LLM agent

A plain chatbot answers whatever you ask and then waits for your next message. An LLM agent is handed a goal instead of a single question — say 'book me a table for four on Friday' — and then works toward it on its own, deciding what to do, taking an action, looking at the result, and deciding again. It is the same underlying model, but wrapped in a loop that lets it act rather than just reply.

Mechanically, an agent is a model running inside a loop. On each turn it sees the goal plus everything that has happened so far, and produces either a final answer or a request to use a tool. The surrounding program runs that tool, appends the result to the conversation, and calls the model again. This repeats until the model decides the goal is met or some limit is reached.

The catch is reliability. Because each step depends on the last, a single wrong turn can send a whole run off course, and small per-step error rates compound over a long task. Good agents are built with guardrails, verification, and limits — not just a clever prompt.