Agentforce topics, actions, and reasoning
Key Takeaway
For each message, an agent runs fixed logic first and reasons with a large language model (LLM) second. An action that must run every time, before any reasoning, goes in the reasoning instructions.
Agentforce agents are conversational AI assistants that automate routine tasks and help with complex ones. An agent is built from subagents, which are the jobs it can do, and actions, which are the tools it uses to get information or perform tasks. The Atlas reasoning engine picks the subagent that fits a request, runs fixed logic, and then uses the LLM to choose actions and write the response.
What an agent can do
An agent can find opportunities for action, anticipate next steps, and start tasks on its own, with or without a human in the loop. It does this only within the use cases and guardrails you specify.
You can create an agent from scratch or from a template. Agents are available in Lightning Experience in these editions:
- Enterprise Edition
- Performance Edition
- Unlimited Edition
- Developer Edition
The required add-on licenses vary by agent type.