Can the model call anything it wants?

Only what it has been given. The tool list, the permissions behind each tool and any approval checkpoints are enforced by the system around the model, not by the model’s judgment.

What happens when a tool call fails?

The error is returned to the model, which can retry, choose a different approach or report that it is blocked. Handling that well is most of what separates a robust agent from a demo.

Function Calling

Function calling is the mechanism by which an AI model invokes an external tool as part of producing a response, and it is what agent behaviour is built on.

The model is given a set of tools with typed parameters. Instead of answering directly, it can emit a structured request to call one, the surrounding system executes it, and the result comes back for the model to continue from. That loop, repeated, is what an agent is: the model decides what to do next, the system does it, and the outcome informs the next decision.

Because the model chooses the tool and the arguments, function calling is also where most of the risk sits. The guarantees that matter are enforced outside the model: which tools are exposed to this person at all, what each one is permitted to do, and which calls stop for approval before they execute.

Frequently Asked Questions

No. Function calling is the mechanism. An agent is what you get when it runs in a loop with a goal, memory of what it has already tried, and rules about where to stop.