A voice agent is not a chatbot

People use these interchangeably and then are surprised by which one disappoints them. They are not the same category of thing.

A chatbot has time. A visitor typing into a box will tolerate a two second pause. They can re-read. They can scroll back. If it misunderstands, they rephrase, and the cost of that is mild irritation.

A voice agent has none of that. Silence longer than about a second reads as a dropped call. There is no scrollback. If it mishears, the caller cannot see what it heard, so they repeat themselves louder, which does not help. And the caller cannot tell whether it is thinking or broken.

That changes what you build.

Latency is the product. Not accuracy. A slightly worse answer delivered immediately beats a better answer two seconds later, because the second one has already lost the caller. Everything gets designed around that budget.

Interruption has to work. Real people talk over each other constantly. A system that finishes its sentence while the caller is talking feels like a recording, and callers hang up on recordings.

It must know its own edge. The best thing a voice agent does is recognize that this call is past its competence and say so, cleanly, and take a message. The failure everybody has experienced is the system that will not admit it cannot help.

Names and numbers get confirmed, always. Speech recognition is very good and still wrong often enough on proper nouns that an unconfirmed callback number is a lost job. Read it back.

The other difference is stakes. A chatbot that fails costs you a form fill. A voice agent that fails costs you a customer who was ready to buy and now thinks you are the kind of business that makes people fight a robot.

Which is why the only version worth shipping is the one that hands off well.