Hacker Timesnew | past | comments | ask | show | jobs | submitlogin

Vibing out slop and then developing that into something notslop can be one way to learn the actual domain.


It's true, and I use LLMs for the vast majority of my development work. The thing I greatly disagree with the anti-LLM crowd is that they often act like usage of LLMs necessitates skill decay. I have learned so much more using LLMs than at any point in my career.

Slopping out something as a way to get to a proof of concept for user feedback is a blessing. It really helps you go from a vague idea to a much clearer understanding of what you do and don't want to build. Scrapping the code from that prototype and then rebuilding it slowly and durably is now happening from a much more informed perspective.

I think what I see as the demonic aspect is the lure of no longer needing to do the slow and durable work.


Can, but do people do that?

I'm reminded of a story of a professor who visited a programming class taught by another professor who was teaching kids about binary and registers and stuff which was boring, and the visiting professor showed the kids how to write 10 PRINT "HELLO YOURNAME!": GOTO 10 which they'd never seen before and got them all excited to learn more about computers. That's the good kind of slop leading into learning.


> Can, but do people do that?

Yes, of course they are. Not the people who are vibe-puking feed-content to try and make a quick buck; The people that are hustler first, craftsperson second. But there are plenty examples of it being done better if you start following projects. A simple example: https://github.com/simonw/tools/commits/main/

Its worth remembering that the word we use for what we do is _development_, and the vast majority of software running right now is undergoing continual development. I swear, there must be whole generation of coddled-coders out there who have never had to work on a legacy codebase, and these chuckleheads must be the source of most anti-clanker sentiment.

In 2013 I started working for couchsurfing, a company that was trying to go from a volunteer-coded and operated web app to a "proper" startup with an engineering team and a set of phone apps to go along with the application. The legacy codebase was worse than anything claude or codex will produce these days; It was a true cronenburg horror of php and mysql filled with a decade of twists, turns, dead and hidden features. Every step of that development project involved de-slopping, and none of it had been produced by AI.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: