The two words don't mean the same thing
Almost everything sold today as «artificial intelligence for customer service» is one of two fairly different things. It pays to know which one you're being offered before you sign, because the price looks similar and the result does not.
A chatbot follows a script
A classic chatbot is a decision tree. Somebody sat down and drew the possible questions and the matching answers, and the program walks that drawing. That's why it asks you to type «1», «2» or «3»: it isn't a design quirk, it's that it can only move along the branches somebody drew.
It works as long as the conversation resembles the drawing. The moment someone writes «hi, I saw the ad, would this work for a pharmacy?», the chatbot has no branch for that and answers the only thing it knows: «I didn't understand your answer. Please choose an option from 1 to 3.»
An agent understands, decides and acts
An AI agent doesn't walk a drawing: it interprets what it's told. Underneath it has a language model —the same family of technology as the AI assistants you already know— and, above all, tools: it can check your calendar, work out a price range, save the contact, alert your team or hand the conversation to a person.
That's the difference that matters. A chatbot answers. An agent acts.
Four differences you can spot without being technical
- Memory. The agent remembers what was said ten messages ago; the chatbot starts from zero on every branch.
- Going off script. You can ask an agent two things at once, change your mind halfway through, or write with typos. Not a chatbot.
- Tools. The agent reads from and writes to your systems; the chatbot only shows text somebody typed in advance.
- The handover. A good agent knows when a case isn't its to handle and passes it to a person with the context already gathered.
How to tell them apart in two minutes
You don't need to understand the technology. Try these four things in the chat you're being shown:
- Ask a question that isn't on the menu, in your own words.
- Ask two things in the same message.
- Change your mind halfway: «actually for the other branch, not this one».
- Write to it in another language, or send it a voice note.
A chatbot breaks on the first one. An agent handles all four and, when it doesn't know something, says so instead of making it up.
So chatbots are useless now?
Not at all, for what they're good at. If your case is a narrow, highly repeated task —checking an order status from a tracking number, say— a menu is cheaper, more predictable and easier to audit. The problem isn't the chatbot: it's being sold one while being told it's the other thing.
What to ask whoever is selling it to you
- Can it check my calendar or my systems, or does it only return text?
- What does it do when it doesn't know the answer?
- How and when does it hand the conversation to a person?
- Does it remember the conversation if the customer comes back tomorrow?
- Where are those conversations stored, and for how long?
If the first two answers are vague, what you're being shown is a menu under a different name.
At Intellectum we build agents, not menus. IntelliA, the assistant that answers on this very site and on WhatsApp, checks the calendar, puts together price ranges and logs the contact before anyone on our team reads it. You can try it right here and run the four tests above on it.