The word agent gets thrown around a lot right now, often to mean nothing more than a chatbot with a new coat of paint. There is a real distinction underneath, and once you see it, the whole field gets easier to reason about.
A chatbot answers. An agent acts.
A chatbot is a conversation. You type, it replies, and nothing happens in the world until you type again. It is a very capable text box. An agent is different in one specific way: you give it a goal, and it decides what steps to take to reach that goal, then takes them. It can look things up, use tools, send messages, and check its own work, looping until the job is done or it needs your input.
Example
Same request, two very different tools
Ask a chatbot to "keep an eye on what our competitors ship and tell me when it matters," and it will explain how you might do that. Ask an agent the same thing, and it can actually read their changelogs and announcements every morning, decide which changes matter for your business, and message you a short note the day one does. One gives you instructions. The other does the task.
The shift: from answers to outcomes
That is the mental model to hold on to. A chatbot produces answers. An agent produces outcomes. To get outcomes, an agent needs a few things a plain chatbot does not have:
- A goal, written by you in plain language, that it works toward.
- Tools it is allowed to use, like searching the web, reading your calendar, or sending an email.
- The ability to run on its own, on a schedule or when something happens, without you prompting each step.
- Some memory of what it has already done, so it does not repeat itself or lose the thread.
Notice what the competitor example asked for: not a lookup, a judgment. Deciding which of thirty changes matters to you is work a script cannot do and you never have time to do. That middle ground is exactly where agents live. Next we will look at the tasks in an ordinary working week that sit in it, so the idea stops being abstract and starts being useful.