What it is
Function calling is how a language model translates natural language intent into a structured request — like updating a record or sending an email — that a system can execute. It's the bridge between reasoning and actual doing.
Why it matters
Without reliable function calling, an agent can only chat; with it, an agent can actually close a case or update a Salesforce record, turning conversation into action.
Key components
- structured parameters
- tool schema
- action mapping
- execution layer
How it connects
Agentforce works the same way: its reasoning engine picks an action from a topic and fills in that action's inputs, which then run as Apex, Flow, or API calls.
Good to know
Tight input validation matters — poorly constrained function calls are a common source of tool misuse and unintended actions.
Related terms
MCP (Model Context Protocol)
Anthropic's open standard for connecting AI models to external data sources and tools. Think of it as a universal adapter for AI.
Topics and Actions
The building blocks of Agentforce agents — topics define what an agent can discuss, actions define what it can do.
Copilot Actions
Pre-built AI capabilities in Salesforce that handle common tasks like summarizing records, drafting emails, and generating reports — no configuration required.
Tool Misuse
Attacks where the agent stays inside its allowed permissions but is tricked into using its tools in damaging combinations or with attacker-chosen inputs.
