What is Function Calling?
Function calling is a technical capability provided by many modern AI models that allows them to reliably output structured data (typically JSON) designed to trigger external code. You provide the AI with a description of the functions available to it. If the AI decides a function is needed to answer the prompt, it outputs the exact arguments required.
How does it work?
A developer tells the AI: "I have a function called get_weather(location)." The user asks: "Is it raining in London?" The AI realizes it doesn't know the current weather, so instead of generating a text response, it outputs a JSON object: {"name": "get_weather", "arguments": {"location": "London"}}. The developer's software runs the function and feeds the result back to the AI.
What is it commonly confused with?
Function calling and Tool Calling are often used interchangeably, but there is a sensible distinction. Function calling specifically describes the model's technical ability to produce structured arguments matching a declared schema. Tool calling is the broader software pattern in which an AI system selects and requests those external capabilities.
Why does it matter?
Language models are trapped in the past by their training data cutoff and cannot interact with the outside world. Function calling provides a reliable, mathematically structured bridge between the probabilistic AI model and the deterministic, rigid APIs of traditional software.