Agentic Development · Execution
A model output formatted as a structured call to a specific tool with named arguments, instead of plain text describing what it wants to do.
Example: Instead of writing 'let me check the weather', the model returns {"function": "get_weather", "args": {"city": "Almere"}} that your code can execute directly.
In practice: Teams enforce this with a schema and a validator, since models occasionally return malformed arguments, like a string where a number was expected.