Totally agree. AI has gone completely off the rails with weird language, that is frankly hard to follow. It's suggestions are often quite difficult to reason about and I think that has more to do with the models choices than with our distance from the code. It could make much simpler, clearer, more helpful suggestions but often makes convoluted suggestions that might be technically right but make no sense.

Claude dumps 3 pages of response, you try to read some of it but it's so fragmented and weird, so you skim to the bottom to see what it's asking for next and it says something like "Do you want me to charge the gap on cell length fortification or drop the bagel bites on stdout with Q1 and Q3 before testing the edges on spec? Say the word and I'll refactor the baseline.", and you're like "WTF is Q1 and Q3? Bagel bites? Am I tripping? I've never heard of any of these terms before. Have I been hacked?" LoL