AI agent, chatbot, automation: what are the differences?
Chatbot, automation, assistant, AI agent: what each one does, the effort it takes, and which to choose for the work you want to hand over.
Chatbot, automation, assistant, agent: the words get mixed up in sales pitches. Yet they do not do the same work, and they do not take the same effort to set up. Here is how to tell them apart.
The four tools, in one sentence each
- A chatbot answers questions in a chat window, often from a bank of answers.
- An automation (also called RPA or a workflow) always runs the same sequence of actions: "when a form arrives, create a record in the CRM".
- An AI assistant (ChatGPT, Copilot, Le Chat…) helps a person write, summarise, search; that person stays in charge.
- An AI agent understands a request, chooses the steps and acts in your tools, following your rules, and hands over to a person when a decision commits the company.
The comparison table
| Chatbot | Automation | AI assistant | AI agent | |
|---|---|---|---|---|
| Understands free text | often partly | no | yes | yes |
| Acts in your tools | rarely | yes | no, the person acts | yes |
| Handles varying cases | no | no | yes, with the person | yes, within a set framework |
| Works without being prompted | no | yes | no | yes |
| Set-up effort | low | low to medium | almost none | medium: the work and its rules must be described |
When to choose what?
The work is always identical: automation
If the data always arrives in the same format and the sequence of actions never changes, an automation is enough. It costs little, it is reliable, and it needs no AI. We say so when that is the case: it is one of the possible conclusions of the Flash Diagnostic.
A person wants to go faster: the assistant
Drafting a first version, summarising a document, rewording an email: an off-the-shelf assistant does this work very well, provided you set rules of use (which data to put in, who reviews). See Off-the-shelf tool or custom agent: which to choose?
The work varies, but follows rules: the agent
Freely written emails, invoices in different formats, customer requests that are never quite alike, with decisions to make along the way: that is agent territory. It reads, it understands, it applies your rules, and it flags what falls outside the framework.
Example: a contact form can be routed by an automation. But qualifying the request (quote, complaint, technical question), finding the customer in the CRM, suggesting a suitable reply and alerting the salesperson if the prospect matters: that is an agent’s work.
The trap to avoid
Calling a chatbot that only answers, or an automation with an AI step added, an "agent" creates expectations that will not be met. Before signing, ask the provider:
- In which tools does the agent act, concretely?
- What does it do when it does not know?
- Who approves the actions that commit the company?
- How is the gain measured, and from what starting point?
If the answers stay vague, that is a warning sign. For the full definition, read What is an AI agent?
Frequently asked questions
Is the chatbot on my website an AI agent?
Not necessarily. If it only answers from a list of questions, it is a chatbot. If it checks your order tracking, opens a ticket in your tool and passes the case to a person when it does not know, it behaves like an agent.
Is traditional automation obsolete?
No. For a sequence of actions that never changes, on well-structured data, it remains the simplest and cheapest solution. An agent becomes useful as soon as free text has to be read or cases vary.
Can the two be combined?
Yes, and it is common: an automation triggers the agent on each new request, the agent handles the case, then another automation files the result.