I really don't think that's how it works. Smart, experienced developers who thought coding agents were junk for most of 2025 and think they're useful now in 2026 are not saying that because they got "worn down over time".
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.
As a smart, experienced developer who's getting worn down over time, I disagree.