Chatbot or AI agent: the same question, two answers
A customer asks if it is in stock and can arrive tomorrow. The chatbot answers politely. The agent answers correctly. Here is the difference.
- Part 2/3
- About a 3-minute read
- Written by Minh Khang
The same question
On the left an ordinary chatbot, on the right an agent connected to stock and delivery.
Polite, but the customer still does not know if it is in stock and may shop elsewhere.
Answers both questions with real numbers and leads the customer to the order.
How do chatbots and AI agents differ?
| Criterion | Chatbot | AI agent |
|---|---|---|
| What it does | Answers questions in the chat window. | Finishes a job: looks up, creates, sends, updates. |
| Connected to | A script or prepared documents. | Real tools: stock, calendar, orders, email. |
| What it knows | The current conversation. | The conversation, the data it looks up and each step's result. |
| Steps | One question, one answer. | Many steps, deciding the next one itself. |
| When it goes wrong | A wrong answer, a confused customer. | A real mistake: wrong order, wrong message, wrong refund. |
| Build and run | Quick and cheap; edit the script and you are done. | Work to connect systems, plus people to approve and monitor. |
When is a chatbot enough, and when do you need an agent?
A chatbot is enough when
- Customers keep asking the same few things: hours, delivery fees, returns.
- The answers sit in your policy and rarely change.
- You only need fast replies when nobody is on shift.
You need an agent when
- A correct answer needs live data: stock, delivery slots, order status.
- One request triggers several steps across several apps.
- You already have someone to approve and a way to log what the agent does.
Common questions
Usually not. Most Messenger and WhatsApp bots follow a script or answer from prepared documents. It only becomes an agent when it is connected to the shop's real tools, such as stock, orders and delivery slots, and allowed to call them to finish a job.
Most small shops should start with a chatbot for repeated questions. Move to an agent only when customers leave because they wait for someone to check stock or order status, and you already have a person to approve what the agent does.
The agent: besides paying for the AI model, there is the work of connecting it to your software and the time a person spends approving and monitoring. A scripted chatbot is cheap and quick to change. Try one small job for two weeks to see your real numbers.
The shop is, towards the customer, just as with a staff mistake. The 2024 Air Canada chatbot case showed a business cannot blame "the bot". That is why anything involving money needs guardrails and approval.
Watch the true story
Chatbot thường và agent nối đơn hàng thật khác nhau chỗ nào?
The video loads in YouTube's player when you press play.
Words in this guide
Each word has its own page in the AI glossary.
Sources
The dated facts on this page come from the original sources below.
- BC Civil Resolution Tribunal (CanLII)
Moffatt v. Air Canada, 2024 BCCRT 149