How language models call external functions — and why an agent can only truly act once it has tools.
Category · Agentics
Tools are everything.
Tool use describes an LLM's ability to invoke external functions — database queries, API calls, code execution, search. Without tools, a language model is just a well-spoken sparring partner. With tools, it becomes a productive unit.
Most agent architectures stand or fall on the tool layer: how clean the descriptions are, how strict the validation, how reliable the error semantics.
What matters day to day.
Tool descriptions have to be as precise as API docs. Inputs are validated before execution. Outputs are structured, not "vaguely natural-language". And every tool call lands in the audit log.
See also
