K.A.A.L OS/A voice agent needs clear boundaries

AI & INTERFACES / WORKING NOTE / 2 MIN

A voice agent needs clear boundaries

Turn-taking, interruptions, and honest feedback are part of the product.

By Viral Prajapati

A voice interface has to answer a simple question at every moment: whose turn is it? Listening, thinking, and speaking should be visible states, with a way to interrupt and a way to recover.

The conversation belongs to a session. Its history, pending requests, and audio playback need a shared lifecycle. Closing the conversation should close the connection, cancel playback, and release speech input.

Speech recognition varies across browsers and environments. A text input is a useful companion: it makes the conversation usable when a microphone is unavailable or a technical term is transcribed incorrectly.

An agent should say what it can actually do. Discussing a discovery call is different from booking one. The interface and the model instructions should keep that boundary clear.