Tool schemas are the contract between model reasoning and application execution. They tell the model what operations exist, what each operation is for, and what arguments are valid. They also tell your runtime what must be validated before anything touches a database, API, queue, or file system.

A strong schema reduces ambiguity. It helps the model choose the right tool, extract arguments from natural language, avoid unsupported parameter values, and ask for clarification when the request is underspecified.

Weak schemas push that work into chance. Generic names, broad tools, missing enums, and vague descriptions lead to wrong calls, fabricated parameters, overuse of tools, or unnecessary user follow-ups.

In this chapter, you will design schemas the way a production AI engineer designs APIs: narrow capability, clear intent, explicit constraints, predictable errors, and test cases for the requests most likely to confuse the model.