“Agent” has become the word every chatbot vendor reaches for this year. The label tells you nothing. The test is simple: does it take action in another system, or does it just answer questions from a static list and stop? Most of what gets called an agent still stops.
The one question that separates them
Ask what happens after it answers. A real agent qualifies the lead and routes it. It books the appointment and confirms it. It flags the exception and hands it to a person with full context. If the answer is “then someone reads the transcript,” you bought a chatbot.
Three questions to ask a vendor
What business rules does it actually know: yours, or a generic template? What happens when it is wrong, and how would you find out? What does it do besides talk? If the demo only shows a conversation, you have not seen the product yet.
Why the knowledge layer is the actual product
The difference between an agent that works and one that embarrasses you in front of a customer is not the model. It is the quality of the rules, escalation logic, and business knowledge it runs on. That layer is unglamorous, takes real time to build, and is exactly the part most vendors skip to ship a faster demo.