A lot more tricks can be learned from just watching AI work. Instead of allowing AI to work autonomously, go back to the old days where you manually approve every command the AI runs. Just recently while doing performance optimization work, I found Opus using the `perf` command in ways I didn’t know possible. Just give AI a real task and carefully read what commands are used by the AI to solve the problem; most likely the AI uses a trick or two that you didn’t know existed.
This makes me think of something I hate about how LLMs write CLI scripts. Even with the same model on the same day, there's never any consistency with the way CLI arguments are written or handled. Sometimes they want to split on an equal sign, and specifically write that into the code. Other times they write code to pair keys and values based on spacing and order, even and odd.
--date=2026-09-16
--date 20260916
and all variations thereof. It's annoying enough to have to check which script expects what format that I probably should've wasted some time or tokens to align them all, or at least put it into an MD file. But it's just an illustration of how LLMs can make CLI commands more obscure than they need to be.
kccqzy · · focus · HN ↗
noduerme · · focus · HN ↗
--date=2026-09-16
--date 20260916
and all variations thereof. It's annoying enough to have to check which script expects what format that I probably should've wasted some time or tokens to align them all, or at least put it into an MD file. But it's just an illustration of how LLMs can make CLI commands more obscure than they need to be.