I was getting useful coding work done with GPT 3.5. I think devs saying "the models are finally good enough" this year are just trying to save face from their own previous irrational denials.