Hacker News new | ask | show | jobs
by steeve 164 days ago
nobody cares about the website being done with AI because the code of the project itself is not AI

you need to touch grass

1 comments

The code of the project absolutely does look like it was done with AI lol, it’s a single commit…
Claude did rewrote lots of my original messy code. No shame in that? But in the end the interest was in the underlying architecture, applied to nats protocol. Anyway.
Which is totally fine, I use Claude Code a bunch myself. I said nothing about shame, just that one single commit plus a website that looks vibe coded together has all the hallmarks of AI-driven code.
Dude, when I move projects to GitHub I also often collapse everything into a single commit.

I do this to avoid having to check e-mail addresses and names in commits - maybe I mistakenly made a commit from my work account etc.

After the “initial” commit making it all public, I start to work “in the open”. I see many others doing it the same way.

That is NOT a reliable indicator of slop!

And the author has admitted at least some assistance here:

https://news.ycombinator.com/item?id=46452907 https://news.ycombinator.com/item?id=46452841

There may have been less pushback if this had been expressed up front. But also, what is it ? Is it to "test the architecture applied to nats" or is it to be a fully fledged NATs replacement (as per the impression given by table at the bottom of the website) - which becomes much harder if AI has significantly re-written the authors original code (and commented it badly).

The website being AI coded I can take or leave.

Fair point. And as what it is, not a nats replacement, certainly dont have the time to maintain that this way, a test/tech demo/fun side project that yielded super interesting results is probably the answer. As usual I'm probably way too enthusiast when I see some nice results like that and the goal here was to talk about that, but it shifted super fast. So yes Claude rewrote lots of parts, and that's what I love about it. Testing an idea happens in way less time than before, and I find that super cool.
It is super cool. People just want others to be honest about it in my experience, for better or worse.
Nothing individually is a good indicator of slop in itself, a human could also have written this readme full of Claud-isms and a borked ASCII schema or the code littered with idiosyncratic comments.

It's the convergent set of clues that makes the case.

Good thing it's not the only signal then.