"AI agent" has become the label of the year, and many products that were called chatbots last year now carry it. The difference isn't a matter of branding. It changes what you pay, what can go wrong and how much setup is involved, so it's worth being able to tell them apart.
One question tells them apart
Ask the vendor, or yourself: "After a conversation ends, what has changed in my business?"
- If nothing has changed, and the customer simply got an answer, it's a chatbot.
- If something has changed (a booking exists, a sheet was updated, a reminder was scheduled, a ticket was raised), it's acting as an agent.
Side by side
| Chatbot | AI agent | |
|---|---|---|
| Main job | Answer questions | Complete tasks |
| Works from | Your FAQs, price list, policies | Your information plus live systems |
| Connects to | Usually just the chat channel | Calendar, CRM, sheets, billing, WhatsApp templates |
| Typical example | "Our timings are 10 to 8" | "Booked you for 6:15 tomorrow. Reminder at 9 am." |
| What can go wrong | A wrong or vague answer | A wrong action: double booking, wrong reminder |
| Setup effort | Lower | Higher: connections, permissions, testing |
| Running cost | Mostly AI usage and messages | The same, plus more AI steps per conversation |
Modern chatbots are built on the same language models as agents, so they're far better than the old button-menu bots. A good chatbot understands free-text questions in Hinglish and answers naturally. It just doesn't do anything afterwards.
For a fuller explanation of agents, see what agentic AI is.
Which one does your business need?
Start with a chatbot if:
- most messages are the same handful of questions (timings, prices, location, availability),
- the next step is simple and a person can handle it (call back, walk in),
- you want to learn how customers talk to an AI before connecting it to your systems.
Consider an agent if:
- most conversations end in the same task, such as booking, rescheduling, order status or payment reminders,
- that task is rule-based and happens many times a day,
- delays cost you customers, for example enquiries that arrive after hours.
Many businesses go from one to the other: a chatbot first, then booking added once the answers are reliable.
Questions to ask a vendor selling an "agent"
- What actions can it take, exactly? Ask for a list, such as "create booking", "send reminder" or "update order status".
- Which actions happen without a person approving them? Refunds, discounts and cancellations usually shouldn't.
- Where is every action logged, and can I see the log?
- What happens when it's unsure? It should hand over to a person, not guess.
- How do I pause it? One switch, not a support ticket.
If the vendor can't answer the first question with a specific list, you're probably looking at a chatbot with a new name. That's not necessarily bad, but you should pay chatbot prices for it.
Cost, roughly
Agents usually cost more to set up than chatbots because each connection to a system has to be built, tested and secured. Running costs scale with the number of conversations and actions, and on WhatsApp, Meta charges per delivered template message on the WhatsApp Business Platform, with marketing messages costing far more than utility or login messages. Our AI cost guide explains each line, and our own prices are published for both chatbots and agents.
Checklist: chatbot or agent?
- List your 20 most recent customer conversations
- Mark how many ended in a task (booking, order, reminder) vs just an answer
- If most were answers only, start with a chatbot
- If most ended in the same task, write that task down step by step
- Decide which actions an agent could do alone and which need approval
- Ask vendors for a list of actions, the approval rules and the log
For the guardrails every agent should have, see human in the loop.
