Yes, Opus 4.6 is pretty impressive. I'm using it heavily as a sort of executive dysfunction and dysthymia therapy tool and it can run, like, 10 projects of varying sizes in the background while I work/focus on one of them myself. And with 4.6 it's finally good enough code (sometimes better than I would've written, don't think that any amount learning would've allowed me to write a dependently typed language) that if you have a strict planning workflow with tight guardrails (thank the deities for NixOS VM tests), then you can leave it doing it's own thing for quite a while and get out good enough code out of it.

But if you actually go and submit that upstream yourself as-is then wat. The model still does not actually understand what it had written. That it can appears as if it does is very impressive, but that is gone as soon as the context is and can't replace proper understanding and learning. And why are you sending out code nobody understands (or very soon won't), then it basically is offloading the comprehension of your code on the upstream, which sucks. If you write with models, which I think finally stopped being a malpractice with Opus 4.6, then the minimum courtesy you can do is working with its output enough that you don't have to add it as a co-author and have your intellectual property claim hold up in court (which is actually an even stronger reason to reject purely slopped contributions - nobody wants to suddenly learn that their GPL claim was oveturned in court because some silly contributor didn't have IP over the code they "written").

And youre all still successfully ignoring all the moral issues of using these LLMs:

1. Burning the planet even more than were already were: check
2. Preventing people from buying new computers or upgrading old ones: check
3. Stealing everyones copyrighted works and profiting off of them: check