←back to thread

1479 points sandslash | 4 comments | | HN request time: 0.832s | source
1. amai ◴[] No.44316317[source]
The quite good blog post mentioned by Karpathy for working with LLMs when building software:

- https://blog.nilenso.com/blog/2025/05/29/ai-assisted-coding/

See also:

- https://news.ycombinator.com/item?id=44242051

replies(2): >>44316635 #>>44323062 #
2. yusina ◴[] No.44316662[source]
Brutal counter take: If AI tooling makes you so much better, then you started very low. In contrast, if you are already insanely productive in creative ways others can hardly achieve then chances are, AI tools don't make much of a difference.
replies(1): >>44316717 #
3. floppyd ◴[] No.44316717{3}[source]
As someone who is starting very low — I very much agree. I'm basically a hobbyist who can navigate around Python code, and LLMs have been a godsend to me, they increased my hobby output tenfold. But as soon as I get into coding something I'm more familiar with, the LLMs usefulness plummets, because it's easier and faster to directly write code than to "translate" from English to code using an LLM (maybe only apart from using basically a smarter one-line tab completion)
4. mkw5053 ◴[] No.44323062[source]
I like the idea of having a single source of truth RULES.md, however I'm wondering why you used symlinks as opposed to the ability to link/reference other files in cursor rules, CLAUDE.md, etc. I understand that functionality doesn't exist for all coding agents, but I think it gives you more flexibility when composing rules files (for example you can have the standard cursor rules headers and then point to @RULES.md lower in the file)