When I hit the microphone icon to speak my query, a modal comes up and I have to hit record to capture. It would be nice to just hold a button down as a 'push to talk' button. I would hold (home or whatever) and a little red indicator comes up next to the text box. When I let go, the text box is populated with the generated text.

Much more 'flow' in that case rather than multiple clicks each turn of the conversation.