Input variables versus hardcoded action values
Key Takeaway
By default, the LLM fills each action input from the conversation (slot filling). When you assign a variable to an input, the agent can use only that variable's stored value to fill it.
An action input in an Agentforce agent can get its value in two ways. By default, the agent fills it from the conversation, which is called slot filling. You can instead assign a variable to the input, so the action can only use that stored value. Assigning variables to action inputs makes an agent more secure and consistent, because important values no longer depend on what the LLM infers or remembers.
Two ways an action input gets its value
An action input is agent populated by default, but you can assign a variable to it instead.
| Agent populated (default) | Variable assigned to the input | |
|---|---|---|
| How the input gets its value | The LLM reasons over the conversation to fill the input. This is called slot filling. | The agent can use only the assigned variable to fill the input. |
| How stable the value is | Depends on what the LLM infers, or on information it pulls from its context memory. | Stays stable across subagents, actions and turns. |