> If chunks of code are small, modular, self-contained and trivial to copy-and-paste, the community can thrive via horizontal gene transfer. For any function (gene) or class (operon) that you write: can you imagine someone going "yoink" without knowing the rest of your code or having to import anything new, to gain a benefit? Could your code be a trending GitHub gist?

This is interesting. I don't do frontend, but from what I understand this was the biggest factor in the popularity gains for libraries like shadcn. Get everything as a c/p block of code. Adapt for your use-cases, but you now "own" that piece of code.

Incidentally this is why I think the first "AI" libraries were such a mess. People were trying to build heavily abstracted one-liners instead of building blocks (langchain et. all).

Coming back to the article, I think there's a lot of merit behind this idea. For one, we could have a repository of "things that have been done before", and have LLMs treat that as a "base example" from where they can start to edit and adapt to the current task. My only concern is that it'll inevitably become enshitified by marketing and spam. And some security concerns on top. But it's an interesting direction.