Well so far we've been using humans to clean up the code that humans ruined in the first place, it's hardly a logical fallacy

LLMs are as good as the operator directing them.

I’d say I fall in the “AI skeptic but willing to use it” category. If LLMs can actually clean up after themselves it would actually be a game changer.

I came here to post the same thing. But this is addressed later in the same paragraph:

> But hopefully better development practices, with a human in the driver’s seat, and a focus on reducing technical debt and writing idiomatic Zig, mean that in a few weeks or months there will be a presentable codebase that serves as a drop-in replacement for Rust Bun 1.4.0.

So the intention here is simply more steering. Or perhaps better steering (code structure appears to be a matter of taste... I had a very perplexing chat with a friend yesterday who insisted that four backend processes were required to serve a single HTTP request...)

This was always the plan. Code is no longer written by humans, code is LLLM output of broken slop that is mostly software-shaped, impossible for humans to read and make sense of it (because it makes no sense). So the only solution is that all code is intended to be read and written by machines. The human-in-the-loop era was always a stopgap.

This is the present and future, willingly or not.

Engineering is about design and plan and architectural integrity and qa. Not writing code anymore.

Yeah, good luck with that one though. LLMs are terrible at deslop, or they wouldn't slop in the first place.

You have no idea, to slop-circle is getting more common.

In 2025, sometimes I decided to put a slop LinkedIn post to ChatGPT (back then) to deslopify.

Now, the slop circle is everywhere.

Your PO uses automations to create tasks? Half of the comments are bot slop? Links to documents never edited or read by another human? A lazy three sentence description of a task requirement you would have gotten previously sounds like a dream. Now everything generates pages long texts and you can't possibly go through the slop without AI agents. You need your agents comb through the slop and demystify the task for you.

Same with the other end... Slop PRs will be reviewed by your agents, hoping you catch 2-3 issues so that you can pretend you did a review. The author didn't do a review on their generated slop, but sure, let's pretend reviewers still review code.