Hey, thanks for the report. This is actually the current design behavior for voice input in Cursor.
The microphone button is replaced by the Send button after transcription completes, which prevents continuous voice input. This is a known limitation that several users have reported.
There’s an existing feature request for this exact functionality here: Allow multiple voice recordings or re-recording in Voice Mode
I’d recommend adding your feedback and use case to that thread to help the team understand the demand for ChatGPT-style continuous voice input. The team is aware of this limitation and may consider it for future improvements.