Tool calling turns a model's words into a callable, typed action your app runs.
DefinitionWhat it means
Tool calling, also called function calling, is the interface by which a model, given a list of declared functions with JSON-schema signatures, outputs a structured call naming a function and its arguments rather than plain text; the host application parses that call, executes the real function, and feeds the result back to the model as an observation it can reason over.
Why it mattersWhy you should care
Tool calling is what turns a language model into a system that can look things up, write to a database, or trigger a workflow, so almost every production agent, copilot, or automation built in 2026 depends on reliable schema adherence and error handling around it; teams evaluate models partly on how consistently they format calls and recover from tool failures.
At a glanceSee it
The agent loop — the model keeps calling tools and feeding their results back until its stop reason flips from tool_use to done, which the linear happy-path never shows.
Validation and failure handling — malformed arguments bounce back as a typed error for the model to retry, and a thrown or timed-out tool returns error text rather than crashing the run.
Where you see itIn the wild
- API docs for OpenAI, Anthropic, and Gemini function-calling schemas.
- Agent frameworks like LangChain and the Claude Agent SDK wiring tools to models.
- Design discussions about schema design and handling malformed tool calls.