AI agent vs. button chatbot: what changes for the customer asking about price
A button chatbot asks; an AI agent understands. The real differences, when buttons are enough and how an AI sales agent's risks are kept under control.

In this article
Monday, 8:15 a.m. A customer snaps a photo of a machine's nameplate and records a 40-second voice note: "Hi, I need the replacement part for this one. Do you have it and how much is it?"
The business's chatbot replies instantly:
"Hi! Choose an option: 1. Sales. 2. Support. 3. Hours."
The customer types "1". The bot replies: "Type the name of the product you're looking for."
The customer doesn't know the product's name. That's why they sent the photo.
It answered fast, but it didn't answer anything. That's the difference between a button chatbot and an AI sales agent.
In this article you'll see what each one does, which differences your customer notices, when a button menu is enough and how the risks of an AI agent are kept under control.
In short: a button chatbot forces the customer to speak the menu's language. An AI agent speaks the customer's language: free text, voice notes, photos and documents. Buttons work for simple, predictable flows. If your customers ask different things, send voice notes or need a quote, the menu falls short.
What a button chatbot is (and why it worked for years)
A button chatbot is a decision tree. Each customer message leads to a branch: "if they pick 1, show this; if they pick 2, show that."
It worked for years for one simple reason: it's predictable. You know exactly what it will say, because you wrote every reply.
The problem is that your customers aren't predictable. They don't read menus. They don't pick options. They write the way they'd write to a person: with typos, three messages in a row, a voice note while driving or a photo of what they need.
When the message doesn't match any branch, the bot repeats the menu. And the customer leaves.
What an AI sales agent is (no jargon)
An AI sales agent is a system that reads the whole conversation, understands what the customer wants and replies with your business's information: your catalog, your prices, your shipping rules and your policies.
It doesn't follow a tree. It follows instructions, the way a new salesperson would after you handed them a manual:
- This is what we sell and at what price.
- This is what we never promise.
- When the customer wants to negotiate or is ready to pay, hand the conversation to the team.
The easiest way to picture it is to compare a form with a person at the counter. The form only accepts what has a box. The person listens, asks for what's missing and helps you even if you don't know the exact name of what you're looking for.
The differences your customer notices
This table sums up what changes from the customer's side, which in the end is the only side that matters.
| Situation | Button chatbot | AI sales agent |
|---|---|---|
| Customer writes free text | Sends them to the menu | Understands the question and answers |
| Sends a voice note | Can't understand it | Listens and replies |
| Sends a product photo | Can't understand it | Looks at it and finds the item |
| Asks something off-menu | Repeats the options | Answers with business info or hands off |
| Price depends on size or quantity | Can't quote | Asks for the details and applies your rules |
| Wants to talk to a person | Only if that option exists | Hands off with the full context |
| Writes at 11 p.m. | Replies with the menu | Handles the whole conversation |
A button chatbot asks. An AI agent understands.
The voice note test
There's a one-minute test to know whether your automated support works: send your own WhatsApp a voice note and a photo, the way a customer in a hurry would.
If the reply is a menu, your system isn't helping customers. It's filtering them.
Voice notes aren't a quirk of a few customers. In 2022, WhatsApp said on its official blog that its users send an average of 7 billion voice messages a day. That's a global figure that includes personal chats, not just messages to businesses, but it shows how the people writing to you communicate.
It isn't an edge case in sales either. In the 29 days we measured at JAVAR, a Colombian food-equipment company, its agent received 107 voice notes from customers. At Optimus, in Chile, customers sent 134 photos and the agent sent 1,026 product images. Every one of those messages would have hit a wall with a button menu.
When a button chatbot is enough
To be fair: not every business needs an AI agent. A button menu works well when:
- You sell a single product or service at a fixed price.
- The conversation always follows the same path (for example, confirming an appointment that's already booked).
- You get few messages, almost all during business hours.
- Your customers don't send voice notes or photos.
If that list describes you, a simple, well-written menu may be enough. What doesn't make sense is using buttons for a business that lives on quoting, advising or answering different questions in every chat.
The risks of an AI agent, and how to control them
An AI agent isn't magic. It has real risks, and any serious provider should explain how they control them.
Making things up
It's the best-known risk. You control it by limiting the agent to the information you give it (catalog, prices, policies) and giving it a clear rule: if it doesn't know something, it says so and hands off to the team.
And it isn't a theoretical risk. In February 2024, British Columbia's Civil Resolution Tribunal, in Canada, ruled on a claim against an airline whose website chatbot told a customer they could apply for a bereavement fare after traveling, which the company's policy didn't allow. The company argued it shouldn't be liable for what the chatbot said. The tribunal rejected that: a company is responsible for all the information on its website, whether it comes from a static page or a chatbot. It ordered the airline to pay CA$650.88 in damages, plus interest and tribunal fees.
It's a small-claims decision from one Canadian province, not a general law. But the lesson travels: what your agent says, your business says.
Promising what it shouldn't
Discounts, delivery dates, warranties. You control it with a list of things the agent never does, just as you'd tell a new salesperson: "discounts are approved by a person."
The customer wants a person and can't find one
You control it with handoff rules: when the customer asks for someone, wants to negotiate or is ready to pay, the conversation goes to the team with the full context and the agent stops replying in that chat.
Nobody reviews what it says
An agent needs supervision, especially in the first weeks. Reviewing real conversations once a week and adjusting its instructions is part of the job, not an extra.
How to choose: three questions
If you're torn between buttons and an agent, answer these:
- Do your customers send voice notes, photos or documents? If yes, a menu won't understand them.
- Do your prices depend on something? Size, quantity, shipping zone. If there's math involved, someone (or something) has to ask and quote.
- How many messages arrive after hours? In our published cases, 6 out of 10 replies went out after business hours. We explain it in what slow replies really cost.
Two or three "yes" answers mean your business has outgrown what a menu can do.
Frequently asked questions
Does an AI agent replace my salespeople?
No. It responds, answers repeat questions, qualifies leads and filters out the merely curious. When the customer is ready to buy or wants to negotiate, your team steps in. The agent takes the repetitive work off your salespeople so they can focus on closing.
Can it make up prices?
A well-configured agent only answers with the information you gave it. If the customer asks about something that isn't there, the right move is to say so and hand the conversation to a person. Take it seriously: in 2024, a Canadian tribunal ordered a company to pay for wrong information its chatbot gave a customer.
What happens if the customer wants to talk to a person?
The agent hands the conversation to your team with the full context and stops replying in that chat. Your salesperson doesn't have to ask again for what the customer already said.
Does it work on WhatsApp, Instagram and Messenger?
Yes. The agent handles all three channels with the same information and the same rules, and your team sees every conversation in one place.
A button menu was the best option for years, back when there wasn't another one. Today your customers expect to be answered the way a person would answer.
Run the voice note test this week. If your system replies with a menu, you know what to change.


