I actually started writing a very similar essay, but the hyperbole got too out of hand – open source isn't dying anytime soon.

I do think that SDKs and utility-focused libraries are going to mostly go away, though, and that's less flashy but does have interesting implications imo.

https://meelo.substack.com/p/a-mild-take-on-coding-agents

I'm inclined to agree somewhat about libraries. I'm not entirely certain that it is a bad thing.

Perhaps it would be more accurate to say libraries will change in form. There is a very broad spectrum of what libraries do. Some of the very small may just become purpose written inline code. Some of the large, hated-but-necessary libraries might get reduced into manageable chunks if people who use them can utilise AI to strip them down to the necessary component. Projects like that are things that are a lot of work for an individual that make it easier to just bite the bullet and use the bloated mass library. Getting an opportunity to make an AI do that drudge work might lower the threshold that some of those things will be improved.

I also wonder about the idea of skills as libraries. I have already found that I am starting to put code into skills for the AI to use as templates for output. Developing code in this way would let you add the specific abilities of a library to any skill supporting AI.

A simple is this https://htmlpreview.github.io/?https://github.com/Lerc/JustS... which was generated by a skill that contains the source for the image decoders within the skill itself.