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

 

Question 341

What is the primary purpose of agent memory?

1. Manage information that an agent needs to retain during or across tasks
2. Increase API quotas
3. Replace authorization
4. Encrypt databases

Correct Answer: 1

Explanation

Agent memory allows an AI agent to retain useful information needed to continue tasks or provide continuity across interactions. Depending on the architecture, memory can include short-term conversation context, task state, or approved long-term information. Memory should be designed around the actual business requirement and should follow privacy, retention, and access-control policies. Storing unnecessary information can create additional risk. Architects should also determine what information should be forgotten, summarized, or retained and for how long.

Question 342

Which type of memory is most appropriate for information needed only during the current conversation?

1. Long-term memory
2. Short-term memory
3. Permanent database storage
4. Archived storage

Correct Answer: 2

Explanation

Short-term memory is designed for information needed during an active interaction or task. It can contain recent conversation messages, temporary decisions, intermediate results, and task context. This information helps the agent maintain continuity without repeatedly requesting the same details. Long-term memory is more appropriate when information must persist across separate sessions. The architecture should define retention behavior clearly because not every piece of conversation data should be stored permanently. Privacy and data-minimization requirements should also guide memory design.

Question 343

A customer asks an agent to remember a preference for future conversations. What capability is relevant?

1. Long-term memory
2. API throttling
3. Prompt injection
4. Model routing

Correct Answer: 1

Explanation

Long-term memory can store approved information that should remain available across separate conversations. In this example, a customer preference may be stored in an appropriate persistent system and retrieved during future interactions. The organization should determine what information is appropriate to retain and establish consent, privacy, access, and retention controls where applicable. Long-term memory should not become an unrestricted storage mechanism for all conversation content. Only information with a legitimate business purpose should generally be retained.

Question 344

Which capability helps an agent maintain the current progress of a multi-step workflow?

1. Task state
2. Model catalog
3. Data residency
4. Embedding refresh

Correct Answer: 1

Explanation

Task state records information needed to understand the current position of an agent within a workflow. It can identify completed actions, pending steps, retrieved values, decisions, and other information required to continue processing. Maintaining task state is particularly useful when workflows involve multiple tools or agents. It can prevent duplicated actions and help an agent resume work after an interruption. Task state should be protected appropriately because it may contain business information or other sensitive details related to the workflow.

Question 345

An agent needs to call three independent APIs. Which approach may reduce total execution time?

1. Sequential execution
2. Parallel execution
3. Repeated retries
4. Manual approval for every call

Correct Answer: 2

Explanation

Parallel execution can reduce total execution time when multiple API calls are independent and do not depend on each other’s results. Instead of waiting for one call to complete before starting another, the orchestration layer can initiate the operations concurrently. The architecture must still consider API quotas, concurrency limits, failures, and resource capacity. If one operation depends on another, sequential execution remains necessary. Properly identifying dependencies allows teams to improve performance without compromising workflow correctness.

Question 346

What should an AI agent use to determine whether a tool call is allowed?

1. Authorization and permission controls
2. Model temperature
3. Context window size
4. Document chunk size

Correct Answer: 1

Explanation

Authorization and permission controls should determine whether an agent is allowed to execute a particular tool operation. Natural-language instructions can describe boundaries, but technical controls should enforce them. Permissions should follow least privilege so the agent receives only the access needed for its assigned tasks. Tool-level authorization can prevent an agent from performing sensitive operations even if a user or malicious input attempts to instruct it to do so. Logging and monitoring should also record important tool actions for accountability.

Question 347

Why is tool input validation important?

1. It prevents malformed or unsafe values from reaching backend systems
2. It guarantees perfect model reasoning
3. It removes the need for authorization
4. It increases model context automatically

Correct Answer: 1

Explanation

Tool input validation checks whether values provided by an agent meet expected formats, ranges, types, and business rules before the tool executes. This can prevent malformed requests, invalid identifiers, unexpected commands, or potentially unsafe values from reaching backend systems. Validation should be implemented at the tool or service boundary rather than relying only on model instructions. Authorization remains necessary because a correctly formatted request may still be unauthorized. Together, validation and authorization provide stronger protection for automated actions.

Question 348

Which technique can help reduce the number of tokens sent in a long-running conversation?

1. Context summarization
2. Unlimited history
3. Adding duplicate documents
4. Increasing prompt repetition

Correct Answer: 1

Explanation

Context summarization can condense older conversation information while retaining details that remain important to the current task. This reduces the amount of conversation history that must be included in future model requests and can help manage context-window limits and token costs. The summary should preserve important facts, decisions, constraints, and unresolved tasks. Teams should evaluate summarization quality because losing a critical detail can cause incorrect behavior. Summarization can be combined with selective retrieval and task-state management.

Question 349

A company wants to prevent an agent from using an outdated policy document. Which control is most useful?

1. Document freshness management
2. Higher model temperature
3. Larger context windows
4. Unlimited retrieval

Correct Answer: 1

Explanation

Document freshness management helps ensure that outdated or superseded information is not used as authoritative knowledge. Organizations can use effective dates, document status, ownership, review schedules, and automated ingestion processes to keep knowledge repositories current. Outdated documents can be removed, archived, or excluded from retrieval when appropriate. This is especially important for policies, pricing, procedures, and other information that changes frequently. Retrieval quality should be monitored because a technically functioning search system can still produce incorrect answers from stale content.

Question 350

What is a key purpose of reranking in a RAG pipeline?

1. Put the most relevant retrieved content first
2. Increase user permissions
3. Replace all source documents
4. Disable semantic search

Correct Answer: 1

Explanation

Reranking evaluates retrieved candidates and reorders them according to their relevance to the user’s request. An initial retrieval step may return several potentially useful documents, but some may be more relevant than others. Reranking helps prioritize stronger evidence before the context is provided to the language model. This can improve answer quality and reduce irrelevant context. Reranking should work alongside appropriate indexing, query processing, access controls, and evaluation rather than being treated as a replacement for the retrieval system.

Question 351

Which search method combines semantic and keyword-based retrieval?

1. Hybrid search
2. Static search
3. Manual search
4. Random search

Correct Answer: 1

Explanation

Hybrid search combines keyword-based matching with semantic retrieval methods. Keyword matching can be effective for exact names, identifiers, codes, and terminology, while semantic retrieval can identify content with related meaning even when the wording differs. Combining the approaches can improve retrieval quality across diverse enterprise content. Additional filtering and reranking may further improve results. The effectiveness of hybrid search should be measured using representative queries because the best retrieval configuration depends on the organization’s data, terminology, and business requirements.

Question 352

What should an organization consider when selecting chunk sizes for RAG documents?

1. Content structure and retrieval requirements
2. Only the model’s name
3. User password length
4. Number of employees

Correct Answer: 1

Explanation

Chunk size should be selected based on the structure of the source content and the type of information users need to retrieve. Very large chunks may contain unnecessary information and consume more context, while very small chunks may separate related concepts and reduce useful context. Appropriate overlap can help preserve continuity between sections. Teams should test different approaches using representative queries and retrieval evaluations. Metadata, document structure, and expected answer requirements should also influence the chunking strategy.

Question 353

Which feature can help an agent retrieve only records the current user is permitted to access?

1. Access-aware retrieval
2. Temperature adjustment
3. Prompt repetition
4. Model compression

Correct Answer: 1

Explanation

Access-aware retrieval applies user or service permissions when selecting information for an agent. The retrieval layer should ensure that documents or records are returned only when the requesting identity is authorized to access them. This is particularly important in enterprise RAG systems where a common knowledge index may contain information belonging to different departments or users. Authorization should be enforced technically rather than relying on the model to hide unauthorized content. Access-aware retrieval should also be tested with users having different permission levels.

Question 354

An agent needs current inventory information before recommending products. Which approach is most appropriate?

1. Retrieve current inventory through an approved data connection
2. Rely only on model training
3. Store inventory permanently in the system prompt
4. Ignore inventory changes

Correct Answer: 1

Explanation

Current inventory should be retrieved from an approved business data source when the information changes frequently. A connector, API, or other authorized data connection can provide the latest inventory status at runtime. This avoids relying on stale information embedded in model training or static instructions. The integration should include authentication, authorization, validation, error handling, and monitoring. If inventory information is unavailable, the agent should follow a defined fallback or clarification process rather than inventing availability.

Question 355

Which architecture pattern is appropriate when specialized agents collaborate on different parts of a business process?

1. Multi-agent architecture
2. Single static prompt
3. Database backup
4. Token compression

Correct Answer: 1

Explanation

A multi-agent architecture divides responsibilities among specialized agents that collaborate to complete a broader workflow. For example, one agent may handle customer identification, another may analyze billing information, and another may manage service requests. An orchestration layer can coordinate these agents and manage handoffs. Each agent should have clearly defined responsibilities and appropriate permissions. Multi-agent designs can provide specialization but also introduce additional complexity, so organizations should use them when the business workflow genuinely benefits from separated capabilities.

Question 356

Why should an organization define agent ownership?

1. To establish responsibility for maintenance, governance, and operational decisions
2. To give everyone administrator access
3. To eliminate monitoring
4. To avoid documentation

Correct Answer: 1

Explanation

Defined ownership establishes who is responsible for maintaining an AI agent and responding to operational or governance issues. Owners may oversee configuration changes, access reviews, evaluations, incident handling, knowledge updates, and lifecycle decisions. Clear ownership prevents situations where an agent remains active without anyone responsible for its security or performance. Ownership should be documented and aligned with organizational processes. In larger environments, technical, business, security, and compliance responsibilities may be shared across different roles.

Question 357

Which practice helps detect whether an AI solution’s performance is declining over time?

1. Drift monitoring
2. Removing evaluation datasets
3. Disabling telemetry
4. Increasing permissions

Correct Answer: 1

Explanation

Drift monitoring helps identify changes in data, user behavior, or system performance that may cause an AI solution to behave differently over time. Changes in the underlying business environment can reduce model or retrieval effectiveness even when the application itself has not been modified. Monitoring can track relevant quality, usage, and operational metrics against established baselines. When drift is detected, teams can investigate data, prompts, retrieval, models, or business processes and determine whether retraining, reconfiguration, or other action is required.

Question 358

What is a key benefit of using an API gateway with multiple AI services?

1. Centralized traffic and access management
2. Automatic removal of all security risks
3. Elimination of model evaluation
4. Unlimited backend access

Correct Answer: 1

Explanation

An API gateway can provide a centralized point for managing traffic between applications and multiple backend services. Depending on the platform, it can support authentication, authorization, routing, rate limiting, monitoring, and policy enforcement. This can simplify governance when an AI solution interacts with several APIs or services. The gateway does not automatically eliminate security risks, and backend services should continue to enforce their own authorization where appropriate. Centralized management should complement, not replace, service-level security controls.

Question 359

A model change reduces cost but also decreases response quality below the business requirement. What should the team do?

1. Deploy it immediately
2. Ignore the quality reduction
3. Reassess the model against the required acceptance criteria
4. Remove all evaluation metrics

Correct Answer: 3

Explanation

A lower-cost model should not be adopted solely because it reduces operating expenses if its quality falls below defined business requirements. The team should compare the model using representative workloads and established acceptance criteria. The evaluation should consider accuracy, groundedness, safety, latency, and cost as appropriate for the use case. If quality is insufficient, the team can consider routing only suitable tasks to the smaller model or using a more capable model for important scenarios. Architecture decisions should reflect both business value and technical performance.

Question 360

What should happen to unnecessary agent permissions after a system is retired?

1. They should remain permanently active
2. They should be reviewed and revoked
3. They should be shared with users
4. They should be copied to another system automatically

Correct Answer: 2

Explanation

When an agent is retired, associated permissions, credentials, service identities, connectors, and access paths should be reviewed and revoked when they are no longer required. Leaving unused permissions active creates unnecessary security exposure and can result in orphaned access. Retirement should also include dependency review, documentation, data-retention decisions, and appropriate audit records. If another system replaces the retired agent, its permissions should be independently designed according to least privilege rather than simply inheriting all of the previous agent’s access.