What is function calling? AI that uses your software

On its own, the model only produces text. With function calling, you describe the tools available to it, for example "find order by number" or "create appointment". When needed, it replies with a request to use one of those tools with certain parameters. Your program runs it and sends the result back.

It is the mechanism agents are built on. The delicate part is not the technique but the choice: which tools you give the model, and which actions need confirmation.

An example in an SMB

A garage's chatbot has two tools: check free slots and book. The customer writes "can I bring the car in on Thursday morning for its service?". The model calls the first tool, suggests the available times, and only after the customer confirms does it call the second.

The most common mistakes

• Giving tools that delete or pay without a confirmation step.
• Describing the tools vaguely: the model uses them badly.
• Not logging the calls made: without a record, you cannot work out what went wrong.

Why it matters to you

Function calling is what separates an assistant that talks from one that acts. When a supplier offers an assistant that reads orders or books appointments, ask for the list of tools the model can call and which actions go through a confirmation: that is where the risk is decided.

All the words of AI, explained simply, are in the glossary.