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

 

Question 201

What is a key purpose of defining an AI agent’s objectives?

1. To identify the results the agent should achieve
2. To increase storage capacity
3. To remove access controls
4. To disable monitoring

Correct Answer: 4

Explanation

Clear objectives help define what an AI agent is expected to accomplish. They provide direction for selecting tools, knowledge sources, workflows, and evaluation criteria. For example, an agent may be designed to resolve routine support requests or retrieve approved information for employees. Objectives should be specific enough to guide behavior without unnecessarily expanding the agent’s scope. They also provide a foundation for measuring success after deployment. A well-defined objective helps prevent the agent from performing unrelated tasks or taking unnecessary actions.

Question 202

Which feature allows an AI application to retrieve information from a vector database based on semantic similarity?

1. Role-based access control
2. Vector search
3. Password rotation
4. File compression

Correct Answer: 1

Explanation

Vector search retrieves information by comparing numerical representations of content, commonly called embeddings. Instead of requiring exact keyword matches, it can identify documents or passages that are semantically related to a user’s query. This makes vector search useful in RAG architectures where relevant knowledge must be retrieved before generating an answer. Retrieval quality depends on factors such as embedding quality, indexing, chunking, and query design. Access controls should still be applied so that users and agents retrieve only information they are authorized to access.

Question 203

An agent needs to complete three tasks where the second task depends on the result of the first. What workflow is appropriate?

1. Parallel execution
2. Random execution
3. Sequential execution
4. Uncontrolled execution

Correct Answer: 3

Explanation

Sequential execution is appropriate when one task depends on the result of a previous task. For example, an agent may first retrieve an order, then validate its status, and finally perform an approved action based on that status. Executing dependent steps in parallel could produce incorrect or incomplete results because required information may not yet be available. The workflow should maintain task state and define how errors are handled at each step. Independent tasks can still be parallelized when dependencies do not exist.

Question 204

What should an AI solution use to protect sensitive credentials?

1. A user prompt
2. Secure secrets management
3. A public document
4. A model response

Correct Answer: 2

Explanation

Sensitive credentials such as API keys, passwords, and tokens should be stored using secure secrets-management mechanisms. Credentials should not be placed directly in prompts, source code, public documents, or user-visible responses. Secure storage can support controlled access, encryption, auditing, and credential rotation. The application should also use least-privilege permissions so that a credential grants only the access required for its purpose. Protecting secrets is an important part of securing AI applications that connect to external APIs and business systems.

Question 205

Which approach can improve an agent’s ability to answer questions using current company information?

1. Increasing model temperature
2. Removing all knowledge sources
3. Using only historical prompts
4. Retrieving information from approved current sources

Correct Answer: 4

Explanation

Current company information should generally be retrieved from approved sources rather than relying only on information contained in the model or old prompts. Retrieval allows an agent to access updated policies, records, documentation, or other business information when needed. The sources should be authoritative, relevant, and properly secured. Refresh processes are also important because outdated indexes can lead to incorrect answers. Retrieval should be evaluated regularly to verify that the system returns information that is both current and appropriate for the user’s request.

Question 206

What is a benefit of using specialized agents in a multi-agent architecture?

1. Each agent can focus on a defined responsibility
2. Every agent receives unrestricted permissions
3. All agents must use identical instructions
4. Security controls become unnecessary

Correct Answer: 3

Explanation

Specialized agents can focus on specific responsibilities within a larger workflow. For example, one agent may handle customer inquiries while another specializes in billing or technical support. Specialization can simplify instructions and allow each component to use tools and knowledge relevant to its role. However, multi-agent architectures require clear orchestration, communication, permissions, and error handling. Specialization does not remove the need for governance or security. Each agent should receive only the access necessary for its assigned responsibility.

Question 207

Why should an AI application validate tool inputs before execution?

1. To prevent invalid or unsafe values from reaching business systems
2. To increase token usage
3. To eliminate authentication
4. To make every request autonomous

Correct Answer: 1

Explanation

Tool input validation helps prevent malformed, incomplete, unexpected, or unsafe values from being sent to connected systems. An AI model may generate incorrect parameters even when the user’s request is legitimate. Validation can check data types, required fields, permitted ranges, identifiers, and business rules before a tool executes. This is especially important for actions that modify records or trigger financial or operational processes. Input validation should complement authorization and other technical controls rather than relying on the model to produce valid values.

Question 208

Which factor is important when designing an AI application for high availability?

1. Prompt wording only
2. Failure handling and service redundancy
3. Font selection
4. Document formatting

Correct Answer: 2

Explanation

High-availability AI applications should consider what happens when individual services or dependencies become unavailable. Failure handling, redundancy, fallback services, retries, health monitoring, and appropriate workload distribution can reduce the impact of service interruptions. External model providers, databases, retrieval systems, and APIs may each introduce availability risks. The architecture should identify critical dependencies and define recovery procedures. Testing failure scenarios is also important because a system that works under normal conditions may behave differently when one of its supporting components becomes unavailable.

Question 209

What does a Model Context Protocol server commonly expose to an AI application?

1. Hardware drivers
2. User passwords
3. Tools or resources through a standardized interface
4. Operating system updates

Correct Answer: 3

Explanation

A Model Context Protocol server can expose approved tools, resources, or other capabilities through a standardized interface that an AI application can interact with. This can simplify integration by providing a consistent mechanism for discovering and using external capabilities. However, availability of a tool does not automatically grant authorization to perform every operation it supports. Authentication, permissions, and data boundaries should still be enforced. Organizations should also evaluate the trustworthiness and security of MCP-connected resources before allowing agents to use them.

Question 210

What can prompt templates provide for enterprise AI applications?

1. Consistent instructions for recurring tasks
2. Unlimited database permissions
3. Automatic compliance approval
4. Permanent model accuracy

Correct Answer: 4

Explanation

Prompt templates can provide standardized instructions for recurring AI tasks. For example, an organization may use a template that defines how customer summaries should be structured or how an agent should respond when information is unavailable. Templates can improve consistency and make prompts easier to manage and version. However, they do not replace security controls, evaluation, or business validation. Templates should be tested with representative inputs because changes in instructions can affect model behavior, output quality, token consumption, and downstream workflows.

Question 211

Which practice helps identify unexpected increases in AI operating costs?

1. Cost and usage monitoring
2. Removing telemetry
3. Increasing model size automatically
4. Disabling usage records

Correct Answer: 1

Explanation

Cost and usage monitoring can identify unexpected increases in AI spending. Teams can track factors such as token consumption, request volume, model usage, tool calls, and other service charges. Sudden increases may indicate higher traffic, inefficient prompts, unnecessary model calls, or incorrect routing. Monitoring allows teams to investigate these changes and apply appropriate controls, such as caching, model routing, quotas, or prompt optimization. Cost monitoring should be combined with quality measurements so that cost reductions do not unnecessarily reduce required AI performance.

Question 212

What should an agent do when a user requests an action outside its defined scope?

1. Execute it anyway
2. Refuse, redirect, or escalate according to the workflow
3. Grant itself additional permissions
4. Ignore authorization requirements

Correct Answer: 2

Explanation

An agent should not automatically perform an action outside its defined scope. Depending on the business process, it can explain that the request is outside its capabilities, redirect the user to an appropriate process, or escalate the request to an authorized person. Scope boundaries should also be enforced through technical permissions. Clear handling of out-of-scope requests helps prevent unintended actions and makes agent behavior more predictable. Testing should include requests that intentionally fall outside the agent’s approved responsibilities.

Question 213

Which approach can improve the quality of a RAG knowledge base?

1. Adding unrelated documents
2. Removing document metadata
3. Using authoritative and relevant source content
4. Ignoring document freshness

Correct Answer: 4

Explanation

A RAG system depends heavily on the quality of its underlying knowledge sources. Authoritative, relevant, accurate, and appropriately maintained documents provide better evidence for generated responses. Unrelated or outdated content can reduce retrieval quality and increase the likelihood of unsupported answers. Organizations should establish ownership for important sources and define processes for updates, indexing, and removal of obsolete information. Evaluation datasets should include realistic queries so teams can verify that the retrieval system consistently finds useful information from the knowledge base.

Question 214

Why might an organization choose RAG instead of putting frequently changing information directly into a model prompt?

1. RAG can retrieve updated information at request time
2. RAG removes the need for data security
3. RAG guarantees perfect answers
4. RAG eliminates all model costs

Correct Answer: 3

Explanation

RAG can retrieve current information from approved knowledge sources when a user submits a request. This is useful when business information changes frequently, because the application does not need to rewrite a large static prompt every time source content changes. The retrieved information can then be supplied to the model as contextual evidence. RAG still requires good indexing, access control, source governance, and evaluation. It does not automatically guarantee accurate answers if the underlying data is incomplete, outdated, or poorly retrieved.

Question 215

What is an important consideration when an AI agent uses a third-party connector?

1. The connector should be authorized and appropriate for the required data
2. The connector should have unrestricted access
3. Authentication should be disabled
4. All retrieved data should be public

Correct Answer: 2

Explanation

Third-party connectors can provide useful integration capabilities, but they should be evaluated for authorization, security, data handling, reliability, and business suitability. The connector should have only the permissions necessary for the agent’s tasks. Organizations should understand what data is sent through the connector and where it may be processed or stored. Connector credentials should also be protected. Monitoring connector activity can help identify unexpected usage, failures, or unauthorized access attempts after the integration is deployed.

Question 216

Which testing approach uses realistic examples that represent expected production usage?

1. Representative scenario testing
2. Random hardware testing
3. Interface color testing
4. Unrelated document testing

Correct Answer: 1

Explanation

Representative scenario testing uses examples that closely reflect how users are expected to interact with the AI solution in production. These scenarios should cover common requests as well as important edge cases and failure conditions. Testing representative inputs can reveal issues that may not appear in simple demonstrations. Teams should define expected outcomes or evaluation criteria where possible. This approach is useful for testing prompts, models, retrieval, tool selection, safety controls, and overall workflow behavior before and after deployment.

Question 217

What is the main purpose of an AI incident-response process?

1. To define how problems involving an AI system are identified and handled
2. To increase model temperature
3. To remove audit records
4. To prevent all user feedback

Correct Answer: 4

Explanation

An AI incident-response process defines how organizations identify, investigate, contain, and resolve problems involving an AI system. Incidents may include security issues, unauthorized data access, unsafe outputs, service failures, or unexpected agent actions. A documented process establishes responsibilities, escalation paths, communication procedures, and recovery steps. Monitoring and audit information can support investigation. Incident response should also feed lessons back into system improvements, testing, and governance so that recurring issues can be addressed systematically.

Question 218

Which design can reduce unnecessary exposure of sensitive data to an AI model?

1. Sending every available database field
2. Applying data minimization
3. Removing all authorization
4. Copying entire databases into prompts

Correct Answer: 2

Explanation

Data minimization limits the information provided to an AI model to what is actually required for the task. For example, an agent answering an order-status question may need an order number and status but not a customer’s complete profile. Reducing unnecessary data exposure can lower privacy and security risks and may also reduce prompt size and processing costs. Data minimization should be combined with authorization, classification, retention, and other appropriate controls to ensure that sensitive information is handled correctly.

Question 219

What is one purpose of using evaluation gates in an AI deployment pipeline?

1. To automatically increase model size
2. To bypass testing
3. To require defined quality checks before promotion
4. To disable version control

Correct Answer: 3

Explanation

Evaluation gates can require an AI configuration to satisfy predefined quality or safety criteria before it is promoted to another environment. For example, a new prompt or model version may need to meet minimum accuracy, groundedness, safety, latency, or cost thresholds. This creates a controlled release process and reduces the chance that an untested change will reach production. Evaluation gates should be based on representative datasets and meaningful business requirements. Failed gates should trigger investigation rather than being automatically bypassed.

Question 220

Which action is appropriate when retiring an AI agent?

1. Remove it immediately without documentation
2. Disable all monitoring first
3. Continue granting access indefinitely
4. Revoke access and follow the organization’s retirement process

Correct Answer: 1

Explanation

When an AI agent is retired, its access to models, tools, data sources, credentials, and business systems should be reviewed and removed according to the organization’s lifecycle process. Relevant documentation, audit records, and retention requirements should also be addressed. Simply deleting an application without considering connected identities or credentials can leave unnecessary access behind. A controlled retirement process helps ensure that unused resources are decommissioned securely and that stakeholders understand the status of the former AI solution.