View Full Microsoft AB-100 Exam Dumps and Practice Test Dumps.

 

Question 141

What is the primary purpose of agent planning?

1. To define the steps needed to achieve a goal
2. To increase storage capacity
3. To remove security controls
4. To disable model evaluation

Correct Answer: 1

Explanation

Agent planning helps an AI system determine the steps required to accomplish a defined goal. A plan may involve selecting tools, retrieving information, completing intermediate tasks, and deciding what should happen next based on results. Planning is especially useful for complex workflows that cannot be completed through a single response. The plan should remain within the agent’s authorized scope and should be evaluated for reliability. For high-impact operations, additional validation or human approval may also be appropriate.

Question 142

Which capability allows an agent to use a business system to perform an approved action?

1. Context window
2. Tool integration
3. Model compression
4. Prompt formatting

Correct Answer: 2

Explanation

Tool integration allows an AI agent to interact with external systems and perform approved operations. A tool could expose functions for retrieving customer information, creating records, checking inventory, or initiating a workflow. The agent should not receive unrestricted access simply because a tool is available. Authentication, authorization, input validation, and auditing should be applied to tool operations. Clear tool definitions also help the agent understand what an action does and when it is appropriate to invoke it.

Question 143

A RAG system retrieves outdated product documentation. What should the team investigate first?

1. Screen resolution
2. Model temperature only
3. Data ingestion and refresh processes
4. User interface colors

Correct Answer: 3

Explanation

Outdated retrieval results can indicate that the underlying knowledge source or indexing process is not being refreshed correctly. The team should investigate data ingestion, document updates, indexing schedules, synchronization failures, and source ownership. Retrieval quality depends not only on the language model but also on the freshness and accuracy of the indexed information. Establishing refresh processes and monitoring data freshness can help prevent agents from providing obsolete information. The source should also be authoritative and appropriate for the business scenario.

Question 144

Why might an AI solution use a queue for incoming requests?

1. To permanently delete requests
2. To manage workload spikes and processing capacity
3. To remove authentication
4. To bypass rate limits

Correct Answer: 2

Explanation

A queue can help manage bursts of incoming requests when processing capacity is limited or when tasks do not need to be completed immediately. Requests can be placed in the queue and processed as resources become available. This can help smooth workload spikes and protect downstream services from sudden excessive traffic. Queue-based designs should define appropriate retry, timeout, prioritization, and failure-handling behavior. For interactive scenarios where immediate responses are required, the architecture should also consider the additional latency introduced by queued processing.

Question 145

Which practice supports secure AI application development?

1. Applying security testing throughout the lifecycle
2. Granting all users administrator access
3. Disabling authentication during deployment
4. Publishing secrets in configuration files

Correct Answer: 1

Explanation

Security testing should be incorporated throughout the AI application’s lifecycle rather than being performed only after deployment. Teams should evaluate authentication, authorization, data protection, prompt injection, tool permissions, sensitive information handling, and other relevant risks. Security controls should also be reviewed whenever the model, prompts, tools, or data sources change. Testing helps identify vulnerabilities before they affect production users. Secure development should work together with monitoring, governance, access management, and operational controls to provide broader protection.

Question 146

An organization needs the same AI behavior across several applications. What can improve consistency?

1. Using unrelated prompts in every application
2. Standardized prompts and reusable components
3. Removing evaluation datasets
4. Giving each application different security rules

Correct Answer: 2

Explanation

Standardized prompts and reusable components can help maintain consistent AI behavior across multiple applications. Organizations can define approved instructions, reusable tools, evaluation criteria, and governance requirements that teams can apply across solutions. Reuse also reduces duplicated development effort and makes maintenance easier. However, shared components should still respect application-specific permissions and data boundaries. Changes to common prompts or components should be versioned and tested because an update can affect multiple applications simultaneously.

Question 147

What should an AI system do when retrieved information does not contain enough evidence to answer a question?

1. Clearly indicate insufficient information or escalate
2. Invent missing facts
3. Ignore the retrieved sources
4. Provide unrestricted access to all data

Correct Answer: 1

Explanation

When available evidence is insufficient, the AI system should avoid presenting unsupported information as fact. Depending on the scenario, it can state that the available information is insufficient, ask the user for clarification, retrieve additional approved information, or escalate the request to a human. This behavior improves trust and reduces unsupported responses. The solution should define appropriate fallback behavior during design and test cases where retrieval returns incomplete, conflicting, or irrelevant information.

Question 148

Which metric is particularly relevant when users require fast AI responses?

1. Data retention
2. Latency
3. Document size
4. User password length

Correct Answer: 2

Explanation

Latency measures the time between a request and the corresponding response or processing result. It is especially important for interactive AI applications where users expect rapid responses. Latency can be affected by model selection, prompt size, retrieval operations, tool calls, network communication, and service capacity. Teams should establish acceptable response-time targets based on the business scenario. Monitoring latency in production can also identify performance changes and help determine whether optimization or architectural changes are necessary.

Question 149

What is the purpose of defining an AI solution’s business KPI?

1. To determine measurable business outcomes
2. To increase token usage
3. To remove user feedback
4. To prevent model updates

Correct Answer: 1

Explanation

A business KPI provides a measurable way to determine whether an AI solution is delivering its intended business value. Depending on the use case, KPIs could include reduced processing time, improved task completion, lower support workload, increased user satisfaction, or reduced operational cost. KPIs should be connected to the original business objective rather than selected only because they are easy to measure. Defining them before deployment or a pilot makes it easier to evaluate results consistently and identify areas requiring improvement.

Question 150

Which approach can help ensure an AI agent uses only approved knowledge sources?

1. Allowing unrestricted web access
2. Defining controlled knowledge sources
3. Removing source validation
4. Ignoring data ownership

Correct Answer: 2

Explanation

Controlled knowledge sources allow an organization to determine which information an agent is permitted to use when generating responses. Approved sources can be selected based on accuracy, ownership, relevance, privacy, and business requirements. Restricting knowledge sources also reduces the risk of an agent relying on outdated, unauthorized, or inappropriate information. Access controls should remain in place when sources contain sensitive data. Source governance and periodic review are important because business information and organizational requirements can change over time.

Question 151

What does model routing allow an AI application to do?

1. Select an appropriate model based on request requirements
2. Remove all models from the application
3. Disable monitoring
4. Store passwords inside prompts

Correct Answer: 1

Explanation

Model routing allows an AI application to select among available models based on factors such as task complexity, required quality, latency, cost, or capability. A simple request may be routed to a smaller model, while a complex reasoning task may require a more capable model. Routing rules should be tested against representative workloads to confirm that the selected models meet business requirements. Monitoring is also important because changes in traffic or model behavior may require adjustments to routing decisions.

Question 152

A company wants an agent to summarize documents while preserving important factual details. What should be evaluated?

1. Keyboard compatibility
2. Summary accuracy and relevance
3. Screen size
4. Network cable length

Correct Answer: 2

Explanation

For document summarization, evaluation should determine whether important facts are preserved while unnecessary information is removed. Relevant measures can include factual accuracy, completeness, relevance, readability, and consistency with the source material. Representative documents should be included in the evaluation dataset, including documents with different lengths and structures. Human review may also be useful for higher-risk content. The goal is not simply to produce shorter text, but to ensure that the summary remains useful and does not introduce unsupported information.

Question 153

Which component can connect an AI solution to Microsoft Power Platform business processes?

1. Power Platform connectors or flows
2. A display driver
3. A file compression utility
4. A browser theme

Correct Answer: 1

Explanation

Power Platform connectors and flows can provide integration between AI solutions and business processes across supported services. For example, an AI application may trigger an approved workflow, retrieve information from a connected service, or initiate a business process. Integration should still use appropriate authentication, authorization, and data-access controls. The available connector capabilities and permissions should be evaluated against the business requirement. Proper error handling and monitoring are also important when AI-driven workflows can modify business data.

Question 154

Why is human oversight useful for certain AI decisions?

1. It guarantees the model is always correct
2. It provides review for decisions that require human judgment
3. It eliminates all operational costs
4. It removes the need for access controls

Correct Answer: 2

Explanation

Human oversight provides an additional review mechanism when AI-generated recommendations or actions have significant consequences. A human reviewer can assess the available information and approve, reject, or modify the proposed action. This is particularly useful for sensitive decisions involving financial, legal, customer, employment, or other high-impact processes. Human oversight should be designed as part of the workflow rather than added informally after deployment. The system should clearly communicate what the AI proposes and what the reviewer is expected to evaluate.

Question 155

What should an organization consider when deciding whether to build or buy an AI capability?

1. Requirements, cost, integration, and maintenance
2. Only the model’s name
3. Only the user interface color
4. Only the number of prompts

Correct Answer: 1

Explanation

Build-versus-buy decisions should consider multiple business and technical factors. These can include functional requirements, total cost of ownership, integration needs, security, customization, maintenance responsibilities, scalability, vendor dependencies, and available internal expertise. A purchased solution may reduce development effort, while a custom solution may provide greater control for specialized requirements. The appropriate choice depends on the organization’s specific circumstances. A structured comparison helps ensure that important long-term operational and governance considerations are not overlooked.

Question 156

An agent repeatedly chooses an incorrect tool for a task. What should the team review?

1. Tool descriptions and agent instructions
2. Monitor brightness
3. Database storage size only
4. User keyboard settings

Correct Answer: 1

Explanation

Incorrect tool selection can result from unclear tool descriptions, ambiguous instructions, overlapping capabilities, or insufficient examples of when a tool should be used. The team should review how each tool is described and whether the agent’s instructions clearly define its purpose and limitations. Tool selection should then be tested using representative scenarios. Permissions should also restrict the potential impact of incorrect selections. Improving tool definitions and instructions can make agent behavior more predictable without simply adding unnecessary tools.

Question 157

Which practice helps protect sensitive information in AI application logs?

1. Logging every secret in plain text
2. Applying data minimization and access controls
3. Making logs publicly accessible
4. Disabling all security monitoring

Correct Answer: 2

Explanation

Data minimization means collecting and retaining only the information needed for legitimate operational purposes. AI logs may contain prompts, tool results, identifiers, or other information that could expose sensitive data if handled improperly. Organizations should limit sensitive information in logs where possible and apply appropriate access controls, retention policies, and protection mechanisms. Logging requirements should be balanced with the need for troubleshooting and auditing. Regular reviews can help identify unnecessary sensitive information being captured by the application.

Question 158

What is one advantage of using structured output from an AI model?

1. It can make downstream processing more predictable
2. It removes the need for validation
3. It guarantees factual accuracy
4. It eliminates security risks

Correct Answer: 1

Explanation

Structured output can make AI responses easier for applications to process consistently. For example, a system may require specific fields in a JSON structure so that downstream software can interpret the result without relying on unpredictable free-form text. However, structured output does not automatically guarantee that the values are correct or safe. Applications should validate required fields, data types, ranges, and business rules before using the output. Structured responses are therefore useful for integration but should remain subject to validation.

Question 159

Which capability is most useful for identifying unexpected changes in AI application behavior after deployment?

1. Static documentation
2. User interface themes
3. Production monitoring and telemetry
4. Larger prompt templates

Correct Answer: 3

Explanation

Production monitoring and telemetry provide information about how an AI application behaves after deployment. Teams can monitor metrics such as latency, errors, token usage, tool failures, request volumes, and evaluation signals. Comparing current behavior with expected baselines can help identify unexpected changes. Monitoring should also support appropriate alerting and investigation processes. Because AI systems can be affected by changing data, prompts, models, and user behavior, ongoing observation is important rather than relying only on testing performed before deployment.

Question 160

What should happen before deploying a major AI model or prompt change to production?

1. Skip testing to save time
2. Validate the change against defined evaluation criteria
3. Remove the previous version immediately
4. Disable production monitoring

Correct Answer: 2

Explanation

A major model or prompt change should be validated against defined evaluation criteria before production deployment. Testing can compare the new configuration with the existing version using representative datasets and scenarios. Teams should examine relevant measures such as accuracy, groundedness, safety, latency, cost, and task completion. If the change does not meet the required criteria, it should be revised or rejected. Versioning and rollback capabilities are also valuable because they allow teams to restore a previous configuration if unexpected issues occur.