"AI agent" is one of the most overloaded terms in the industry right now. It's used for everything from a chatbot with a clever prompt to a fully autonomous system. That vagueness makes it hard to reason about what you're actually building — so here's a definition that holds up in practice.
A working definition
An agent is a language model placed inside a control loop, given tools to act on the world and memory to carry state, working toward a goal with some degree of autonomy over how it gets there. The key words are loop and autonomy. A model that answers one question is not an agent. A model that decides what to do next, does it, observes the result, and repeats until the goal is met — that's an agent.
The four moving parts
Almost every agent, regardless of framework, is built from the same components:
- The model — the reasoning engine that decides what to do.
- Tools — the actions it can take: search, code execution, API calls, database queries.
- Memory — the state it carries across steps and sessions.
- The control loop — the orchestration that runs perceive → plan → act → observe until the task is done or a limit is hit.
Autonomy is a dial, not a switch
The most useful mental shift is to stop asking "is this an agent?" and start asking "how much autonomy does it have?" A tool that drafts an email for you to approve sits low on the dial. A system that triages tickets, takes action, and only escalates the ambiguous ones sits higher. The right level depends entirely on the cost of a mistake — which is why the same underlying model can power a cautious assistant or a bold automation.
Why the distinction matters
Calling everything an agent leads to bad decisions. If you think you're building an agent but you really need a well-prompted single response, you'll add orchestration complexity you don't need. If you think a chatbot will do but the task genuinely requires multi-step tool use, you'll ship something that can't actually complete the job. Naming the shape of the problem correctly is the first design decision.
The takeaway
An agent is a model in a loop with tools, memory, and a goal. It's not magic and it's not a synonym for "AI feature." Deciding how much autonomy the task can safely support — and building only as much machinery as that requires — is what separates a system that ships from a demo that impresses.
