AI explained
AI agents vs chatbots: what's the difference, and which do you need?
By Intact Media Last reviewed 7 min read
On this page
Three levels, in plain English
1. Rule-based chatbot. The classic website pop-up with buttons: “Book a service”, “Opening hours”, “Talk to us”. It follows a decision tree. Predictable, but frustrating when the visitor’s question doesn’t fit a button.
2. AI chatbot. Understands questions typed in everyday language, such as “do you do gutter cleaning in Tauranga and how soon?”, and answers from information you’ve given it: services, areas, hours, approved pricing wording, policies. It can collect contact details and pass the conversation to your team.
3. AI agent. An AI that can also do things. It decides which step to take next and uses tools you connect, such as checking availability, booking appointments, creating CRM records or sending a confirmation. The word “agent” is used loosely in marketing, so ask exactly what actions it can take.
What can each do for a small business?
| Task | Rule-based | AI chatbot | AI agent |
|---|---|---|---|
| Answer common questions | Limited | Yes | Yes |
| Handle unexpected questions | No | Yes, within approved info | Yes, within approved info |
| Capture lead details | Yes | Yes | Yes |
| Book appointments | Via a link | Via a link | Directly |
| Update your CRM | Basic | Basic | Yes |
| Multi-step tasks | No | No | Yes, with limits |
What are the risks?
Wrong answers. AI models can sound confident and be wrong. Limit the AI to approved information, tell it to say “I’m not sure, let me get someone” when it doesn’t know, and review conversations regularly.
Too much access. An agent connected to your systems should only be able to do the specific tasks you’ve chosen. Logging and human approval for anything important are essential.
Privacy. The Privacy Act 2020 applies to information visitors share in a chat. The Privacy Commissioner’s AI guidance recommends transparency, a privacy impact assessment and human review before acting on AI outputs. Tell visitors they are chatting with an AI.
MBIE’s voluntary Responsible AI Guidance for Businesses is a useful checklist for these decisions.
Which do you need?
For most small NZ service businesses, the sweet spot is an AI chatbot with a small number of actions: answer questions from approved information, qualify the enquiry, and book or hand over. That covers the bulk of website conversations with low risk.
Full AI agents make sense where there is a high volume of repetitive, multi-step work with clear rules, and someone in the business to supervise it.
If you’re not sure, AI consulting starts by mapping where AI would actually save time, and where it wouldn’t. Or see how we build AI chat and website agents.