salesforce agentforce Practice Questions & Answers (Set 16) | CodeWme
#1 Universal Containers wants to ensure that no customer Personal Identifiable Information (PII) is stored by the external Large Language Model (LLM) provider when using Agentforce. Which feature of the Einstein Trust Layer guarantees this? Select 1
✅ Answer: Zero Data Retention Policy
The Zero Data Retention policy ensures that the external LLM provider does not store or learn from the data sent to it for processing, maintaining data privacy.
#2 An Agentforce Specialist is creating a prompt template and needs to dynamically retrieve a list of related Opportunities that match a complex set of criteria (e.g., probability > 50% and close date this quarter). Which resource type should be used in Prompt Builder? Select 1
✅ Answer: Template-Triggered Prompt Flow
Template-Triggered Prompt Flows allow for complex logic, filtering, and data retrieval (like querying specific Opportunities) to be used as a resource within a prompt template.
#3 A Service Agent at Cloud Kicks needs to transfer a chat session to a human agent if the customer expresses frustration. The specialist wants to detect this frustration automatically. Which capability should be configured? Select 1
✅ Answer: Conversation Sentiment Analysis with a Conditional Action
Agentforce can use sentiment analysis to detect negative customer sentiment and trigger a conditional action (like transfer to human) based on the sentiment score.
#4 Universal Containers is preparing to deploy an Agentforce Service Agent to production. They want to verify that the agent respects the 'Do Not Contact' preference stored on the Contact record. How should this be enforced? Select 1
✅ Answer: Use a Flow action that checks the 'Do Not Contact' field and returns an error or alternative path if true.
For deterministic enforcement of business rules like 'Do Not Contact', using a Flow action with logic is more reliable than relying on soft instructions.
#5 What is the primary role of the 'Atlas Reasoning Engine' in Agentforce? Select 1
✅ Answer: To plan and execute a sequence of actions based on user intent and available tools.
The Atlas Reasoning Engine is responsible for the 'cognitive' process: understanding intent, selecting the right topic/action, planning the steps, and executing them.
#6 A developer wants to test a new prompt template for a Sales Agent to ensure it handles edge cases correctly before activating it. Where should this testing be performed? Select 1
✅ Answer: Prompt Builder Preview / Workbench
Prompt Builder includes a preview/test mode where developers can run the template against specific records to verify the output.
#7 Universal Containers uses Data Cloud to ground their agent. They want to ensure the agent only uses 'Active' product documentation when answering queries. Which Data Cloud feature enables this? Select 1
✅ Answer: Filter Criteria in the Search Index/Retriever
You can define filter criteria (e.g., Status = 'Active') in the Retriever definition or Search Index to restrict the scope of data available to the agent.
#8 Which permission set license is typically required for a user to configure and manage Agentforce Agents? Select 1
✅ Answer: Agentforce / Einstein Copilot Administrator
Specific permission set licenses (like Agentforce Admin or similar) are required to access the Agent Builder and configuration tools, beyond just standard Admin rights.
#9 Cloud Kicks wants their agent to be able to check inventory levels from an external ERP system. What is the recommended integration pattern? Select 1
✅ Answer: Create an External Service based on an API spec and expose it as an Agent Action.
External Services allow you to ingest an API specification (OpenAPI) and automatically create invokable actions that agents (and Flows) can use to call external systems.
#10 When defining a new Topic for an agent, what is the purpose of the 'Classification Description'? Select 1
✅ Answer: It is used by the reasoning engine to determine if the user's input matches this topic.
The Classification Description provides the semantic 'hook' for the LLM to understand what this topic covers and when to select it over others.
#11 A stakeholder wants to know how often the Service Agent is successfully resolving customer queries without human intervention. Which metric in Agentforce Analytics should be monitored? Select 1
✅ Answer: Deflection Rate / Resolution Rate
Deflection Rate (or Resolution Rate) specifically measures the percentage of interactions handled entirely by the agent without escalating to a human.
#12 Which component is necessary to allow an Agentforce Agent to send an email to a customer? Select 1
✅ Answer: A Flow Action (Send Email)
To perform an output action like sending an email, the agent typically calls a Flow that contains the 'Send Email' core action.
#13 Universal Containers wants to restrict the Agentforce Service Agent to only answer questions about 'Solar Panels' and ignore questions about 'Competitor Products'. Where should this instruction be placed? Select 1
✅ Answer: In the Agent/Topic Instructions
Instructions (System Prompt) are the correct place to define behavioral guardrails and scope, telling the agent what it should and should not talk about.
#14 What happens if an Agentforce Agent enters a loop where it keeps calling the same action repeatedly without progress? Select 1
✅ Answer: The Reasoning Engine detects the loop and terminates the interaction or asks for clarification.
The Atlas Reasoning Engine has built-in loop detection and safety mechanisms to prevent infinite loops, often stopping the execution and informing the user.
#15 To use a specific Salesforce Flow as an action for an Agent, what must be true about the Flow? Select 1
✅ Answer: It must be an Autolaunched Flow or Screen Flow with the 'Make available to Agent' setting enabled.
Flows must be explicitly configured (via checkbox or metadata setting) to be discoverable and executable by Agentforce agents.