Voice Agents Can Just Do Things — Charlie Guo, OpenAI

A useful voice interface can respond by changing the application instead of speaking. This OpenAI developer-experience talk proposes connecting spoken requests to the actions an app already exposes, such as its endpoints and event handlers. Those actions become tools the model can call.

The result can be visible feedback: a completed field, a changed button or a cursor moving through the interface. This makes voice a way to operate existing software, with feedback chosen to fit the task. The design question becomes which actions benefit from speech and how the application should show what happened.