Chatbot, copilot, assistant, agent. The words are used interchangeably in marketing and they mean genuinely different things in terms of what you are buying, what it costs, and what can go wrong. Getting the distinction right before you talk to a vendor will save you a budgeting argument later.
The three shapes, in plain terms
| What it does | Who drives | What you pay for | |
|---|---|---|---|
| Chatbot | Answers questions from a defined body of knowledge | The user, one question at a time | Conversations or seats |
| Copilot | Suggests, drafts and completes inside a tool someone is already using | The user, continuously | Seats, almost always |
| Agent | Carries out a multi-step task end to end and reports what it did | A trigger — an email, a file, a schedule | Work completed, or a project |
The practical difference is who is holding the thread. With a chatbot or a copilot, a person is present for every step and the value scales with how many people are using it. With an agent, the work happens whether or not anyone is watching, and the value scales with volume of work rather than headcount.
Why this decides your pricing model
Seat pricing makes sense when the value is proportional to the number of people using the thing. It stops making sense when the work is document-shaped. If your case is "four hundred drawings a month need checking", a per-seat licence is the wrong instrument: you either buy seats for people who barely touch it, or you buy too few and quietly cap the process you were trying to scale.
Where each one actually earns its money
None of these is better than the others. They fit different work, and most companies eventually run all three.
- 1Chatbots earn their keep on repeated questions with one right answerOpening hours, order status, where a document lives, what a specification says. High volume, low stakes, and the cost of being wrong is that somebody asks a human instead.
- 2Copilots earn their keep where a skilled person is already workingDrafting, summarising, writing code, first-pass analysis. The person's judgement is the product; the copilot removes the typing.
- 3Agents earn their keep on work that arrives whether or not anyone is freeInbound email, incoming drawings, quotation requests, catalogue updates. The work has a trigger, a defined outcome, and a queue that builds up when nobody gets to it.
The part vendors are vague about
"Agent" has become the fashionable word, so a great deal of software that is really a chatbot with a longer prompt is now sold as an agent. Two questions cut through it.
- What triggers it without a person present, and what happens to the result when nobody is logged in? If the honest answer is "a user opens it and types", you are buying a chatbot.
- What does it write to, other than the screen? An agent that cannot put its output into your systems — your inbox, your PIM, your documents, your ERP — is a demo with good manners.
What this means for how you buy
Chatbots and copilots are products: you evaluate them on a trial, you pay per seat, and switching is relatively cheap. Agents behave more like integration projects, because most of the work is in the connections to your systems and the rules that encode how your company actually operates. That makes them slower to start and considerably harder to rip out — which cuts both ways, and is worth knowing before you sign rather than after.