AI chatbots vs AI agents: what's the difference?

What a chatbot does

A chatbot replies to questions using content you have given it — your pages, documents and FAQs. Done well, it deflects a large share of routine enquiries. It does not change anything in your systems.

What an agent does

An agent is connected to tools and allowed to use them. It can check availability, create a record, send a quote request onward or update a status, then report what it did.

• Chatbot: 'Our lead time is usually two weeks.'
• Agent: 'I've checked your order, it ships Thursday, and I've emailed you the tracking link.'

Which one to choose

If most of your inbound messages are questions with stable answers, start with a well-fed chatbot. If your team's time goes on doing things after the question is answered, the agent is where the saving sits.

Many businesses end up with both: one conversation, with the ability to act when it is safe to do so.

What each needs from you

A chatbot needs accurate, current content. An agent needs that plus clean access to your systems and clear rules about limits, approvals and when to hand over to a person. The second is more work, and worth more.

A practical side-by-side comparison

A chatbot is built to hold a conversation. It answers questions, often from a fixed script or a knowledge base, and its job ends when the conversation ends. An AI agent is built to complete a task. It can still talk, but it also takes actions in other systems: creating a booking, updating a record, sending a follow-up email or raising a ticket.

That difference changes how you plan, test and govern each one. A chatbot mostly needs good content and sensible guardrails on what it says. An agent needs all of that plus carefully scoped permissions, clear hand-over rules and monitoring of the actions it takes on your behalf.

• Chatbot: answers questions; low risk; quick to launch.
• Agent: answers and acts; needs permissions and logging.
• Chatbot success: fewer repetitive questions for staff.
• Agent success: tasks completed end to end without re-keying.

Which one should you start with?

If your main problem is that staff spend hours answering the same questions, start with a well-trained assistant on your website or inbox. It will teach you what customers really ask and give you the content you need later.

If the real pain is what happens after the conversation, such as typing details into three different systems or forgetting to follow up, an agent is the better investment. Many businesses take a staged approach: launch an assistant first, then add actions one at a time once the answers are reliable.

Be wary of tools sold as agents that are really scripted chatbots, and of agents given broad access to everything on day one. Start narrow and widen permissions only when the logs show it is working.

Typical UK costs and running considerations

Off-the-shelf website chatbots are often available on monthly subscriptions, which suits simple question answering. A custom assistant trained on your own documents, with your branding and lead capture, usually involves a one-off build plus modest monthly running costs. Agents that connect to several systems cost more to build because each integration needs designing, securing and testing.

Whichever you choose, budget a few hours a month for someone to review conversations, correct weak answers and update the knowledge base. That review time is what keeps quality high after launch.