AI agent

An AI agent is a software system that pursues a goal on its own. Given an objective, it perceives the relevant state of its environment, reasons about what to do next, plans a sequence of steps, and then acts — typically by calling external tools or APIs — before observing the result and deciding on its next move. That perceive–reason–act cycle, run in a loop, is what makes it an agent rather than a one-shot responder.

Under the hood, the reasoning is usually powered by a large language model, but the model alone is not the agent. The agent is the wider system: the model plus the tools it can invoke, the loop that lets it take many turns, and the memory that carries context forward. Each turn, the agent feeds its running transcript back into the model, lets it choose an action through tool calling, executes that action, and returns the real result as the next observation.

Why it matters: this loop lets an AI agent handle messy, multi-step work a single prompt cannot — booking travel across several sites, triaging a support queue, debugging code by running tests, or researching a question by searching and reading. Because it acts on fresh data and reacts to outcomes, it can recover from a failed step instead of confidently returning a wrong answer. A concrete example: a research agent asked "summarize our top three competitors" will search the web, open pages, extract facts, notice a gap, search again, and only then write the summary — looping until the goal is satisfied.