Voice-dictate prompts and prose, keyboard for syntax — 3–4× throughput where it counts, sub-500ms latency

Ryan Shrott (Medium)·low signal

Speaking runs 150–200 WPM vs. typing at 40–60, a 3–4× gain that pays off specifically on AI prompts, PR descriptions, tickets, docstrings, and review replies — not on raw syntax. The discipline that makes it work: match tool to task (prose by voice, syntax by keyboard, edits by keyboard) and keep transcription latency under ~0.5s, because a longer delay between speaking and seeing text makes your brain distrust and abandon the tool. On-device Whisper builds (Wispr Flow, Superwhisper, DictaFlow) keep it fast and private for everyday hardware.