They are not the same thing
A chatbot answers questions. It retrieves from material you control and responds. The failure mode is a wrong answer, which is embarrassing and correctable.
An agent pursues a goal across multiple steps and takes actions — booking, updating a record, sending something. The failure mode is a wrong action, which may not be correctable. That difference should drive the decision, and usually the branding drives it instead.
Three conditions before deploying either
- The volume justifies it. Below a threshold, a person answering is cheaper and better.
- The answers already exist and are current. A system over out-of-date documentation will answer confidently from stale material — and confident wrong answers damage trust faster than no answer.
- Being wrong is cheap to correct. This is where chatbots and agents diverge sharply, and where most deployments should stop at a chatbot.
Where a chatbot earns its place
High question volume against documented answers, with an escalation path when confidence is low. Support deflection on a product with real documentation is the clearest case. Qualification before a human sees the enquiry is the second, provided your team can state the criteria.
Where an agent does
A bounded task with a reversible outcome and a review step: drafting a response someone approves, assembling a quote a person signs off, updating a record with an audit trail. The pattern is narrow scope, low cost of error, human in the loop.
The pattern to avoid is an agent with broad permissions on a customer-facing surface, because the risk arrives before the experience does and a visible failure sets the internal programme back further than the project was worth.
When neither is the answer
A five-page site with modest traffic does not need either. Neither does a business whose documentation is out of date — fixing the documentation is the higher-return project, and it has to happen first regardless.
The test: if a competent person could not answer reliably from the material you have, a model will not either. It will just be more fluent about being wrong.