Why do LLMs love fallbacks and defaults and the silent swallowing of errors? In almost every way GPT-5 is a better at designing abstractions and structuring code than me, but GOODNESS GRACIOUS these models love to build a rat’s nest of fallbacks.