Tool Use (AI)
Tool use is the capability for a language model to call external functions — search, code execution, APIs — with structured arguments and use the results in its next response.
In plain English
Tool use is the model feature that turned language models into agents. Without it, the model can only emit text. With it, the model emits structured tool calls — "run this SQL," "fetch this URL," "write this file" — and the runtime executes them and feeds the results back. The model then uses those results in its next turn.
Under the hood the mechanism is simple: the developer declares a list of tools with names, descriptions, and JSON schemas for their inputs; the model's output is parsed; tool calls are dispatched; results are added to the next prompt. But the cultural shift is huge. Once a model can reliably call tools, it is no longer a text generator — it is a decision-maker that happens to express its decisions as text.
Why it matters for Black Box
Every Black Box specialist operates by tool use. The Coding specialist calls file-write and shell tools. The Browser specialist calls navigation and scraping tools. The Content specialist calls publish and schedule tools. Tool use is how the product does work instead of talking about work.
Examples
- A model calling a
search_webtool and using the three returned snippets to write a summary. - A model calling a
send_emailtool with a subject, body, and recipient. - A model calling a
run_teststool and deciding whether to retry based on the output.