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

 

Question 41

Which Microsoft service can help developers build generative AI applications?

1. Microsoft Intune
2. Microsoft Defender
3. Microsoft Foundry
4. Microsoft Entra ID

Correct Answer: 3

Explanation

Microsoft Foundry provides capabilities for building and managing AI applications and agentic solutions. It supports working with models, agents, evaluations, and other components needed to develop AI-powered applications. Architects should select Foundry when the solution requires development and customization capabilities beyond simpler low-code scenarios. The choice should also consider security, data requirements, model selection, monitoring, cost, and operational needs. Understanding the development requirements helps determine whether Foundry or another Microsoft AI capability is appropriate for the proposed business solution.

Question 42

What does RAG primarily provide to an AI application?

1. Real-time network monitoring
2. Retrieved information for grounding responses
3. Password management
4. Application licensing

Correct Answer: 2

Explanation

Retrieval-augmented generation, or RAG, combines information retrieval with generative AI. Relevant information is retrieved from approved sources and provided to the model as context for generating a response. This approach can help an AI application use organization-specific or current information without relying entirely on the model’s pre-existing knowledge. Architects should evaluate retrieval quality, source permissions, data freshness, security, and response accuracy. RAG is particularly useful when business users need answers based on documents, knowledge bases, or other changing organizational information.

Question 43

Which design helps separate responsibilities among specialized AI agents?

1. Multi-agent architecture
2. Single database
3. Static website
4. Manual workflow

Correct Answer: 1

Explanation

A multi-agent architecture divides responsibilities among multiple specialized agents. For example, one agent may retrieve information, another may analyze it, and another may complete a business action. An orchestration approach coordinates these agents and manages how work moves between them. This design can simplify complex scenarios by giving each agent a focused responsibility. However, architects should carefully define communication, permissions, error handling, monitoring, and escalation. Multi-agent designs should be used when their additional complexity provides meaningful value for the business process.

Question 44

A company wants an AI agent to access a customer’s order information before responding. What should be considered first?

1. Screen resolution
2. Data access and authorization
3. Printer settings
4. Desktop wallpaper

Correct Answer: 2

Explanation

An agent accessing customer information must have appropriate data access and authorization controls. The architect should determine which data the agent requires, which users can access it, and how permissions are enforced when information is retrieved. The design should follow least-privilege principles so that the agent cannot access unnecessary information. Logging, monitoring, data protection, and secure integration should also be considered. AI capabilities do not bypass existing security requirements. Proper authorization helps prevent sensitive customer information from being exposed to unauthorized users.

Question 45

What is the main purpose of an AI evaluation process?

1. To replace business requirements
2. To measure solution quality against defined criteria
3. To remove monitoring
4. To increase storage capacity

Correct Answer: 2

Explanation

AI evaluation measures whether an AI solution meets predefined quality and business requirements. Evaluation criteria may include accuracy, relevance, groundedness, safety, task completion, latency, or other scenario-specific measurements. Architects should establish these criteria before deployment so that results can be compared consistently. Evaluation can also be repeated after changes to prompts, models, knowledge sources, or agent workflows. A structured evaluation process helps organizations identify weaknesses and make evidence-based improvements rather than relying only on informal user feedback.

Question 46

Which factor can influence the cost of a generative AI solution?

1. Token usage
2. Keyboard type
3. Monitor size
4. Office lighting

Correct Answer: 1

Explanation

Token usage can influence the cost of generative AI workloads because models process input and output tokens. Longer prompts, larger context, and lengthy generated responses can increase consumption. Architects should therefore consider prompt design, model selection, response limits, caching opportunities, and request patterns when estimating costs. Cost should be evaluated alongside quality and performance because a lower-cost model may not satisfy business requirements. Monitoring actual usage after deployment can help identify unexpected consumption and support ongoing optimization of the AI solution.

Question 47

What should an architect do before granting an agent access to a business system?

1. Remove authentication
2. Review required permissions
3. Disable logging
4. Give administrator access by default

Correct Answer: 2

Explanation

Before connecting an agent to a business system, the architect should determine exactly which permissions are required for the agent’s intended tasks. Access should follow the principle of least privilege and should not automatically provide administrative rights. The design should address authentication, authorization, secrets, monitoring, auditing, and failure handling. Limiting permissions reduces the potential impact if an agent behaves unexpectedly or is manipulated through malicious input. The architect should also periodically review permissions to ensure they remain appropriate as the business process changes.

Question 48

A business wants an agent to summarize long documents for employees. Which AI capability is most relevant?

1. Generative AI
2. Network routing
3. Device management
4. DNS resolution

Correct Answer: 1

Explanation

Generative AI is well suited to tasks such as summarizing documents because it can process provided content and produce a concise natural-language response. The architect should define requirements for summary length, accuracy, format, and handling of sensitive information. For business documents, grounding can help ensure that the response is based on the supplied source material. Evaluation should verify that important information is not incorrectly omitted or altered. Appropriate access controls are also necessary when documents contain confidential or restricted business information.

Question 49

Which approach can reduce unnecessary model costs for simple requests?

1. Always use the largest model
2. Disable monitoring
3. Route simple tasks to suitable lower-cost models
4. Increase prompt length

Correct Answer: 3

Explanation

Routing simple requests to suitable lower-cost models can reduce AI consumption costs while preserving the required quality. Not every request needs the capabilities of the most advanced model. Architects can establish routing rules based on task complexity, expected quality, latency, and other requirements. The selected model should still be evaluated to confirm that it meets business expectations. Monitoring usage and response quality helps determine whether the routing strategy is working effectively and whether particular workloads require a different model.

Question 50

What does human-in-the-loop mean in an AI workflow?

1. A person reviews or approves certain AI actions
2. AI operates without any oversight
3. Users are removed from the process
4. Data is automatically deleted

Correct Answer: 1

Explanation

Human-in-the-loop means that a person participates in a process by reviewing, approving, correcting, or otherwise supervising certain AI-generated results or actions. This approach can be useful when decisions have significant business, financial, legal, or operational consequences. The architect should identify which steps require human involvement and define clear approval or escalation procedures. Human review should complement, rather than replace, appropriate testing and technical safeguards. The amount of human oversight should reflect the risks and requirements of the particular business scenario.

Question 51

Which capability allows an AI agent to perform an operation in another application?

1. Action or tool integration
2. Screen brightness
3. File compression
4. Device wallpaper

Correct Answer: 1

Explanation

Action or tool integration allows an AI agent to interact with external applications or services and perform supported operations. This extends an agent beyond conversational responses and enables it to participate in business workflows. Architects should carefully define which tools are available, what permissions they require, what inputs they accept, and what outputs they return. Error handling and user confirmation may also be necessary for sensitive operations. Tool access should follow least-privilege principles so that the agent receives only the capabilities required for its intended purpose.

Question 52

Why is data classification important for AI solutions?

1. It determines monitor resolution
2. It helps identify how information should be protected
3. It changes model architecture automatically
4. It removes the need for access controls

Correct Answer: 2

Explanation

Data classification helps organizations identify the sensitivity and handling requirements of information used by AI solutions. Different types of data may require different protection, access, retention, and compliance controls. Architects should understand which information an agent can access and ensure that sensitive data is handled according to organizational policies. Classification can also help determine appropriate grounding sources and sharing boundaries. It does not replace access control, encryption, or other security measures, but it provides an important foundation for designing appropriate data protection.

Question 53

A company wants to prevent an agent from performing high-impact actions without user confirmation. What should be implemented?

1. Human approval
2. Unlimited permissions
3. Anonymous access
4. Disabled auditing

Correct Answer: 1

Explanation

Human approval can provide an important safeguard for high-impact AI actions. Instead of allowing an agent to automatically complete sensitive operations, the workflow can require a person to review relevant information and approve the action. The architect should identify which operations require confirmation based on business risk and organizational policy. Approval workflows should clearly present the proposed action and provide appropriate auditing. This approach helps maintain human oversight while still allowing agents to automate lower-risk activities that do not require individual approval.

Question 54

Which practice helps protect an AI agent from excessive permissions?

1. Least privilege
2. Shared administrator accounts
3. Anonymous access
4. Unlimited tool access

Correct Answer: 1

Explanation

The principle of least privilege requires an AI agent to receive only the permissions necessary to perform its intended tasks. This limits the potential impact of accidental behavior, compromised credentials, prompt manipulation, or incorrect actions. Architects should identify required resources and operations before assigning permissions and avoid granting broad administrative access by default. Permissions should also be reviewed regularly. Least privilege should be applied to the agent, connected applications, tools, data sources, and identities involved in the complete AI workflow.

Question 55

What is an important reason to monitor AI-generated outputs?

1. To detect quality or safety issues
2. To increase monitor brightness
3. To change file extensions
4. To remove user accounts

Correct Answer: 1

Explanation

Monitoring AI-generated outputs can help identify quality, safety, compliance, or operational problems. AI systems may occasionally generate inaccurate, irrelevant, biased, or otherwise inappropriate responses. Monitoring allows organizations to detect patterns and investigate issues after deployment. Architects should define appropriate metrics and establish processes for handling problematic outputs. Monitoring should also respect privacy and data protection requirements. Combining monitoring with evaluation and user feedback provides a stronger foundation for improving the AI solution and maintaining expected performance over time.

Question 56

A developer needs an AI system to return responses in a fixed JSON structure. What should be included in the prompt?

1. Output format instructions
2. Network credentials
3. Printer configuration
4. Screen settings

Correct Answer: 1

Explanation

Output format instructions tell the model how the response should be structured. If an application requires JSON, the prompt can specify the expected fields, structure, and formatting requirements. Structured output can make AI responses easier for downstream applications to process. However, architects should not assume that prompt instructions alone guarantee valid output. The application should validate the response before using it in an automated process. Clear schemas, validation, error handling, and testing can improve reliability when AI output is consumed programmatically.

Question 57

Which factor should be considered when selecting between cloud and local AI processing?

1. Data, compliance, latency, and cost requirements
2. Keyboard color
3. Office furniture
4. Number of printers

Correct Answer: 1

Explanation

The choice between cloud and local AI processing should consider business and technical requirements. Important factors can include data sensitivity, regulatory requirements, network connectivity, latency, scalability, model availability, infrastructure costs, and operational capabilities. Cloud services may provide scalability and managed capabilities, while local processing may be appropriate for specific privacy, latency, or connectivity requirements. Architects should evaluate the complete workload rather than assuming one deployment model is universally appropriate. The selected architecture should satisfy security, performance, cost, and business requirements.

Question 58

What does AI observability help organizations understand?

1. Agent and model behavior in operation
2. Employee vacation schedules
3. Printer toner levels
4. Office temperature

Correct Answer: 1

Explanation

AI observability provides insight into how AI applications, agents, models, and supporting components behave during operation. It can include information about requests, responses, latency, errors, tool calls, resource consumption, and other relevant signals. This information helps teams investigate failures, understand performance, and identify opportunities for optimization. Architects should determine which telemetry is necessary for the business scenario and apply appropriate privacy and security controls. Observability is particularly valuable for agentic systems because multiple steps or tools may contribute to a single outcome.

Question 59

An agent repeatedly gives incorrect answers because its knowledge source contains outdated information. What should be improved?

1. Data freshness
2. Screen resolution
3. Network printer settings
4. User interface color

Correct Answer: 1

Explanation

Data freshness is important when an AI solution relies on changing business information. If the grounding source is outdated, the agent may produce responses that were once accurate but are no longer valid. The architect should establish processes for updating, validating, and retiring knowledge sources. The required refresh frequency depends on the business scenario. Highly dynamic information may need more frequent updates than relatively stable documentation. Monitoring source quality and retrieval behavior can also help identify when outdated information is affecting agent responses.

Question 60

What is the purpose of a responsible AI review before deployment?

1. To assess risks and responsible use
2. To remove security controls
3. To avoid testing
4. To increase application colors

Correct Answer: 1

Explanation

A responsible AI review helps identify risks associated with deploying an AI solution and verifies that its intended use aligns with organizational principles and requirements. Areas of consideration can include safety, reliability, fairness, transparency, privacy, security, accountability, and appropriate human oversight. The review should consider both the AI model and the complete business workflow in which it operates. Conducting the review before deployment can identify issues that require architectural, technical, process, or governance changes before the solution reaches production.