100% Latest & Updated Microsoft AB-100 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 50
Separate specialist agents must delegate tasks to one another across service boundaries while maintaining responsibilities. Which controlled change is most appropriate for this situation?
- Integrate through a governed API or Power Platform connector.
- Define an Agent2Agent delegation contract between the specialist agents.
- Define roles, decision rights and escalation responsibilities in the AI operating model.
- Use an explicitly modeled orchestration flow with auditable steps.
Explanation
Correct Answer
B:A2A addresses agent-to-agent task exchange rather than exposing a simple tool catalog.
Incorrect Answers
A:Direct transactional APIs suit the deterministic system operation without inventing agent negotiation, whereas a2A addresses agent-to-agent task exchange rather than exposing a simple tool catalog.
C:Named governance responsibilities close accountability gaps around high-impact AI behavior, but a2A addresses agent-to-agent task exchange rather than exposing a simple tool catalog.
D:A deterministic workflow represents fixed ordering and retries better than an unconstrained dialogue. A2A addresses agent-to-agent task exchange rather than exposing a simple tool catalog.
Question 51
A non-agent system must send an exact purchase order to ERP through a published REST contract. Which design choice targets the observed problem?
- Use an explicitly modeled orchestration flow with auditable steps.
- Connect through a properly secured Model Context Protocol tool server.
- Integrate through a governed API or Power Platform connector.
- Specify a strict structured-output contract and validate required schema fields.
Explanation
Correct Answer
C:Direct transactional APIs suit the deterministic system operation without inventing agent negotiation.
Incorrect Answers
A:A deterministic workflow represents fixed ordering and retries better than an unconstrained dialogue. Direct transactional APIs suit the deterministic system operation without inventing agent negotiation.
B:MCP is designed to expose interoperable tools and context to AI clients with explicit interfaces, but direct transactional APIs suit the deterministic system operation without inventing agent negotiation.
D:An output contract makes machine consumption reliable and rejects malformed generated structures; direct transactional APIs suit the deterministic system operation without inventing agent negotiation.
Question 52
An approval must execute a fixed sequence of notifications and updates independent of conversational reasoning. Which concrete solution decision should the team make?
- Define an Agent2Agent delegation contract between the specialist agents.
- Build the conversational solution in Microsoft Copilot Studio.
- Connect through a properly secured Model Context Protocol tool server.
- Use an explicitly modeled orchestration flow with auditable steps.
Explanation
Correct Answer
D:A deterministic workflow represents fixed ordering and retries better than an unconstrained dialogue.
Incorrect Answers
A:A2A addresses agent-to-agent task exchange rather than exposing a simple tool catalog, but a deterministic workflow represents fixed ordering and retries better than an unconstrained dialogue.
B:Copilot Studio is oriented to conversational agents and tool integration for business workflows. A deterministic workflow represents fixed ordering and retries better than an unconstrained dialogue.
C:MCP is designed to expose interoperable tools and context to AI clients with explicit interfaces; a deterministic workflow represents fixed ordering and retries better than an unconstrained dialogue.
Question 53
Ten agents use conflicting instructions about how to handle unsupported facts and citations. Which recommendation addresses the identified failure most directly?
- Complete privacy and permitted-use review before enabling data reuse.
- Publish a governed prompt library with approved templates and version ownership.
- Separate trusted system instructions from untrusted user and retrieved content.
- Version prompts and run regression evaluations against representative cases.
Explanation
Correct Answer
B:Central prompt artifacts make instructions reviewable and consistent across teams.
Incorrect Answers
A:Data governance requires a valid purpose and lawful processing basis for model improvement. Central prompt artifacts make instructions reviewable and consistent across teams.
C:Instruction hierarchy and content boundaries reduce inadvertent promotion of untrusted text, whereas central prompt artifacts make instructions reviewable and consistent across teams.
D:Prompt versioning plus stable test cases establishes the causal effect of a change. Central prompt artifacts make instructions reviewable and consistent across teams.
Question 54
Model replies change after a prompt tweak and the team cannot tell whether the change improved accuracy. What should the solution architect recommend for this specific need?
- Specify a strict structured-output contract and validate required schema fields.
- Separate trusted system instructions from untrusted user and retrieved content.
- Version prompts and run regression evaluations against representative cases.
- Perform stakeholder impact review and change-management planning before rollout.
Explanation
Correct Answer
C:Prompt versioning plus stable test cases establishes the causal effect of a change.
Incorrect Answers
A:An output contract makes machine consumption reliable and rejects malformed generated structures; prompt versioning plus stable test cases establishes the causal effect of a change.
B:Instruction hierarchy and content boundaries reduce inadvertent promotion of untrusted text, whereas prompt versioning plus stable test cases establishes the causal effect of a change.
D:Adoption planning addresses human workflows and acceptance beyond technical model performance, but prompt versioning plus stable test cases establishes the causal effect of a change.
Question 55
A support agent is given raw user input inserted into a high-trust system instruction with no boundary. What is the most defensible intervention for this case?
- Publish a governed prompt library with approved templates and version ownership.
- Specify a strict structured-output contract and validate required schema fields.
- Publish shared identity and evidence contracts for inter-agent messages.
- Separate trusted system instructions from untrusted user and retrieved content.
Explanation
Correct Answer
D:Instruction hierarchy and content boundaries reduce inadvertent promotion of untrusted text.
Incorrect Answers
A:Central prompt artifacts make instructions reviewable and consistent across teams, whereas instruction hierarchy and content boundaries reduce inadvertent promotion of untrusted text.
B:An output contract makes machine consumption reliable and rejects malformed generated structures. Instruction hierarchy and content boundaries reduce inadvertent promotion of untrusted text.
C:Consistent identity semantics prevent one agent from trusting an unverified claim passed by another, but instruction hierarchy and content boundaries reduce inadvertent promotion of untrusted text.
Question 56
Agents should return a validated JSON claim summary instead of free-form paragraphs for downstream software. Which architecture response most directly addresses the risk?
- Specify a strict structured-output contract and validate required schema fields.
- Publish shared identity and evidence contracts for inter-agent messages.
- Publish a governed prompt library with approved templates and version ownership.
- Version prompts and run regression evaluations against representative cases.
Explanation
Correct Answer
A:An output contract makes machine consumption reliable and rejects malformed generated structures.
Incorrect Answers
B:Consistent identity semantics prevent one agent from trusting an unverified claim passed by another, but an output contract makes machine consumption reliable and rejects malformed generated structures.
C:Central prompt artifacts make instructions reviewable and consistent across teams. An output contract makes machine consumption reliable and rejects malformed generated structures.
D:Prompt versioning plus stable test cases establishes the causal effect of a change; an output contract makes machine consumption reliable and rejects malformed generated structures.
Question 57
High-volume helpdesk routing contains short repetitive requests and tolerates a lower-cost fast model. Which solution approach best meets the stated requirement?
- Use a governed model router choosing supported models by quality and cost requirements.
- Select a higher-quality reasoning model based on evaluation outcomes.
- Prefer a capable smaller model meeting the quality threshold for routine classification.
- Use Microsoft Foundry agent and model development capabilities.
Explanation
Correct Answer
C:Model right-sizing reduces per-request spend without sacrificing a validated simple task.
Incorrect Answers
A:Routing permits per-request selection while maintaining a declared quality and policy envelope, but model right-sizing reduces per-request spend without sacrificing a validated simple task.
B:Demanding analysis warrants measured accuracy rather than choosing solely the cheapest tokens. Model right-sizing reduces per-request spend without sacrificing a validated simple task.
D:A code-first controlled engineering platform supports custom runtime, evaluation and model choices, whereas model right-sizing reduces per-request spend without sacrificing a validated simple task.
Question 58
A complex acquisition document requires multi-source reasoning and a strict accuracy target. Which option most directly closes the documented gap?
- Keep a conventional deterministic application experience instead of adding an agent.
- Use a governed model router choosing supported models by quality and cost requirements.
- Assess targeted model customization against a tested baseline and maintenance cost.
- Select a higher-quality reasoning model based on evaluation outcomes.
Explanation
Correct Answer
D:Demanding analysis warrants measured accuracy rather than choosing solely the cheapest tokens.
Incorrect Answers
A:A static task has no defensible reasoning or tool-selection benefit from generative AI. Demanding analysis warrants measured accuracy rather than choosing solely the cheapest tokens.
B:Routing permits per-request selection while maintaining a declared quality and policy envelope; demanding analysis warrants measured accuracy rather than choosing solely the cheapest tokens.
C:Fine-tuning or domain customization is justified only by demonstrated gaps and sustainable data governance, whereas demanding analysis warrants measured accuracy rather than choosing solely the cheapest tokens.
Question 59
Different departments send tasks of widely varying complexity and budget, but a single fixed model is wasteful. Which approach best satisfies the actual business constraint?
- Use a governed model router choosing supported models by quality and cost requirements.
- Assess targeted model customization against a tested baseline and maintenance cost.
- Use Microsoft Foundry agent and model development capabilities.
- Prefer a capable smaller model meeting the quality threshold for routine classification.
Explanation
Correct Answer
A:Routing permits per-request selection while maintaining a declared quality and policy envelope.
Incorrect Answers
B:Fine-tuning or domain customization is justified only by demonstrated gaps and sustainable data governance, whereas routing permits per-request selection while maintaining a declared quality and policy envelope.
C:A code-first controlled engineering platform supports custom runtime, evaluation and model choices. Routing permits per-request selection while maintaining a declared quality and policy envelope.
D:Model right-sizing reduces per-request spend without sacrificing a validated simple task, but routing permits per-request selection while maintaining a declared quality and policy envelope.
Question 60
A company has a unique domain language with enough lawful representative data and generic models consistently fail evaluations. Which specific remediation should the business adopt?
- Prefer a capable smaller model meeting the quality threshold for routine classification.
- Assess targeted model customization against a tested baseline and maintenance cost.
- Version prompts and run regression evaluations against representative cases.
- Select a higher-quality reasoning model based on evaluation outcomes.
Explanation
Correct Answer
B:Fine-tuning or domain customization is justified only by demonstrated gaps and sustainable data governance.
Incorrect Answers
A:Model right-sizing reduces per-request spend without sacrificing a validated simple task, whereas fine-tuning or domain customization is justified only by demonstrated gaps and sustainable data governance.
C:Prompt versioning plus stable test cases establishes the causal effect of a change, but fine-tuning or domain customization is justified only by demonstrated gaps and sustainable data governance.
D:Demanding analysis warrants measured accuracy rather than choosing solely the cheapest tokens. Fine-tuning or domain customization is justified only by demonstrated gaps and sustainable data governance.
Agentic AI Business Solutions Architect (AB-100) Practice Tests