update on this since a lot of people seem to relate.
i think i’ve found a solution to my problem.
the issue is not necessarily that i change my mind about implementation.
i feel pressure to speak the perfect prompt as soon as the mic turns on. this is because i was using 1 to 1 speech to text which transcrivbes everything.
what i really need is a post-processor that will take my ramblings, rewording, and filler words and distill it into an actionable prompt for the agent.
i’ve been trying out this app Utter(not affiliated) that was recommended in the comments and it’s been great so far. (gonna check out superwhisper too)
took some fiddling to get it set up, but i’m using the free on device models and i had claude write a custom post-process prompt.
now, i feel like i can speak for however long i want and come out the other side with a polished prompt.
still adapting to using voice first, but this is much better than /voice in Claude Code or anything else i’ve tried.
i feel like i don’t think linearly enough for speech to text prompting.
i frequently change wording as i write so the pressure of having to voice the right prompt first try is frustrating