# What is AI Agent?

Also called agent, LLM agent.

An AI agent is a software system that uses a language model to decide what actions to take toward a goal, then carries those actions out through tools such as APIs, databases, or a browser. Unlike a single question and answer exchange, an agent runs over multiple steps, observing results and adjusting until the goal is met or it stops.

The defining feature is the loop. A model receives a goal and the current context, chooses either to answer or to call a tool, and the surrounding program executes that call and feeds the result back. That cycle repeats. Everything else, including memory, planning, and safety checks, is scaffolding around this basic exchange between a model that decides and a runtime that acts.

What makes agents useful is that the sequence of steps is not written in advance. A traditional program encodes every branch a developer anticipated. An agent is given a goal, a set of tools, and constraints, and works out the path at runtime. That flexibility handles messy real-world tasks, and it is also the source of the unpredictability that agent engineering spends most of its effort managing.

The term is used loosely. Some products call any chatbot an agent, while stricter definitions require tool access, multi-step execution, and some ability to pursue a goal without a prompt for every step. A useful test is whether the system can take an action in the world and then react to what happened. If it only produces text on request, most practitioners would call it an assistant rather than an agent.

Agents sit between two neighboring ideas. Below them are plain model calls and fixed pipelines, which are predictable but rigid. Above them are multi-agent systems, where several agents coordinate and each still contains the same loop. Understanding a single agent well is the prerequisite for reasoning about orchestration, memory, and autonomy limits, because every one of those concepts modifies some part of that loop.

## Key points

- An agent decides its own next step rather than following a fixed script.
- Tool access is what separates an agent from a chatbot.
- Agents run in a loop of decide, act, observe, repeat.
- The term is used loosely and definitions vary across vendors.

## In practice

A support inbox agent is given the goal of triaging new messages. It reads an incoming email, calls a search tool to look up the sender's order history, sees the order shipped late, drafts an apology with a replacement offer, and posts it as a draft for review. No developer wrote that exact sequence. The agent chose each step from the tools available.

## Related terms

- [Agentic AI](/en/glossary/agentic-ai)
- [Agent Loop](/en/glossary/agent-loop)
- [Tool Use](/en/glossary/tool-use)
- [Autonomous Agent](/en/glossary/autonomous-agent)
- [Multi-Agent System](/en/glossary/multi-agent-system)

[Back to the AI Glossary](/en/glossary)