You are building a Databricks agent that answers customer questions about their orders. The agent must call an internal `get_order_details(order_id: str, include_shipping: bool)` function. During testing, the foundation model frequently returns the arguments as free-form prose (e.g., 'the order id is 12345 and yes include shipping'), which your Python code cannot parse reliably to invoke the function. You want the model to reliably emit arguments matching the function signature. Which approach best resolves this?