evidence before altering the deployment. In a workflow where retraining uses revised labels and a different source snapshot, model behavior changes without reproducible training provenance. What should the team examine first?

  1. Record recall at K and citation-grounding errors separately
  2. Trace each release to labels, parameters and dataset hash
  3. Inspect chunk boundaries and parent document links
  4. Review agent capability manifest against requested functions

Explanation

Correct Answer

B:

In the setting where retraining uses revised labels and a different source snapshot, tracing each release to labels, parameters and dataset hash can isolate the failure in which model behavior changes without reproducible training provenance, because dataset lineage and model versioning support repeatable assessment.

Incorrect Answers

A:

In the process where stakeholders report hallucinations but cannot isolate search errors, the diagnostic procedure could explain generator tuning fails because the required documents were never retrieved; this particular incident instead turns on whether dataset lineage and model versioning support repeatable assessment and why model behavior changes without reproducible training provenance.

C:

In the separate scenario where retrieved policy fragments lose context when documents are split, the same investigation could explain answers cite isolated sentences without section qualifiers; it does not isolate the reported cause in a system where retraining uses revised labels and a different source snapshot.

D:

Employees want curated assistance without owning a separate LLM runtime; its matching diagnostic check is reasonable for a custom service would duplicate built-in host orchestration; the present investigation needs evidence for the different failure where model behavior changes without reproducible training provenance.

Question 837

A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: An updated agent performs differently months after release. Known defect: Old scores cannot be reproduced because prompts and tests changed. Which KPI should be selected?

  1. Candidate releases satisfying every critical acceptance metric
  2. Model changes with complete actor and approval records
  3. Critical releases with reproducible evaluation and sign-off records
  4. Retrieval regressions following index promotions
  5. True performance drift detected before critical SLO breach

Explanation

Correct Answer

C:

The use case involves an updated agent performs differently months after release and the baseline failure that old scores cannot be reproduced because prompts and tests changed; the KPI tracks critical releases with reproducible evaluation and sign-off records, so improvement has an interpretable operational meaning.

Incorrect Answers

A:

The premise of this indicator is predefined gates weigh accuracy, latency and business risk; the architecture being evaluated instead serves an updated agent performs differently months after release, and its success is measured by critical releases with reproducible evaluation and sign-off records.

B:

The alternative tracks model changes with complete actor and approval records, arising from the mechanism where deployment audit links actor, approved artifact and release time; its movement alone says little about the failure that old scores cannot be reproduced because prompts and tests changed.

D:

For the workflow where a search index changes embedding model and chunking approach, retrieval regressions following index promotions is an actionable measure; it does not establish whether old scores cannot be reproduced because prompts and tests changed in the situation under review.

E:

The premise of this indicator is drift monitoring and outcomes justify retraining when performance shifts; the architecture being evaluated instead serves an updated agent performs differently months after release, and its success is measured by critical releases with reproducible evaluation and sign-off records.

Question 838

A product owner must measure whether an AI-enabled change actually improves this process: Sales and finance agents must agree on revenue and margin calculations. The earlier problem was: Separate assistants report contradictory totals for the same period. Which outcome indicator is most relevant?

  1. Percentage of agent metrics matching reconciled source calculations
  2. Percentage of agent releases passing common governance gates
  3. Share of grounding records with complete governance metadata
  4. Proportion of tasks assigned to an explicit execution pattern

Explanation

Correct Answer

A:

Percentage of agent metrics matching reconciled source calculations; this outcome measure belongs to the workflow where sales and finance agents must agree on revenue and margin calculations; a change in it can help detect whether the defect that separate assistants report contradictory totals for the same period is resolved.

Incorrect Answers

B:

Independent teams choose inconsistent evaluation and security methods; its intended measurement is percentage of agent releases passing common governance gates; substituting that result here would not capture whether separate assistants report contradictory totals for the same period has been corrected.

C:

That scorecard concerns the activity in which search spans internal manuals with different owners and permissions and its result of share of grounding records with complete governance metadata; the reported shortcoming here is separate assistants report contradictory totals for the same period.

D:

That scorecard concerns the activity in which a business process mixes deterministic calculations and open-ended judgments and its result of proportion of tasks assigned to an explicit execution pattern; the reported shortcoming here is separate assistants report contradictory totals for the same period.

Question 839

A solution architecture is being reviewed for this business situation: A Copilot prompt feeds records into downstream automation. The observed defect is that generated output occasionally omits required keys. Which architectural decision addresses the underlying requirement?

  1. Test structured outputs against explicit format contracts
  2. Validate custom extraction models by field and document class
  3. Use delegated user identity for user-scoped data actions
  4. Test prompt changes on paraphrases and edge conditions

Explanation

Correct Answer

A:

For the need that a Copilot prompt feeds records into downstream automation, the relevant mechanism is that schema validation catches responses that violate integration contracts; that architectural dependency addresses the observed failure where generated output occasionally omits required keys.

Incorrect Answers

B:

When a document model performs well overall but fails on tax identifiers, stratified field metrics protect high-impact values; in the stated workflow, however, generated output occasionally omits required keys; the relevant architectural behavior is that schema validation catches responses that violate integration contracts.

C:

Delegated authorization evaluates each caller's effective permissions; its main purpose is agent looks up records on behalf of employees with different rights, not correction of the reported defect where generated output occasionally omits required keys in the target process.

D:

The technical behavior of this alternative is that regression sets cover lexical variation and task boundaries; the user case being assessed is a Copilot prompt feeds records into downstream automation with a different failure: generated output occasionally omits required keys.

Question 840

Troubleshooting starts from a business complaint: Parallel agent updates create conflicting authoritative values. The affected process is: Two applications try to overwrite the same customer credit field. Which investigation is most directly tied to the reported defect?

  1. Audit samples for missing owner, classification and source identifiers
  2. Inspect write conflicts and source-of-truth definitions
  3. Inspect baseline equivalence and external trend factors
  4. Compare metered usage with adoption and concurrency assumptions

Explanation

Correct Answer

B:

The reported fault is that parallel agent updates create conflicting authoritative values in a system where two applications try to overwrite the same customer credit field; evidence from inspecting write conflicts and source-of-truth definitions can show whether field ownership and integration contracts prevent competing writes.

Incorrect Answers

A:

Auditing samples for missing owner, classification and source identifiers explores a separate connection, where metadata enables filtering, review, retention and traceable answers; the diagnosed requirement is two applications try to overwrite the same customer credit field, and the actual defect is parallel agent updates create conflicting authoritative values.

C:

Management attributes a sales lift to Copilot without a comparison; its matching diagnostic check is reasonable for seasonality produces the same change in unassisted teams; the present investigation needs evidence for the different failure where parallel agent updates create conflicting authoritative values.

D:

The proposed inspection is associated with an agent scales from a trial to enterprise usage, where workload forecasts include model usage, connectors, storage and observability; it would not expose the failure mode that parallel agent updates create conflicting authoritative values with the required specificity.

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