Agents
Tool calling
Letting a model invoke functions you define — how an AI system does anything beyond producing text.
You describe the functions available ("create a shipment, given an address and a weight"), and the model decides when to call them and with what arguments. Your code executes the call and returns the result.
This is the mechanism behind every agent. It is also the security boundary that matters, because each tool is a door into your systems.
Good practice: make every tool explicit and narrowly scoped, validate arguments against a schema before executing anything, and gate the consequential ones. The model proposes; your backend decides.