i think a big thing about pure vibecoding rn is that like, the ai's make so many odd architectural decisions that i never would for w/e reason. and i think people who do purely vibecode are oft not considering what exactly the model will do, and wont prompt to prevent it from making poor decisions.

i think it sucks how most AI research is it at private, for profit institutions with clear profit incentive, and thats the research the field takes the most seriously as well.