The expensive mistakes are rarely technical. Usually someone built exactly what was asked for instead of what was needed.
I care about clean code and fast queries. I care more about whether the thing we built solved the problem someone actually had.
At Kinobi I build for universities and career centres. I'm not in their meetings, but I read the support tickets, the feedback and how people use the product, and most of what I build starts there rather than from a spec handed to me.
Before this I mentored 35+ engineers and ran bootcamp tracks, and won a couple of international hackathons back when I was a student. These days I pair with Claude on a lot of work. It's a real speed-up on the mechanical parts and no help at all on deciding what's worth doing.