skip to content
ainoya.dev

I’ve been trying out the voice input feature in Antigravity (specifically testing the UI in version 1.15.8), and it sparked a realization about how we interact with voice UIs.

Most tools display text in real-time as you speak. While technically impressive, this creates a cognitive burden. As the words appear, your brain involuntarily switches from “speaking” mode to “editing” mode. You start spotting errors or rethinking phrasing mid-sentence, which breaks your train of thought.

Antigravity behaves differently. It uses a simple recorder UI that hides the transcription process entirely. You speak, and the text generates only after you finish.

This visual-free approach removes the distraction of immediate self-correction. Because modern AI models are now accurate enough to be trusted without constant supervision, we no longer need to monitor the output in real-time.

I also suspect a technical upside: by decoupling the UI from the immediate transcription, the system might be free to prioritize capturing the broader context rather than striving for strict, word-for-word accuracy.

By hiding the visual feedback, the tool allows you to focus entirely on articulating the thought itself, making the drafting process significantly smoother.