Not every chatbot with a tool is an agent. We break down the loop of planning, acting and observing that separates true agents from scripted automations.

“Agentic AI” has become one of those terms that gets stuck on everything. Underneath the marketing, there’s a real and useful definition — and knowing it will save you from both overhyping and underestimating what these systems can do.

The simplest definition

An AI agent is a system that can pursue a goal by choosing its own sequence of actions, rather than following a fixed script written by a human.

The key word is “choosing.” A traditional automation runs the same steps every time; an agent decides, based on what it currently sees, what to do next.

The loop: plan, act, observe

Most agents run some version of the same three-step loop:

  • Plan — decide what to try next, given the goal and everything observed so far.
  • Act — actually do it, using a tool: send an email, run code, search the web, call an API.
  • Observe — look at what happened, and feed that back into the next planning step.

This loop repeats until the goal is met, a limit is reached, or the agent asks a human for help. It’s the same pattern whether the agent is booking a flight, debugging code, or researching a topic.

Agent vs. assistant vs. automation

It helps to place these three on a line:

  • Automation follows a fixed script: if X happens, always do Y. No judgment involved.
  • An assistant (like a chat-based helper) answers questions and can use tools, but generally does one step per your request — you’re still driving.
  • An agent is handed a goal and drives itself, taking multiple steps and adjusting its own plan without you approving each one.

Plenty of products live in between these, which is exactly why the term gets used so loosely.

A quick way to tell if something is really an agent

Ask one question: if the first thing it tries fails, does it change its approach on its own, or does it just stop and report back to you?

If it adapts and keeps going, you’re looking at something agentic. If it stops at the first obstacle, it’s closer to a smart assistant — genuinely useful, just not an agent in the strict sense.