@Li @KitsuneofInari So... Nobody can really understand code fully, not even the code they have written. Code has become too complex, and there is just too much of it!

But if you have thought about what your were writing, if you have recorded your thinking in commit messages and comments.

There is a chance you might have a recollection of having created that code when debugging it, ten years later.

If you asked Claude to regexp-slop it for you, not chance.

@Li @KitsuneofInari Also, the fun part of writing software is thinking, coding, testing, seeing people use it...

Doing code-review, not so much. That's what you do to help other people to level up.

But with LLM-generated code, where are the people you want to mentor?

But you still have to code-review the swill.

@halla @Li @KitsuneofInari This is tantamount to saying don’t use tools, or libraries, or graphics cards, if you don’t review their code. I can’t agree with this sentiment. The real issue is the broken trust and overreach of agentic tools.

Should I ditch the Wacom and pick up a papyrus roll and feather quill instead because I don’t review the Krita source code or my tablet’s circuits? No, I trust the authors and their community.

So the question in my mind is: in what skilled hands, with which LLM, and to what extent, does it fall within the boundaries of acceptable use?