Chatbot, copilot, agent: what the words actually mean
Three words that get used interchangeably by people selling you things. They are not interchangeable, and the difference is exactly where the value is.
These three words appear in every AI pitch you'll sit through this year, frequently in the same sentence, frequently meaning whatever the moment requires. Here is the plain version, and why the distinction decides what you should pay for.
A chatbot answers
You type, it replies. It can be superb — answering questions from your documents, drafting text on request — but it acts on nothing. It cannot read your inbox, update a record, or file anything. A chatbot is a conversation. When someone says "we tried AI and it was useless", nine times out of ten they bought a chatbot and expected it to do work. It was never going to; it can only talk about work.
A copilot helps a person work
A copilot sits inside a person's tools and assists whoever is at the keyboard — drafting the reply for approval, summarising the thread, suggesting the code. Genuinely useful, and the easiest to adopt since it changes no processes. But its output scales with the human's hours: it makes each person faster and does nothing at three in the morning, because it only acts when someone is driving.
An agent does the work
An agent is given a job rather than a keystroke: watch this inbox, read what arrives, work out which client it belongs to, file it, acknowledge it, flag anything unclear to a person. It acts on your real systems, on its own initiative, within limits you set — and keeps a log of everything it did. That last part isn't decoration. Because an agent acts without someone watching each step, the guardrails are the product: what it may touch, what waits for approval, what gets recorded. An agent without those isn't a bold version of the same thing; it's an unsupervised temp with your passwords.
Why the words decide the money
The three solve different problems at different prices. Staff drowning in questions that have documented answers? A chatbot is cheap and quick. Skilled people slowed by drafting and summarising? Copilots, per seat. Whole jobs — the reading, routing, filing, chasing — happening hundreds of times a week? That's agent work, and the only kind where the work happens without a person in the loop for every step.
So when a supplier says "AI", make them pick a word. Then make them show you which of your tasks it touches, and where the limits live. If they can't do the first, they're reading from a script. If they can't do the second, keep your data and your money where they are.
How this was made
Drafted by our research agent from the sources above, then edited, fact-checked and approved by a human before publishing. We publish this note on every article because we think you should know.
More articles
What should we automate first?
A scoring method you can run in a meeting: three questions per task, and the honest list of things that should stay human.
Is our data safe with AI? The questions that actually settle it
Not a reassurance piece. The five questions to put to any AI supplier in writing, what good answers look like, and the red flags that end the meeting.
Telling your team about AI without starting a panic
The moment you announce an AI project, everyone in the room hears one thing: someone is measuring my job. How to have that conversation honestly — and why the honest version is better news than they fear.