Section 9 · Agentforce · 8% of exam
9.1 Agentforce capabilities, use cases, security, and permissions
Exam objective
Describe Agentforce capabilities and suitable use cases, evaluate AI fit, and troubleshoot security or agent permissions.
Agentforce supports AI agents that can interpret requests and use configured topics, instructions, and actions to help complete tasks. It is appropriate when a conversational interaction adds value and the task can be bounded with reliable data and supported actions. A fixed, predictable process may be better served by a simpler automation or self-service design.
An agent’s behavior is constrained by its configuration and access context. Limit data and actions to what the agent needs, protect sensitive information, and plan handoff to a human for uncertainty or out-of-scope requests. Troubleshoot failures by checking agent permissions, record access, action configuration, and the user or agent execution context.
Additional concepts
Agent building blocks
An agent uses configured topics or work areas, instructions, and actions to handle requests. The reasoning process selects relevant capabilities and evaluates action results before responding. Clear descriptions and boundaries help the agent choose the right work; adding more actions alone does not fix an unclear topic or instruction.
Access and execution context
Agent behavior is limited by the permissions and data access of its execution context, together with the access configured for its actions. Troubleshoot in layers: whether the agent is available, whether it can select the action, whether the action is authorized, and whether the underlying user or agent context can access the record. Provide a human handoff for out-of-scope or uncertain requests.
A service agent can answer order-status questions using approved records. Confirm the agent can retrieve only the necessary order fields and that the action respects the intended customer access; test unauthorized and ambiguous requests as well as normal ones.
Do not give an agent broad data or action access simply to make a demo work. Test least-privilege behavior and escalation paths.
Key takeaways
- Use Agentforce where conversational reasoning and supported actions fit the task.
- Scope agent permissions, data access, and actions to the use case.
- Diagnose permission issues across the agent, action, record, and execution context.
