Agentforce Specialist Practice Test

Answer choices

1. A. A stateless MCP tool call with no remote autonomous agent or collaboration context.
2. B. A new unrelated prompt for every message so no task state is shared.
3. C. An A2A stateful task that associates the agents' messages and artifacts throughout the handoff.
4. D. An Embedded Service session that treats the remote agent as a customer website channel.

Wrong-answer review

• A. A stateless MCP tool call with no remote autonomous agent or collaboration context.: MCP addresses AI access to external tools and data rather than stateful autonomous-agent collaboration.
• B. A new unrelated prompt for every message so no task state is shared.: Unrelated prompts would fragment the context and artifacts needed to complete one delegated job.
• D. An Embedded Service session that treats the remote agent as a customer website channel.: Embedded Service hosts customer conversations and does not define cross-platform agent delegation semantics.

Answer choices

1. A. Enter all 120 requests manually in production Preview and record observations.
2. B. Inspect one production session trace and treat it as representative of all paraphrases.
3. C. Run the cases against production because expected results prevent data changes.
4. D. Run a Test Suite in a sandbox with expected subagents, actions, and responses.

Answer choices

1. A. Keep the subagent visible and add a warning to its instructions.
2. B. Change the rule to accept any nonblank verification value.
3. C. Expose the subagent after any action returns, regardless of that action result.
4. D. Use an available-when condition that requires the verification variable to be true.

Answer choices

1. A. Flex, because every record-field workflow requires designer-defined arbitrary resources.
2. B. A general agent action with no association to the target Case field.
3. C. Field Generation, associated with the supported editable Case field.
4. D. A data library, because retrieval configuration determines where generated text is written.

Answer choices

1. A. Automated data stream and mapping setup, document chunking, a search index, and a dedicated retriever.
2. B. Only a file container; the team must separately code chunking, indexing, and retrieval.
3. C. A single full-document prompt that sends every complete article to the model on each request.
4. D. A topic-classification model that replaces the agent's configured topics and actions.

Answer choices

1. A. An agent-to-agent protocol that treats the pricing service as an autonomous collaborator.
2. B. A separate proprietary integration for every AI assistant that may use the service.
3. C. An Embedded Service chat deployment that exposes the pricing service as a website conversation.
4. D. An MCP server that securely processes the permitted data and tool requests through the standard interface.

Answer choices

1. A. Publish the Embedded Service deployment in the target org and verify customer access.
2. B. Retrieve the same metadata again because repeated retrieval automatically publishes the channel.
3. C. Replace the agent user because identities control whether Embedded Service is published.
4. D. Monitor the unavailable channel until production traffic causes it to publish itself.

Answer choices

1. A. Give the Atlas reasoning engine unrestricted account access and rely on it to avoid sensitive fields.
2. B. Describe the sensitive fields in natural-language subagent instructions so the model can decide when to read them.
3. C. Replace the flow with a prompt template that receives the entire account record.
4. D. Limit the information and access controls available to the agent while allowing the approved custom action to run.

Answer choices

1. A. Remove all template descriptions and rely only on creation timestamps.
2. B. Give each template a useful description that clearly states its purpose and differentiating use case.
3. C. Grant Prompt Template Manager access to every user so selection mistakes no longer matter.
4. D. Activate every template without preview so administrators never need to choose among drafts.

Answer choices

1. A. One basic data library with default retrieval and no transformation customization.
2. B. A data graph alone, because flattened JSON automatically performs exact and semantic document search.
3. C. Multiple duplicate libraries solely to justify an ensemble retriever.
4. D. Advanced Data 360 RAG with the required transformations and advanced hybrid search.