honest answer? the biggest change wasnt speed on stuff i already knew how to do. it was how cheap it got to explore stuff i DIDNT know how to do. trying three different approaches to a ui problem used to eat half a day. now its like twenty minutes which honestly changes which ideas even get tried in the first place
what didnt change tho: taste. an ai pair will happily spit out five competent forgettable versions of a screen before it makes one good one, and picking the good one is still 100% a human job. if anything the bottleneck just moved from production to judgement, which nobody warns you about lol
the failure mode you really gotta watch for is accepting the first plausible looking answer just cause it showed up fast. plausible and correct are NOT the same bar, especially in code touching money or auth or user data. the review discipline has to go up just as fast as the output volume does, or your gonna have a bad time 😬
