I got a little surprised observing myself working on software development under time pressure. In such circumstances, I tend to make different engineering choices than I "normally" would. For example, I choose more expensive LLM models (like GPT Sol 6.1 or even Claude Opus 5.5) rather than letting cheaper ones like GPT 6 Luna do the job, even though I see them as capable of doing it too.
But let's focus on another one here. Whenever I feel the clock ticking, I almost always prefer speaking than typing, despite the lower transcription accuracy and the chaotic prompts it leads to.
It's probably some cognitive bias (or just a failure of habit) that makes me over-optimize my prompts when I have time, even though the messy, dictated stream of consciousness gets the job done faster. I do it in GitHub Copilot in VS Code, as well as in chat tools (ChatGpt Plus, Gemini).
I think falling back to dictation when things go chaotic must reflect some belief of mine that it's just faster and more efficient. However, I don't know if it's true. Probably someone measured it already. But since I'm still swamped, I'll stop here and not dig deeper today ;)