Learn how LLMs decide when to call external APIs, how tool schemas drive selection accuracy, and what really happens under the hood.