UUS Practice Tests⌕ Search
IT Certifications · AB-100

Exam AB-100: Agentic AI Business Solutions Architect Practice Test

AB-100 practice addresses the architecture of business solutions that use agentic AI: systems that interpret requests, select actions, and interact with business applications. The subject includes defining an agent’s scope, connecting it to enterprise data and services, and managing risks when it handles sensitive information or performs consequential operations.

Exam overview

Exam providerMicrosoft
Exam codeAB-100
CategoryIT Certifications

About the Exam AB-100: Agentic AI Business Solutions Architect Exam

Architects must align an agent’s capabilities with a business process, choose appropriate interaction modes, and define how it exchanges data with systems such as CRM and order platforms. Key design decisions include limiting permissions, requiring approval for high-impact actions, and handling inconsistent interfaces or slow dependencies. Security work includes resisting instruction-based attacks and controlling access to restricted information. Evaluation should test task-level outcomes, such as whether service requests reach the right team, alongside quality across languages and user groups. For geographically distributed deployments, architects also need to account for data-residency constraints. These practice scenarios exercise judgment about scope, safeguards, integration patterns, resilience, and evidence of readiness rather than memorization of a single implementation.

Sample Practice Questions

Review a selection from this exam's published US Practice Tests practice bank.

Question 1

A company needs an agent to serve users in multiple regions, and some user data must remain within a specified jurisdiction. What should architects do first?

  1. AAssume the model provider will route all data appropriately by default
  2. BMap data flows and applicable residency constraints, then select and configure services accordingly
  3. CStore all data in the region with the lowest service cost
  4. DDisable logging without checking whether other components retain data
Show answerCorrect answer: B

Residency requirements must be assessed across the full data flow, including model calls, storage, telemetry, and integrations. The architecture can then be configured and validated against the organization's obligations.

Question 2

An agent may receive messages that contain instructions to ignore its rules and disclose restricted information. Which defense is most appropriate?

  1. ATreat every instruction found in retrieved documents as trusted because it came from a search result
  2. BRely only on a longer system prompt to prevent all harmful behavior
  3. CTreat external content as untrusted data and enforce access and action policies outside the model
  4. DDisable logging so malicious instructions are not recorded
Show answerCorrect answer: C

Prompt injection can appear in user input or retrieved content. Treating that content as untrusted and enforcing permissions and action constraints in application logic provides stronger protection than relying on prompt wording alone.

Question 3

An organization is choosing between a general-purpose agent and a narrowly scoped agent for a specialized internal task. Which consideration most directly supports choosing the narrower scope?

  1. AA narrower scope can make the agent's permitted tasks, tools, and expected behavior easier to define and test
  2. BA narrower scope guarantees the agent will never make an error
  3. CA general-purpose agent cannot use business data
  4. DA narrow agent never requires maintenance after deployment
Show answerCorrect answer: A

A bounded purpose can simplify design, testing, and governance by making capabilities and expected behavior more specific. It does not eliminate errors or ongoing maintenance, so those still need to be addressed.

Question 4

A business is evaluating whether an agent correctly routes incoming service requests to the right team. Which evaluation method provides the most direct evidence?

  1. AMeasure routing accuracy on a representative, labeled set of requests
  2. BMeasure how quickly the agent produces a response on one example
  3. CAsk the agent whether it believes its routing decisions are correct
  4. DCount the number of tools the agent can call
Show answerCorrect answer: A

A representative labeled evaluation set allows predicted routes to be compared with expected routes, yielding a direct measure of routing quality. Latency and tool count do not establish correctness.

Question 5

A team is deciding whether a business agent should support users who communicate by voice as well as text. What should the team evaluate before choosing the interaction mode?

  1. AWhether the voice experience fits the task and user context, including privacy, accessibility, noise, and error-correction needs
  2. BWhether voice is more novel than a text interface
  3. CWhether voice eliminates the need to confirm important information
  4. DWhether the agent can use the same wording in speech and text without adaptation
Show answerCorrect answer: A

Interaction mode should fit the task and environment. Voice can raise distinct privacy, accessibility, recognition, and correction concerns, so its suitability should be evaluated rather than assumed.

Question 6

An agent can submit purchase orders through a connected business system. What is the most appropriate safeguard before it submits an order above a defined spending limit?

  1. ARequire an authorized person to review and approve the proposed order
  2. BAllow the agent to submit it if its generated explanation is detailed
  3. CRetry the submission several times if the system responds slowly
  4. DLet the agent approve its own order using a second prompt
Show answerCorrect answer: A

A human approval gate is an appropriate control for consequential actions above a risk-based threshold. The agent can prepare the order, but an authorized person should approve it before execution.

Question 7

An agent needs to produce customer-facing text in several languages. What is the most responsible way to assess whether it is ready for those users?

  1. ATranslate a single sample answer and assume quality is consistent across languages
  2. BEvaluate representative tasks with fluent reviewers for each supported language and check for language-specific failure patterns
  3. CUse only the English evaluation set because the underlying model is multilingual
  4. DEnable every language at once and rely on user complaints to identify problems
Show answerCorrect answer: B

Performance can vary by language, dialect, task, and cultural context. Representative evaluation with fluent reviewers helps reveal quality and safety issues before broad availability.

How to use this practice test

The free US Practice Tests bank contains 50 published questions. Each practice session presents up to 20 questions selected at random, then shows results. Use the results to identify which agent architecture decisions need closer review, such as permission boundaries, approval controls, evaluation evidence, or integration reliability.

Preparation tips

  • For each proposed agent action, map the required identity, data access, write permissions, and approval threshold before selecting a safeguard.
  • Compare broad and narrowly scoped agents by the task boundary, tools exposed, and potential impact of an incorrect action.
  • Practice tracing a request across systems with different APIs and data models; identify where validation, retries, or human intervention may be needed.
  • Design evaluation cases for routing accuracy, multilingual output, and resistance to malicious instructions, and specify observable success criteria.
  • For regional deployments, identify where data is stored and processed before choosing an architecture that meets jurisdictional constraints.
Official sourceOfficial exam page ↗