AI Agent

How it works

In the classic definition an agent perceives its environment through sensors and acts on it through actuators, choosing actions by some policy (Russell & Norvig).

An LLM-based agent realises this as a loop: the model reads the current state and history, decides which tool to call (or answers directly), the runtime executes the tool and feeds the result back, and the model uses that to choose its next step — until the task is done or a limit is hit. The model is the decision-maker; tools supply capabilities it lacks on its own (file access, retrieval, code execution).

Difference from a chatbot

A chatbot produces its output in one reply and does not interact with the outside world. An agent runs a multi-step, feedback-driven process where each observation changes the following decisions. The practical test is whether the system can advance several steps without a human directing each one.

Common misconceptions

• "Agent" is not a kind of model — it is the perceive–decide–act system built around one.
• Agentic does not mean fully automated: most deployed systems keep human approval at critical steps.

Related terms

Sources

1. Russell & Norvig, Artificial Intelligence: A Modern Approach (4th ed.), ch. 2 — Intelligent Agents
2. Anthropic — Building effective agents (2024)
3. Weng, L. — LLM Powered Autonomous Agents (2023)