AI Agent
A program that uses a language model in a loop to decide actions, call tools, and work toward a goal with little step-by-step instruction.
An AI agent is more than a single model call. It is a loop: the model looks at the situation, decides on an action such as calling a tool or asking a question, sees the result, and decides again, until the goal is met. That loop is what turns a chatbot into something that can search, calculate, browse, or change a system on its own.
The agent's power comes from tools. A model alone can only produce text. Give it a tool to run code, query a database, or call an API, and it can act in the world. The framework's job is to run the loop safely, pass the right context each turn, and stop when the work is done or a human needs to step in.
In 2026, most production agents are still relatively simple: one model, a handful of tools, and a clear stopping condition. The complexity comes from reliability, not cleverness, which is why state, observability, and human-in-the-loop matter more than raw agent count.