Agentforce agent building blocks
Key Takeaway
Actions are how an agent gets things done, and subagents group those actions by job. Agent Script lets you decide where the LLM reasons freely and where the agent must run deterministically.
An Agentforce agent is a conversational AI assistant that can plan and carry out tasks on its own, inside the use cases and guardrails you set. Every agent is built from the same building blocks, including a reasoning engine that calls a large language model (LLM). Agent Script is the language that ties these blocks together. It mixes natural language instructions with logic that runs the same way every time.
What an Agentforce agent is
Agentforce agents are goal-oriented, autonomous AI assistants. They can spot chances to act, anticipate next steps and start tasks within the use cases and guardrails you specify. They can do this with or without a human in the loop. You can create an agent from scratch or from a template.
Every agent is built from the same key building blocks:
- Subagents and actions
- Data
- Connections and channels
- The reasoning engine
Agents respect standard Salesforce access controls, such as licenses and permissions. Agents also include AI guardrails. They are integrated with the Einstein Trust Layer, a secure AI architecture built into the Salesforce Platform.