Guide
AI agents, chatbots and automations: what each one actually is
Four words get used interchangeably by people selling four different things. Knowing which is which is the difference between buying what you need and buying a demo.
Automation
What it is: a set of rules that runs when something happens. A form is submitted, so a record is created and an email goes out.
What it is good at: anything where the steps never change. Reminders, notifications, moving data between systems, scheduled reports.
Where it breaks: the moment a step needs judgment. A rule cannot read a messy email and work out what the customer actually wants.
You probably need this if: you are re-typing the same information into two systems, or a task only happens when someone remembers.
Chatbot
What it is: a conversational interface, usually on a website. Older ones follow a decision tree; modern ones use a language model to answer in natural language.
What it is good at: answering common questions instantly, at any hour, in any language you have set up.
Where it breaks: when it has no access to your real data. A chatbot that cannot look up an order can only paraphrase your FAQ page, which is why so many of them annoy people.
You probably need this if: the same questions arrive constantly and the answers exist in a system you could connect it to.
AI agent
What it is: a system that can take actions, not just produce text. It can look something up, write to a record, book a slot, send a message, and decide which of those to do next.
What it is good at: completing a whole task rather than one step. Answering a caller, qualifying them, checking the calendar and booking the appointment is agent work.
Where it breaks: when the scope is not bounded. An agent given vague goals and broad permissions is how projects go wrong. Real ones have a narrow job and clear rules about when to stop and fetch a human.
You probably need this if: the task involves several steps and some judgment, and currently occupies a person who should be doing something else.
Copilot or assistant
What it is: AI that helps a person work faster, rather than doing the work unattended. It drafts, suggests and summarises; the human stays in the seat.
What it is good at: high-stakes work where you want the speed but not the autonomy. Drafting a proposal, summarising a long thread, preparing a first version of a report.
Where it breaks: it saves minutes, not headcount. That is not a criticism, it is the correct expectation.
You probably need this if: the work genuinely requires professional judgment but the blank page is the slow part.
How to choose, in one table's worth of thinking
Does the task have consistent steps and no judgment? Automation.
Is it answering questions people ask? A chatbot, but only if you connect it to real data. Otherwise you are building a worse FAQ.
Is it a multi-step job someone currently does end to end? An agent, scoped narrowly, with an escalation rule.
Does it need a professional's judgment? A copilot. Keep the human in the seat.
Most real systems are a mix. A voice agent that answers the phone is an agent for the conversation, plain automation for writing the booking into the calendar, and a human escalation path for anything unusual. Anyone selling you exactly one of these four for every problem is selling their product rather than solving yours.
The questions worth asking a vendor
What does it do when it does not know? The answer should be "escalates", not "makes its best guess".
What data does it answer from? If it is not your data, it is guessing about your business.
Who owns it if we stop working together? A build you own and a subscription you rent are very different things.
Does it decide anything on its own? Worth knowing for practical reasons and, in Québec, for legal ones. Our Law 25 guide covers what a fully automated decision obliges you to do.