Function Calling and Tool Use in LLMs

Function calling is the mechanism that turns a language model from a text generator into something that can genuinely act on business systems: checking a database, sending an email, or updating a record. It works by letting the model recognize when a task requires an external action and generate a structured request for that action instead of guessing at an answer from memory.

Function calling lets a language model recognise when a task requires an external action, like checking a database or sending an email, and generate a structured request for that action rather than trying to answer from its own knowledge.

This is the mechanism behind most modern AI agents, turning a text generator into something that can genuinely interact with business systems.

We spend significant design time on the function definitions themselves, since a poorly described tool leads the model to call it incorrectly or not at all.

A function definition typically includes a clear name, a description of what it does and when to use it, and a schema for its parameters. The quality of that description matters more than most teams expect: an ambiguous description leads the model to either call the wrong tool, call the right tool with malformed arguments, or fail to call any tool when one was actually needed.

A practical pattern is to keep each function narrowly scoped to a single, well-defined action rather than one large multipurpose function with many optional parameters, since narrow functions are easier for the model to select correctly and easier for engineers to test in isolation.

We also build validation and error handling around every function call, since a model can generate a syntactically valid but logically wrong request, such as an out-of-range date or a missing required field. Returning a clear error back to the model, rather than failing silently, often lets it self-correct and retry with better arguments on the next turn.