Chatbot vs AI agent: what you can ship in 30 minutes

The short version

A chatbot answers you from what it already knows. An agent goes and does something first, then reports back. The dividing line is whether the thing can reach outside itself.

That single distinction decides what you can build, what it costs to run, and how you tell whether it worked. If you are choosing what to build in a first sitting, it is the only distinction that matters.

Side by side

Why the difference is not just semantics

Ask a chatbot to research a topic and it will produce something that looks exactly like research. Structure, confident claims, sometimes citations. It did not go and look. You are reading a recollection of the internet, not a reading of it.

Give an agent a search tool and ask the same thing and it goes, fetches, reads, and tells you what it found. If your instructions ask it to list what it checked, you get something you can go and verify — which is a different kind of answer from one you simply have to trust.

For a lot of tasks the chatbot answer is fine. For anything where being current matters, or where you need to check the work, the difference is the whole point.

What this costs you in practice

The thing that makes agents useful is also the thing that meters them. On Zapier's free plan an agent gets up to 10 "activities" per run and 400 a month. An activity is one instance of reaching outside itself — a search, a page read, an action. Talking to it is not counted.

So a chatbot conversation is cheap and an agent run has a budget. A task defined without a stopping condition can spend that budget before it finishes the job, at which point the run pauses and asks you something instead of delivering. That is the practical cost of the capability, and it is manageable once you know it exists.

Checked on 23 August 2026 against Zapier's usage documentation. Platform limits move, so treat the date as part of the fact — if you are reading this much later, verify before relying on it.

Which should you build first

An agent, and a small one. A chatbot you can build in minutes but it teaches you almost nothing you did not already know from using one.

The agent teaches you the three-part structure — instructions, a tool, something that runs it — and that structure is what every more ambitious build is made of. You learn it once on something small and the pattern transfers.

The build that fits a first sitting is a research-and-summary agent: one job, one tool, output you can immediately judge. About half an hour, no code.

The honest limitation

An agent is not a smarter chatbot. It is a chatbot that has been given a job and some equipment, and it will do that job badly if the instructions are vague — the difference is that a well-written instruction can make it show you what it did, so the badness is findable.

That checkability is worth having. It is also why the useful skill is not picking the right product but writing instructions precise enough to act on — including asking it to show its working. The full walkthrough goes through what that means block by block.