What a chatbot is and what an AI agent is
A chatbot is a program that chats with your customers in writing. The classic kind runs on rules: a numbered menu, keywords or a decision tree that someone designed in advance. If the message matches a rule, it answers. If it doesn't, it shows the menu again.
An AI agent also chats, but it understands the message the way the person wrote it, typos included. Then it decides what to do and does it with tools: it searches your catalog, looks up an order, books an appointment or hands the conversation to your team. We walk through it step by step in what is an AI agent.
The short version: a chatbot answers and an AI agent resolves. You see the difference the moment a customer goes off script, which on WhatsApp happens all day long.
The differences, point by point
The table compares a rules chatbot with an AI agent on what matters to a business that serves customers over chat. What it understands, where its answers come from, what it does next and how much control it leaves you.
Look closely at the last two rows. An agent does more, so you need to see what it did and be able to correct it. Few people ask about that when they choose a tool, and it is what matters most a month later.
| Rules chatbot | AI agent | |
|---|---|---|
| How it reads the message | Looks for keywords or menu options | Understands the intent, even when it's misspelled or mixed |
| Where answers come from | Fixed text someone wrote for each branch | Your catalog, your documents and your business data |
| What it does besides answering | Nothing, or it sends a form | Creates orders, books appointments and checks stock |
| What happens with an unexpected question | Repeats the menu or says it didn't understand | Answers from your data or asks your team |
| Memory between conversations | Starts from scratch every time | Remembers the customer and what they ordered before |
| Handoff to a person | When the customer picks the talk-to-an-agent option | Following the rules you set, at the right moment |
| How you set it up | Drawing every branch of the tree by hand | Telling it how you work and connecting your data |
| How you see what it did | You see the chat, not which rule fired | Every action is noted in the chat |
The step in between: the AI chatbot
Between the two sits a third option that confuses everyone: the AI chatbot. It is a language model, like the one behind ChatGPT, plugged into your WhatsApp. It understands natural language and writes fluently. It lacks two things: access to your data and permission to do anything.
That is why it handles general questions well and fails on specific ones. Ask whether a size is still in stock and it has no way to check. It writes something that sounds right, and sometimes it makes it up. We explain why in why AI makes up answers and what to expect from ChatGPT for customer service.
An AI agent uses the same kind of model and adds tools to look things up and act in your systems, within limits you set. If you are comparing products in this category, our guide on choosing an AI chatbot for business lists what to check.
Where a rules chatbot gets stuck
It gets stuck on the first message nobody planned for. A customer writes: "hi, are you open Saturday? also, has my order shipped?". The chatbot spots "Saturday", sends the opening hours and drops the second question. Or it sends back the whole menu.
Real messages look like that. They bundle two questions, a voice note, a photo or a misspelled word. Every new case means another branch in the tree, and the tree grows until nobody understands it.
The cost lands on your team. Whatever the chatbot can't solve goes back to a person, now with a customer who is less patient than before. We list the usual warning signs in when a WhatsApp chatbot falls short.
Keeping it up to date takes work too. Every promotion, new product and change of hours has to be edited into the tree. If nobody updates it, the chatbot keeps offering what you no longer have, as we describe in a chatbot with outdated stock.
When a chatbot is enough
A rules chatbot is enough when all of these hold at once:
- You get few kinds of questions, and they repeat almost word for word.
- The answers don't depend on data that changes, like stock, prices or open slots.
- Once it answers, there is nothing left to do.
Opening hours, your address and payment methods fit that group. If that is most of what your business gets, a simple menu can work for a while.
As soon as customers ask about an order, an appointment or something they bought last week, the menu falls short. That is when the manual handoffs and the unanswered messages start piling up.
In effort, a rules chatbot asks for work up front and again every time something changes. An agent asks for a different kind of work: organize your information once and review what it did during the first weeks. After that, what you maintain is your catalog and your documents, which you need to keep current anyway.
When you need an AI agent
You need an agent when the answer depends on your data or ends in an action. These are the clearest signs:
- Customers ask about price and stock on products that change every week.
- Orders arrive by chat and someone copies them by hand into a spreadsheet.
- You book appointments on WhatsApp and reply after hours so you don't lose them.
- The same customer comes back and you have to dig up what they bought last time.
- Your team spends more time answering the same thing than solving the hard cases.
In those cases Orbi, Alumbra's AI agent, answers from your catalog, creates the order with the right items and books the appointment in the calendar you already use. When a case needs a person, it hands the conversation to your team in Alumbra Chat.
An agent also pays off when you serve customers on more than one channel. Orbi answers on WhatsApp, Instagram and your website chat with the same information, so you are not maintaining a separate menu for each.
The risks of an agent and how to keep them in check
An agent does more, so it can also get more things wrong. The two most common fears are that it makes up an answer and that it does something it shouldn't. You control both from the setup.
- Orbi answers from your catalog, your documents and your business profile. When it can't find the answer, it asks your team instead of guessing.
- What your team tells it becomes a learning, and you approve it before Orbi uses it.
- You decide what it can do on each channel: which orders it creates, which statuses it can set and when it hands the conversation to a person.
- Every action Orbi takes is noted in the conversation and in the order history, with what it did and when.
- In Channel Mode you talk to Orbi the way a customer would, before you open it to real customers.
If a system acts without leaving a trail, you end up checking everything by hand. That is why the record matters as much as what it can do. We go deeper in transparent AI for customer service and handoff rules.
How to move from a chatbot to an agent without starting over
Moving from a chatbot to an agent takes less work than building the original tree. You don't draw branches: you tell Orbi how your business works and connect your data.
- Gather what you already have in writing: FAQs, price lists, shipping and return policies. Orbi answers from your documents.
- Connect your catalog or add it in the Panel, so it can check price and stock.
- Decide in which cases your team wants to take over the conversation.
- Test it in Channel Mode with the messages that trip up your chatbot today.
- Start with one channel and add the rest once you are happy with it.
If your chatbot runs on the official WhatsApp API, check how your number is connected before you switch. We cover it in requirements to connect the WhatsApp Business API. Alumbra connects as a Meta Tech Provider, through the official route and without sharing passwords. And if you'd rather go step by step, rolling out AI in stages suggests an order.
