The difference is simple to state and easy to get wrong in a purchase decision: a chatbot answers, an agent acts. A chatbot's job ends with a message. An agent plans steps, uses tools, and is responsible for a finished result.
Choosing wrongly costs money in both directions — an agent where a chatbot would do is over-engineering, and a chatbot where work must be completed leaves staff copying answers into other systems by hand.
What a chatbot actually does
A chatbot holds a conversation: answering questions, asking clarifying questions, and pointing people to the right information. Grounded in your documents (a RAG chatbot), it can answer 'what is our refund policy?' accurately and cheaply.
Its boundary is the chat window. It does not file the ticket, update the CRM, or check the order status unless someone builds those actions for it — at which point it is becoming an agent.
- Best for: FAQs, help-centre answers, lead qualification questions
- Output: a message
- Risk profile: low — a wrong answer can be corrected in conversation
What an agent adds
An agent receives a goal, breaks it into steps, calls tools (search, databases, APIs, documents), checks results, and returns completed work: a triaged inbox, a prepared quote draft, a reconciled report.
That power needs guardrails chatbots do not: permissions per tool, approval steps for sensitive actions, logs of everything it did, and an evaluation set proving it completes the job reliably.
- Best for: multi-step work across systems with a defined 'done'
- Output: completed work, not just text
- Risk profile: higher — wrong actions have side effects, so review steps matter
How to choose
Ask what should exist after the interaction. If the answer is 'the person knows what to do next', build a chatbot. If the answer is 'the task is finished and recorded', you need an agent — or an agent that talks like a chatbot on the surface.
Many products need both: a conversational front door, with an agent behind it for the workflows worth automating. Starting with the chatbot and adding agent actions for the two or three most repeated requests is usually the cheapest path.
Common questions
Yes, if it was built with access to your real data and APIs in mind. Adding tools, permissions, and evaluation turns the same interface into an agent gradually.
A chatbot answers in one or a few model calls. An agent may make many calls and tool requests per job, so cost is measured per completed task, not per message.
Related services
Working on something like this?
These are the services behind this guide. Tell us about your project and we will reply with a useful next step — not a sales script.
