I85PREP
⌘ K
AB-100 study materials
AB-100 study resources

AB-100 practice questions

One question at a time. Build your AB-100 understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Plan AI-powered business solutions

An organization is deploying a customer-facing agent and must publish how the agent reaches its recommendations and tell users that they are interacting with AI. Which Microsoft Responsible AI principle does this primarily address?

  1. Fairness
  2. Reliability and safety
  3. Inclusiveness
  4. Transparency Correct answer.

Explanation

Transparency requires that people understand how an AI system behaves and know when they are interacting with AI rather than a person. Fairness concerns equitable treatment across groups, inclusiveness concerns accessibility, and reliability and safety concern consistent, safe operation.

Question 2Plan AI-powered business solutions

A company is defining a Power Platform environment strategy for solutions that include Copilot Studio agents. Which environment should everyday business users work in?

  1. The production environment Correct answer.
  2. A per-maker developer environment
  3. The default environment, shared by all makers and users
  4. A sandbox environment used for testing

Explanation

Production is the managed environment where end users consume released solutions, and it should be separated from build and test activities. The default environment is shared and ungoverned, so it is not the right place for a governed business workload.

Question 3Plan AI-powered business solutions

A leadership team wants an early, visible win from its AI investment. Which use case should the team select first?

  1. A long-term initiative with high strategic value and unresolved data dependencies
  2. A complex, cross-department process with many legacy integrations
  3. A high-value, low-complexity process with available, clean data Correct answer.
  4. A low-value process chosen only because a team already requested it

Explanation

Use case prioritization weighs business value against feasibility, including data readiness, integration complexity and time to deliver. A first pilot should be quick to deploy and easy to measure so it builds credibility for scaling.

Question 4Plan AI-powered business solutions

An agent must answer questions from the company's product manuals and internal policies. Which design choice best reduces fabricated answers?

  1. Add a disclaimer telling users that answers may be inaccurate
  2. Increase the model temperature so responses sound more natural
  3. Rely on the model's pre-trained knowledge of similar industries
  4. Ground responses in the manuals and policies as knowledge sources Correct answer.

Explanation

Grounding supplies authoritative content at query time and instructs the model to answer only from that content, which substantially reduces invented answers. Internal manuals and policies are exactly the kind of source an enterprise agent should cite.

Question 5Plan AI-powered business solutions

A team needs a custom agent that orchestrates several Azure-hosted models, uses a fine-tuned proprietary model, and is called by multiple line-of-business applications through an API. Which platform fits best?

  1. Power BI with Copilot
  2. Power Apps with a Dataverse table
  3. Copilot Studio with a SharePoint knowledge source
  4. Azure AI Foundry Correct answer.

Explanation

Azure AI Foundry provides a model catalog, fine-tuning, orchestration and code-level API access for complex multi-model solutions. Copilot Studio is optimized for low-code agents grounded in business data rather than for fine-tuning and orchestrating multiple Azure models.

Question 6Plan AI-powered business solutions

Makers must not combine the Salesforce connector and the SharePoint connector in the same agent. Which control should you configure?

  1. A conditional access policy in Microsoft Entra ID
  2. An Azure Firewall rule for outbound traffic
  3. A data loss prevention policy with connector classification Correct answer.
  4. A sensitivity label applied to the agent's documents

Explanation

Power Platform data loss prevention policies group connectors into business, non-business and blocked categories, and connectors placed in different groups cannot be used together. This is the supported mechanism for restricting which services a single agent or app can reach.

Question 7Plan AI-powered business solutions

An agent will be able to issue customer refunds of up to $5,000. Which design control is most appropriate?

  1. Allow the agent to act autonomously to reduce handling time
  2. Disable logging so customer data is not stored
  3. Raise the refund limit so fewer escalations occur
  4. Require human approval before the refund is executed Correct answer.

Explanation

Consequential financial actions call for human oversight, so the agent should prepare the transaction and hand off to a person for approval. This keeps the agent in an assistive role for high-impact decisions while still automating the surrounding work.

Question 8Plan AI-powered business solutions

After validation in a sandbox environment, how should a Power Platform solution containing Copilot Studio components be released to production?

  1. Recreate each component manually in the production environment
  2. Export the solution as managed and import it into production Correct answer.
  3. Copy the unmanaged solution files directly into the production environment
  4. Export the default environment as a patch and import it into production

Explanation

Managed solutions are the deployment artifact used for test and production environments, while unmanaged solutions are for development only. Importing a managed solution keeps environments consistent and supports clean upgrades and removal.

Question 9Plan AI-powered business solutions

You are planning an agent expected to handle around 20,000 conversations per month for external customers. Which licensing consideration is most important to validate during planning?

  1. Whether the agent is authored with the classic designer
  2. Whether every user holds a Microsoft 365 E3 licence
  3. Whether tenant message capacity is sufficient for projected volume Correct answer.
  4. Whether the tenant has an even number of environments

Explanation

Copilot Studio agents consume message capacity, which is provisioned through capacity packs or pay-as-you-go billing, so projected conversation volume must be mapped to available capacity before go-live. Capacity planning also depends on how many generative answers and actions each conversation triggers.

Question 10Plan AI-powered business solutions

Which Microsoft service should store business records that agents and Power Platform apps read and write with rich relationships and fine-grained security?

  1. Azure Blob Storage
  2. Microsoft Dataverse Correct answer.
  3. A Microsoft Excel workbook in OneDrive
  4. A SharePoint list

Explanation

Dataverse provides tabular storage with relationships, metadata and row- and column-level security that Power Platform workloads consume natively. This makes it the standard choice for structured business data used by apps, flows and agents.

Question 11Plan AI-powered business solutions

What is the primary purpose of retrieval-augmented generation in an enterprise agent?

  1. To supply relevant enterprise content to the model at query time Correct answer.
  2. To reduce the cost of Azure subscriptions
  3. To train a new foundation model from scratch
  4. To replace the need for data governance

Explanation

Retrieval-augmented generation finds relevant documents and injects them into the prompt so answers reflect current, authoritative enterprise content. It improves accuracy and traceability without changing the underlying model weights.

Question 12Plan AI-powered business solutions

An agent reads inbound customer emails and can call a flow that changes order status. Which risk is most significant for this design?

  1. Token overruns; cap every response at 100 tokens
  2. Model drift; retrain the model weekly on new email
  3. Prompt injection; treat email text as untrusted data Correct answer.
  4. Latency spikes; add a caching layer in front of the model

Explanation

Untrusted content arriving from outside the organization can contain instructions that attempt to redirect the agent, which is the classic prompt injection scenario. The mitigation is to treat email text as data rather than instructions and to require confirmation before any state-changing action.

Question 13Plan AI-powered business solutions

A regulated customer requires that agent conversation data and grounding sources remain within the European Union. Which planning activity matters most?

  1. Enable verbose logging for faster troubleshooting
  2. Confirm the environment region and the data residency commitments Correct answer.
  3. Schedule weekly exports of conversation transcripts
  4. Choose a model with the largest context window

Explanation

Data residency is determined by the geographic region in which the environment and its connected services are provisioned, so region selection must be validated against the regulatory requirement before build. Residency commitments should be confirmed for every component in the architecture, including the model endpoint and search index.

Question 14Plan AI-powered business solutions

A pilot agent has been running for one quarter and sponsors want evidence of business value. Which reporting approach is most appropriate?

  1. Report the total number of prompts that makers tried during the pilot
  2. Compare baseline and post-pilot handling time and resolution rate Correct answer.
  3. Track how many model deployments were made each month
  4. Count the number of Copilot Studio topics created for the agent

Explanation

Business value is demonstrated by movement in business metrics against a measured pre-deployment baseline, such as handling time, resolution rate or cost per case. Build and activity counts describe effort rather than outcomes.

Question 15Plan AI-powered business solutions

What is the primary role of an AI Center of Excellence in a large organization?

  1. To replace business owners in defining requirements
  2. To set standards, share reusable assets and govern AI adoption Correct answer.
  3. To approve every individual prompt written by makers
  4. To operate the organization's Azure billing accounts

Explanation

A Center of Excellence provides shared standards, templates, guardrails and expertise so business teams can build consistently without duplicating effort. It enables scale by codifying patterns rather than by performing every individual task centrally.

Question 16Plan AI-powered business solutions

Users are resisting a newly deployed agent that summarizes case notes. Which action best supports adoption?

  1. Run role-based training supported by a champion network Correct answer.
  2. Add more features before communicating with users again
  3. Publish a long policy document on the intranet
  4. Restrict access to the agent until resistance stops

Explanation

Adoption improves when users see role-relevant value and have peers and trainers who can answer their specific concerns. Targeted training plus champions addresses skill gaps and trust at the same time.

Question 17Plan AI-powered business solutions

An internal agent must return only data that each employee is already permitted to see. Which configuration is required?

  1. Authenticate users and query with their identity Correct answer.
  2. Enable anonymous access to reduce sign-in friction
  3. Use a single shared service account for all queries
  4. Cache all query results in a public storage account

Explanation

Querying with the signed-in user's identity lets the underlying data platform apply its own permission model, often called security trimming, so each user receives only content they can access. Without user identity there is no reliable basis for enforcing per-user permissions.

Question 18Plan AI-powered business solutions

A high-volume classification workload must respond in under a second and can tolerate moderate accuracy. Which model strategy is most appropriate?

  1. Fine-tune a frontier model on the full document corpus first
  2. Route every request through a multi-agent orchestration chain
  3. Deploy the largest available frontier model for every request
  4. Use a smaller, faster model validated on a labelled sample Correct answer.

Explanation

Model selection balances quality, latency and cost against the requirements of the task, and simple high-volume tasks rarely need the largest model. A smaller model benchmarked against labelled data confirms the accuracy is acceptable before committing to cost.

Question 19Plan AI-powered business solutions

A primary agent must delegate specialist work, such as HR policy lookups and IT ticket creation, to separate agents that have their own instructions and knowledge. Which capability should you plan for?

  1. A separate tenant for each department
  2. One knowledge source containing all department documents
  3. A single monolithic topic containing every instruction
  4. Connected agents that the primary agent invokes as tools Correct answer.

Explanation

Connected agents let a primary agent delegate to focused child agents that carry their own instructions and knowledge, keeping each scope small enough to test and maintain. This also limits which documents each specialist agent can reach.

Question 20Plan AI-powered business solutions

A sales team wants AI-generated opportunity summaries and email drafts inside the CRM it already uses every day. Which approach fits best?

  1. Replace the CRM with a Power Apps canvas app
  2. Enable Copilot features in Dynamics 365 Sales Correct answer.
  3. Build a custom model and expose it through a separate web portal
  4. Export CRM data nightly to spreadsheets for sellers to review

Explanation

In-product Copilot capabilities in Dynamics 365 Sales generate record summaries and draft emails using the CRM context sellers already work in. Keeping assistance inside the system of record avoids context switching and additional integration work.

Question 21Plan AI-powered business solutions

An agent must answer questions about the signed-in user's own meetings and files. Which grounding source should you plan to use?

  1. A public web search connector
  2. Microsoft Graph Correct answer.
  3. An Azure SQL table of product prices
  4. A static FAQ document

Explanation

Microsoft Graph exposes Microsoft 365 content such as mail, calendar and files in the context of the signed-in user, which is exactly the personal work data this scenario needs. It also lets existing permissions determine what the user can retrieve.

Question 22Plan AI-powered business solutions

Before releasing an agent to production, what should the evaluation plan include?

  1. A single demo run with the project sponsor present
  2. A labelled test set, human review, and pass criteria Correct answer.
  3. Load testing at three times the expected production volume
  4. Sign-off from the model vendor's support team

Explanation

Quality must be measured against expected answers, judged by people, and compared with thresholds agreed before testing so the release decision is evidence-based. Performance testing is a complementary activity rather than a substitute for quality evaluation.

Question 23Plan AI-powered business solutions

An agent must create a ticket in an external ITSM system after the user confirms the issue. Which component should perform the write operation?

  1. A Power Automate cloud flow called as an action Correct answer.
  2. A Power BI dataset refreshed on a schedule
  3. A topic that only returns a confirmation message
  4. A knowledge source pointing at the ITSM portal

Explanation

Cloud flows expose actions that the agent can invoke with inputs and return results, which is how agents write to external systems through connectors. Knowledge sources are read-only and messages alone do not change any system state.

Question 24Plan AI-powered business solutions

An agent grounded on a SharePoint site returns content from a restricted folder to users who are not authorized to see it. What is the most likely cause?

  1. The agent's topics were authored in the wrong language
  2. The model temperature was set too high during the conversation
  3. The connection uses credentials that can read the folder Correct answer.
  4. The site has more document libraries than the indexer can handle

Explanation

Retrieval honours the identity used to query the index, so a service account with broad access bypasses the restrictions that apply to end users. The fix is to query with the signed-in user's identity or constrain the connection's permissions.

Question 25Plan AI-powered business solutions

A sponsor asks for a complete business case for an agent expected to handle 100,000 conversations per month. Which cost categories should the case include?

  1. Only the one-time cost of building and testing the agent
  2. Only the AI model token charges for each conversation
  3. Consumption, licensing, build and ongoing operations Correct answer.
  4. Only the internal staff time spent during the pilot

Explanation

A credible business case covers total cost of ownership: capacity and licensing, implementation and integration effort, and the recurring run costs of monitoring, evaluation and maintenance. It should also be compared with the current baseline cost of the process.

Question 26Plan AI-powered business solutions

A company wants its new HR agent to be available only to employees inside Microsoft Teams. Which planning decision does this requirement primarily affect?

  1. Channel selection Correct answer.
  2. Model selection
  3. Data retention
  4. Licensing tier

Explanation

Publishing channels determine where an agent is exposed and which audience can reach it. Choosing Microsoft Teams as the only channel scopes the agent to internal employees who use Teams. Model and retention decisions are separate design concerns.

Question 27Plan AI-powered business solutions

Which artifact should be defined first to measure whether an agent delivers business value?

  1. The maximum conversation length
  2. Business success metrics and KPIs Correct answer.
  3. A list of all available connectors
  4. The agent's color theme

Explanation

Business value must be tied to measurable outcomes such as resolution rate, time saved, or cost avoidance. Defining KPIs before build keeps design and evaluation aligned. Connectors and UI details do not establish value.

Question 28Plan AI-powered business solutions

Why should development, test, and production agents be placed in separate Power Platform environments?

  1. To increase the model's accuracy automatically
  2. To reduce the number of available connectors
  3. To isolate changes and control releases Correct answer.
  4. To eliminate the need for authentication

Explanation

Separate environments create boundaries so changes can be validated before they affect production users. They also support controlled promotion and rollback. Model accuracy and authentication are not determined by environment separation.

Question 29Plan AI-powered business solutions

Which Microsoft service should provide sign-in for an agent that must use corporate credentials?

  1. Azure Key Vault
  2. Power BI
  3. Microsoft Purview
  4. Microsoft Entra ID Correct answer.

Explanation

Microsoft Entra ID is the identity provider for Microsoft cloud services and supports corporate sign-in, conditional access, and single sign-on. Key Vault stores secrets, Purview governs data, and Power BI is analytics.

Question 30Plan AI-powered business solutions

An agent must transfer a conversation to a human when it cannot resolve the request. Which design element should be planned?

  1. An escalation path to a human queue Correct answer.
  2. A larger language model
  3. More manually authored topics
  4. A public website knowledge source

Explanation

Escalation design ensures unresolved or high-risk conversations reach a person with the right context. It should define triggers, routing, and handoff information. Model size or topic count does not guarantee a safe handoff.

Question 31Plan AI-powered business solutions

External customers who do not have Microsoft 365 accounts need to use an agent. Which channel is appropriate?

  1. A public website or custom app Correct answer.
  2. A Microsoft Teams channel limited to the organization
  3. A SharePoint site
  4. An internal Outlook mailbox

Explanation

External users need a channel they can reach without organizational credentials, such as a public website or custom app. Teams, SharePoint, and internal mailboxes are typically restricted to the organization. Authentication options should still be planned for the external channel.

Question 32Plan AI-powered business solutions

Which practice helps protect an agent from prompt injection supplied through user input?

  1. Treating input as untrusted Correct answer.
  2. Disabling conversation logging
  3. Increasing the model temperature
  4. Lengthening the system prompt

Explanation

Prompt injection defense starts by treating user and retrieved content as untrusted and constraining what the agent can do with it. High-impact actions should also require validation or approval. Longer prompts and higher temperature do not provide a security boundary.

Question 33Plan AI-powered business solutions

What is the main purpose of publishing an agent to a channel?

  1. To reduce the licensing cost
  2. To train the underlying language model
  3. To encrypt the agent's data
  4. To expose it to intended users Correct answer.

Explanation

Publishing makes the agent reachable through a selected channel and audience. It does not train the model or change encryption. Licensing and security are planned separately.

Question 34Plan AI-powered business solutions

A company wants an agent to answer questions from a large policy library that changes weekly, with minimal manual topic authoring. Which approach should be planned?

  1. A manually authored topic for every policy
  2. Generative answers over the library Correct answer.
  3. A canvas app that displays the documents
  4. A hard-coded FAQ stored in a variable

Explanation

Generative answers use the grounded content to synthesize responses, which scales better than authoring a topic per document. The library must still be curated and refreshed. Hard-coded FAQs become stale and do not scale.

Question 35Plan AI-powered business solutions

An agent must call a line-of-business API that has no prebuilt Power Platform connector. Which capability should be planned?

  1. A Power BI dataset
  2. A custom connector Correct answer.
  3. A Dataverse business card
  4. A model-driven app form

Explanation

A custom connector wraps an API definition so Power Platform and agents can call it through a governed connector. Power BI, business cards, and model-driven forms do not provide API invocation. The connector should be reviewed under DLP policy.

Question 36Plan AI-powered business solutions

Which metric best indicates that an agent is resolving requests without human involvement?

  1. Containment or resolution rate Correct answer.
  2. Number of authored topics
  3. Total number of messages sent
  4. Average prompt length

Explanation

Containment or resolution rate measures conversations completed without escalation or repeat contact. It reflects whether the agent actually satisfied the request. Message counts and topic counts can grow without improving outcomes.

Question 37Plan AI-powered business solutions

A maker wants to package agent components and reuse them across multiple agents in the same environment. What should be planned?

  1. A Power Platform solution package Correct answer.
  2. Anonymous authentication
  3. A separate tenant for each agent
  4. A single monolithic topic

Explanation

Solutions package components for transport and reuse across environments. They support versioning and controlled deployment. Separate tenants, anonymous access, and monolithic topics work against reuse and governance.

Question 38Plan AI-powered business solutions

An agent action must run under the signed-in user's permissions instead of a shared account. What should be configured?

  1. Anonymous access to the connector
  2. A single admin account for all users
  3. Per-user authenticated connections Correct answer.
  4. A shared service principal connection

Explanation

Per-user connections ensure actions execute with the individual user's identity and permissions. This supports least privilege and auditability. Shared or anonymous access can grant more rights than the user should have.

Question 39Plan AI-powered business solutions

Before a pilot, which group should confirm that an agent's answers meet quality and compliance standards?

  1. SMEs and compliance reviewers Correct answer.
  2. The software vendor alone
  3. Only the development team
  4. A random sample of external users

Explanation

Subject matter experts validate correctness, while compliance reviewers check policy and regulatory fit. Development alone lacks business context, and external users should not be first-line validators. Vendor input does not replace organizational accountability.

Question 40Plan AI-powered business solutions

Which approach best estimates an agent's operating cost before launch?

  1. Reuse last year's total IT budget
  2. Count the number of connectors used
  3. Assume the agent has no operating cost
  4. Forecast volume and consumption Correct answer.

Explanation

Operating cost depends on expected usage, licensing model, and consumption-based services. Forecasting conversation volume against those rates produces a defensible estimate. Historical IT budgets and connector counts do not model agent usage.

Question 41Plan AI-powered business solutions

An agent will serve employees in several countries who prefer to interact in their local language. Which planning consideration applies?

  1. Adding more connectors
  2. Increasing the model temperature
  3. Changing the agent's color theme
  4. Language localization Correct answer.

Explanation

Localization covers language, terminology, and regional expectations for each audience. It should be planned in content and testing. Model temperature, connectors, and theming do not address language needs.

Question 42Plan AI-powered business solutions

What is the primary role of conversation analytics after an agent is deployed?

  1. Setting the organization's licensing tier
  2. Encrypting conversation data
  3. Replacing all pre-launch testing
  4. Finding gaps to improve content Correct answer.

Explanation

Analytics reveals unresolved questions, escalation patterns, and content gaps that should drive improvements. It complements rather than replaces pre-launch testing. Licensing and encryption are separate governance decisions.

Question 43Plan AI-powered business solutions

An agent must collect a required order number before running a lookup action. Which design element should be planned?

  1. A public website channel for customers
  2. A validated input question Correct answer.
  3. A higher model temperature setting
  4. A Power BI dashboard for reporting

Explanation

A question node with validation ensures the required value is captured before the action runs. This reduces failed lookups and improves conversation flow. Temperature, channel, and reporting do not enforce input requirements.

Question 44Plan AI-powered business solutions

Which practice improves an agent's ability to answer follow-up questions in context?

  1. Maintaining conversation context Correct answer.
  2. Increasing the model temperature
  3. Disabling authentication
  4. Reducing the number of topics

Explanation

Conversation context allows the agent to interpret pronouns and references across turns. It should be retained within the session and handled carefully for privacy. Temperature and topic count do not provide continuity.

Question 45Plan AI-powered business solutions

An agent's answers must cite the source document for auditability. Which design choice supports this requirement?

  1. Allowing anonymous access
  2. Extending the response timeout
  3. Increasing the model temperature
  4. Enable citations in grounding Correct answer.

Explanation

Citations link generated answers back to the grounding source, which supports verification and audit. The source content must also be accurate and access-controlled. Timeouts and temperature do not provide traceability.

Question 46Plan AI-powered business solutions

A company can buy a SaaS agent or build one with Copilot Studio. Which analysis is most relevant to the decision?

  1. The model's temperature setting
  2. Five-year total cost of ownership Correct answer.
  3. The number of topics in the demo
  4. The agent's color scheme

Explanation

Build-versus-buy decisions should compare total cost of ownership, including licensing, integration, maintenance, and change management. Demo topic count and cosmetic settings do not represent lifecycle cost. The option with the lowest upfront price may not be cheapest over time.

Question 47Plan AI-powered business solutions

A regulated company must retain all agent prompts and responses for seven years. Which planning area should be addressed?

  1. Retention and compliance policies Correct answer.
  2. Prompt wording and tone style
  3. Model temperature and creativity settings
  4. Channel selection for external users

Explanation

Retention requirements must be mapped to Microsoft Purview retention and compliance controls for the relevant data locations. The policy should cover prompts, responses, and related logs. Tone and channel choices do not satisfy legal retention.

Question 48Plan AI-powered business solutions

A company wants to verify that an agent's responses match its brand tone before broad release. Which evaluation method is most appropriate?

  1. Measuring average response latency
  2. Human review with a rubric Correct answer.
  3. Counting every configured connector
  4. Load testing the public web channel

Explanation

Brand tone is a qualitative attribute, so sampled human review against a defined rubric is the most direct evaluation. Automated load and latency tests measure performance, not tone. Connector counts say nothing about response style.

Question 49Plan AI-powered business solutions

An agent grounded on a public website starts giving outdated answers. Which governance action should be taken first?

  1. Add more authored topics
  2. Increase the model temperature
  3. Refresh the knowledge source Correct answer.
  4. Disable citations

Explanation

Outdated answers usually indicate stale or unindexed grounding content, so the source should be refreshed and reindexed. A content update cadence and ownership should also be defined. Temperature and topic count do not fix stale sources.

Question 50Plan AI-powered business solutions

An agent serves external users and must not reveal internal system names or its instructions. Which control is most appropriate?

  1. Restrictive instructions and moderation Correct answer.
  2. A longer response timeout for all users
  3. A higher model temperature for creativity
  4. Anonymous access for every external user

Explanation

System instructions should define what the agent must not disclose, and moderation should filter unsafe or sensitive output. Grounding sources should also exclude internal-only content. Temperature, timeouts, and anonymous access do not prevent disclosure.