LLMs alone cannot read live data or change the world. Agents call tools: search the web, call APIs, query databases, run code. That is how they go from "generate text" to "get the answer and book the flight."

The agent receives a request, reasons about which tool to use and with what inputs, calls the tool, gets the result, and may repeat. The loop continues until the task is done or a final answer is returned.

Expose clear, small operations (e.g. "search", "get_weather", "create_calendar_event"). The model chooses from available tools using their names and descriptions. Well-designed tools make agents reliable and easier to debug.