Guides · long read, updated

AI agent vs chatbot: what each one does, where each one fails, which one you need

A chatbot follows a script: buttons, menus, canned answers. An agent reads your own data, answers questions the script never anticipated, drafts replies for a person to approve and hands over what it cannot answer. This guide sets the two side by side, shows where each fails, and says which one a small business needs, which is more often neither than either.

Updated 2026-09-04 · 2 min read · numbers only with a source

What a chatbot does

A chatbot is a decision tree with a friendly face. It offers buttons, asks a question, matches the reply against a list, and shows a canned answer. It is cheap, predictable and fine for ten fixed questions with ten fixed answers. It fails the moment someone types a question the tree did not anticipate, and it fails silently: the customer gets “I did not understand” and leaves.

What an agent adds

An agent reads: your price list, your calendar, your customer records, your rules. It answers the question as asked, from that data, and drafts the reply for a person to approve. When the data does not contain the answer, it says so and hands over with the whole thread. It can also act, within rules: book a slot, open a ticket, send a confirmation.

CapabilityChatbotAI agent

Answers unexpected questionsNo; shows “I did not understand”Yes, from your own data, or hands over

Reads your systemsNo; answers are typed into itYes: price list, calendar, records

ActsLinks to a formBooks, confirms, opens a ticket, within rules

Says it is an assistantSometimesAlways, by rule and by law in the EU

Hands over with contextRarely; usually a dead endYes: whole thread plus a drafted next step

Learns from gapsSomeone edits the treeUnanswered questions are logged and the data is fixed

Where each one fails

Both fail; they fail differently. A chatbot fails visibly and often: any question off the script. An agent fails rarely and dangerously: when the data it reads is wrong or stale, it answers confidently and wrongly, and only a handover rule or a person catches it. That is why an agent runs with approvals for its first month and reads data someone maintains.

FailureChatbotAI agent

A question off the scriptDead endAnswered from data or handed over

Stale price listWrong answer until someone edits the treeWrong answer until the data is fixed; logged

A complaintA menuHanded to a person without a draft, by rule

A question about money or healthWhatever is in the treeNever answered by the agent, by rule

Nobody maintains itSlowly becomes wrongBecomes wrong faster, because it is trusted more

Cost and control

A chatbot is a monthly tool subscription and someone’s afternoons building the tree. An agent is a fixed build from us and the model’s usage bill, paid to the provider from your own account.

from €500A basic AI agent
Answers repeating questions from your data, drafts for approval, hands over the rest, says it is an assistant.
Details →
priced per caseAn agent that acts
Books, confirms, opens tickets, in several languages, within written rules.
Details →
from €500A process without an agent
Often the honest answer: drafts and reminders on top of the inbox, no conversation interface at all.
Details →

Prices: HeadPills price ladder, checked 2 September 2026

Which one a small business needs

Usually neither, at first. If the ten questions are truly fixed, a good FAQ page and a fast human reply beat both. If the questions vary and the data exists in writing, an agent pays back. A chatbot is right in exactly one case: a strict, short flow such as “choose a service, choose a time”, where a tree is a feature.

01Do the questions repeat?
If not, a person. If yes, continue.
02Are the answers written down?
If not, write them first; nothing can answer from a head.
03Do the questions vary in wording and detail?
If not, a FAQ page or a chatbot. If yes, an agent.
04Is there a person to hand over to?
If not, no agent yet; fix the inbox first.
05Start with approvals
A month of a person approving every draft, then decide what may run alone.

What we do of this

Our part

We build agents from €500 that read your own data, draft for approval and hand over the rest, and we run one on our own enquiries. No client agent is deployed yet; we say so on every page where it matters. When a chatbot or a plain FAQ page is the right answer, we say that too.

Custom AI agents AI automation without an agent What happens when the agent does not know

Questions

Questions from readers

Is an AI agent just a smarter chatbot?

No. A chatbot follows a script; an agent reads your data, answers questions the script never anticipated, drafts for a person to approve and hands over what it cannot answer. They fail differently, and an agent needs maintained data and a handover rule.

When is a chatbot the better choice?

For a strict, short flow where a decision tree is a feature: choose a service, choose a time. For ten fixed questions, a FAQ page and a fast human reply usually beat both.

Can an AI agent replace customer support in a small business?

It can take the repeating questions and the drafting; a person keeps complaints, money, health and judgement. Without a person to hand over to, an agent should not be deployed at all.

Send us the ten questions you get most.

We reply with agent, chatbot, FAQ page or a person, and a price where one applies.

Talk to us See pricing