An AI agent is more than a chatbot. A chatbot answers a message; an agent is given a goal and works toward it over several steps. At each step the model decides what to do, usually by tool calling (searching, querying a database, sending an email, running code), looks at what came back, and decides again.
Most agents are built from the same parts: a language model, a system prompt that sets its job and rules, a set of tools, some form of AI agent memory, and a loop that keeps going until a stop condition is met. The loop is where things go wrong in practice: an agent can repeat the same failing call, act on a wrong assumption, or run up cost, so real agents need limits, logging and error handling.
Related: Agentic AI, Large Language Model, AI Assistant, AI Automation.