AB-100: Agentic AI Business Solutions Architect (AB-100) - Microsoft Actual Exam Questions

The agent must use only open-source pre-trained models to minimize licensing costs

The agent must use only open-source pre-trained models to minimize licensing costs is incorrect because this is a cost/procurement requirement, not a functional requirement related to risk management, accuracy, or trust in the decision-making process. Option

The agent must provide its output to the human radiologist through a mobile application only

The agent must provide its output to the human radiologist through a mobile application only is incorrect because this is a user interface requirement and does not address the core issue of transparency or auditability needed for high-stakes decisions. Option

The agent must be designed with Explainable AI (XAI) capabilities to clearly show which visual features in the scan drove the urgency score right

The agent must be designed with Explainable AI (XAI) capabilities to clearly show which visual features in the scan drove the urgency score is correct because in high-stakes decision- making scenarios, the Explainability of the AI system is paramount for mitigating risk, enabling human oversight, and meeting regulatory audit requirements. XAI allows the radiologist to understand why a recommendation was made, ensuring they can validate the AI's logic before making a final critical decision. Option

The agent must be trained using data that is exclusively anonymized and synthetic to simplify compliance

The agent must be trained using data that is exclusively anonymized and synthetic to simplify compliance is incorrect because while synthetic data simplifies some compliance issues, it often severely degrades model accuracy and performance in real-world scenarios, which is unacceptable for a critical diagnostic tool.

once emitted, the data is funneled into the Azure Monitor pipeline. This step is crucial for: Ingestion: Gathering data from diverse sources. Standardization: Processing, sampling, and preparing the raw data to ensure consistency and adherence to the Azure Monitor schema. Step

The standardized metrics are then written to the native time-series database within Azure Monitor. This highly optimized storage layer: Persists the data for reporting and historical analysis (e.g., typically 93 days). Enables fast retrieval and aggregation, necessary for real-time alerting. Step

The final step involves accessing the stored data to drive operational decisions: Visualization: Tools like Dashboards and Metric Explorer query the data for graphical representation. Alerting: Alert Rules continuously monitor the stored data against defined thresholds, triggering notifications upon breach. Analysis: Tools like Grafana (using PromQL) perform deeper analysis.

Source (Emission)

Ingestion Pipeline(Collection & Standardization)

Storage

Consumption (Usage)

Metrics are stored in the native Azure Monitor Metrics platform (platform/custom or Prometheus)

Metrics are stored in the native Azure Monitor Metrics platform (platform/custom or Prometheus) is correct because

Metrics are emitted from Azure resources, Azure Monitor agents, Application Insights, REST APIs, or Kubernetes clusters

Metrics are emitted from Azure resources, Azure Monitor agents, Application Insights, REST APIs, or Kubernetes clusters. The flow begins at the source, where the raw performance data (metrics) is generated. This includes: Platform Metrics: Automatically emitted by Azure services. Custom Metrics: Generated by applications or pushed via the REST API. Agent Data: Collected by the Azure Monitor Agent from the OS or application components. Step

Metrics are consumed by tools such as Workbooks, Dashboards, Metric Explorer, Alerts, PromQL, or Grafana

Metrics are consumed by tools such as Workbooks, Dashboards, Metric Explorer, Alerts, PromQL, or Grafana is correct because

Metrics are collected and standardized into the Azure Monitor metrics pipeline

Metrics are collected and standardized into the Azure Monitor metrics pipeline is correct because

page vendor contract

page vendor contract is incorrect because this tests the LLM's generative quality, not the resilience or performance of the external tool call (the Power Automate flow), which is the focus of this question. Option

Validate that the agent can gracefully handle the scenario where the external vendor API returns a 503 Service Unavailable error, logging the failure without crashing the conversation right

Validate that the agent can gracefully handle the scenario where the external vendor API returns a 503 Service Unavailable error, logging the failure without crashing the conversation is correct because testing for resilience and error handling of external dependencies (APIs, tools) is critical. A robust agent must use fallback logic to manage external service failures, ensuring the conversation doesn't break. Option

Validate the latency by ensuring the agent executes the Power Automate flow and receives the vendor details within a defined Service Level Objective (SLO), such as 5 seconds right

Validate the latency by ensuring the agent executes the Power Automate flow and receives the vendor details within a defined Service Level Objective (SLO), such as 5 seconds is correct because external tool calls significantly impact the overall Latency of the agent's response. Validation must ensure the end-to-end action meets the established technical performance objectives (SLOs) required for a positive user experience. Option

Validate the agent's ability to accurately summarize a

Validate the agent's ability to accurately summarize a

Validate the model's overall F1 score by comparing its output to a human-curated ground truth of extracted vendor names

Validate the model's overall F1 score by comparing its output to a human-curated ground truth of extracted vendor names is incorrect because this tests the model's accuracy and is typically part of model validation, not the end-to-end integration and resilience testing of the tool call itself. Option

Validate the agent's intent classification by confirming it correctly identifies the user's request as "Invoice Processing"

Validate the agent's intent classification by confirming it correctly identifies the user's request as "Invoice Processing" is incorrect because this tests the agent's NLU (Natural Language Understanding) capability, not its ability to handle external tool execution or failures.

Yes right

No

Microsoft Entra ID: This is the foundational cloud-based identity and access management service that centrally manages user identities, groups, and permissions. For Azure resources, it serves as the ultimate authority for issuing and validating identities. In the context of Managed Identities, Microsoft Entra ID is where the identity itself is registered and managed. Step

Managed Identity: A Managed Identity is an identity registered with Microsoft Entra ID that Azure resources (like the Container App hosting the .NET OpenAI MCP Agent) can use. It eliminates the need for developers to manage credentials directly. Microsoft Entra ID provisions and manages this identity. The Agent's hosting environment then uses this Managed Identity to authenticate to other Azure services (like Azure OpenAI) or external services that trust Entra ID. Step

.NET OpenAI MCP Agent: The .NET OpenAI MCP Agent, deployed within an Azure Container App, is configured to use the Managed Identity assigned to its hosting environment. This allows the Agent to automatically obtain access tokens from Microsoft Entra ID (via the Managed Identity) and present them when making calls to secured services, ensuring that its requests are authenticated and authorized without hardcoding secrets.

Managed Identity

Managed Identity Step 3 →

.NET OpenAI MCP Agent

NET OpenAI MCP Agent Step

Microsoft Entra ID

Microsoft Entra ID Step 2 →