Why haven't I submitted my patches? Because my agent wrote them and I didn't review them. I would feel bad submitting my patches without looking at them, and I don't have time to look at them.
KMyMoney stopped importing (some) OFX files for me. I asked an LLM to debug.
It was a wild journey. An obscure bug in some QT library that was triggered because of something completely unrelated that I had installed briefly in 2019 (and uninstalled years ago).
I had it fix the issue by cleaning the leftovers from 2019, but it's still a QT bug. I had it do a writeup, but I'll never submit it. It's fairly sophisticated and I don't know QT code at all.
I believe I do have some sort of a reproducer, but I likely wouldn't understand even the reproducer script. I think it won't be taken well if someone files a bug, says "Here's a script that reproduces it", but himself doesn't understand what the script is doing.
For bug reports (at least IMO) a minimal example to reliably reproduce the error without any fuss is the basic building block. The reporter doesn't need to understand why it works let alone offer an explanation. In fact for complex projects more often than not any purported explanation will be subtly incorrect and actively misleading. If anything further "documentation" of bugs is unwelcome. Unambiguously demonstrate the defect and GTFO.
I'd still be wasting someone's time.
I'll probably put a write-up on my blog one day, though.
If an end user encounters an error while staying within the bounds of "expected usage" then the bug is guaranteed to be valid. You just don't necessarily know whose bug it actually is (for example a broken library versus broken usage of the library by the consuming program).
I've seen the same thing in Go, for what it's worth.
The code totally works, I'm not a denier there, but the interfaces are heinous if you don't keep an eye out. Mine wrote a Go function that passed a name variable as a string, but also embedded in 2 separate structs that were arguments. I've also seen it do re-validation of data multiple times in internal functions so everything has to return an error.
It's also stuff that sort of doesn't matter if an LLM is the only user. It probably costs some extra quantity of tokens, but not enough to be worth hand-grooming the code.
It's honestly not worth it. I understand both sides here, but people are going to have to come up with new ways of collaborating. Pull requests are likely not it anymore.