We don't need more apps. We need better open source. — Jordan Krueger

AI made it easy to build a new app in a weekend. I’d rather see that weekend spent making the open source tools we already use faster, lighter and easier to run.

Read in full here:

1 Like

Agree with the headline, liked the article, wish he said more.

What I wanted the author to get to was: LLMs have made code so cheap that a Tower of Babel effect is now upon us. The ecosystem is flooded with mediocre, copy/paste tools written in different languages resulting in a lot of noise and squandered effort. The better AI gets at shipping what’s already been created while continuing to fall short of producing things of novelty, the more of this recycling of software we’ll see. Maybe, probably, some of it truly gets better as a result, but hopefully a winner emerges, for standards, for discoverability.

Analyzing the last few years until present, I think GitHub won’t be able to serve the next generation of source control - and we will need a new platform lead by people who know what’s going on and are agile and responsive to the trends (the same way GitHub felt when SourceForge ZIP files were all we had). I think this should-be platform - and for it to get any adoption it’s probably also a Vercel/Heroku type PaaS - simplifies the git workflow and merges the disjointed product development workflows like design, vibe coding, code repositories, deployment/DNS/domains, and possibly even file storage, roles/privileges, and more, into a cohesive UX that is heavily reliant upon LLMs, including decision-making to orchestrate the different product development and IT/administrative tasks needed to build and manage software. I mean it sounds crazy, but, cPanel + AI, but the shell is alive.

The premise of the post - needing better open source - might be limited by open source platforms themselves (e.g. GitHub, GitLab); it might be limited by the way we present open source to the world. Because they haven’t exactly adapted to the changing tides, have they? Until someone does, why would I expect LLMs to produce any other code than what I see on GitHub - and further, why would I expect any developer to not use LLMs when producing code? You’re always going to get GitHub: Rehashed.

I agree, but think the blame is misplaced - it’s not a developer (who are increasingly just LLM pilots) issue but a platform/protocol issue. The way we do open source is now, officially, flawed.

I’d push back a little on the platform part: GitHub didn’t make people write wrappers, the incentive did, since shipping something new gets attention and fixing a slow upstream dependency mostly doesn’t. A new platform with an LLM in the middle could make discovery better, but I’m not sure it changes that incentive. What I’d love is something that says “this already exists, here’s the project to contribute to” before someone starts the weekend app. (I’m with aicost.tools, so I see a lot of near-identical AI wrappers go by.)

Mostly agree, bolting an LLM onto version control would give you a “Vercel v0” like experience, closing the gap between prompt and deployment - since the GitHub repo already hooks into deployment platforms on commit (I guess how v0 is doing it under the hood).

But yeah none of that incentivizes “better software”, or fixing existing software, probably more of the opposite.

That’s why I say open source is now flawed, or at least is no longer an archive of community-supported projects. Modern GitHub is more like a creative zoo of potential ideas, plus cloud storage for code files, and - randomly - a stage for CI/CD.

My GitHub UI this morning has a new toggle in the top-left: A pin icon with “Home” and “Feed”. You can choose either one as the permanent home page. Apparently, some of the users are still using GitHub as “social coding” (one of their original slogans), while others are using it like a personal workspace - presumably for building and deploying software with external tooling.

Outside of gatekeeping open-source contributions (which already happens), or implementing some App Store esque review process to anything called a “release”, or anything classified as a “Community Repo” - giving them special treatment in the UI and enforcing certain open source standards - I wonder how GitHub and those who copy them (GitLab etc.) will pivot to handle the growing use case divide. My projection is that they’ll have to choose one primary reason for existence, and another platform or method will serve the other use case(s).

Your idea of AI basically doing a “Did you mean:” with existing code repos is a great idea, and I would not be surprised if Gemini and other web search capable LLMs implement this alongside the external links they’re already provide for citations. Though, it does seem susceptible to ads… I imagine you’d start getting paid suggestions.

IMO, the emergent winner will be a desktop app (AI-capable IDE/creative suite), that runs local models for code completion and local version control, that somehow simplifies the dev process greatly in such a way that it widens the developer audience, with the online/cloud marketplace of code being a lot more locked down by review processes similar to publishing on the App Store, or getting a book published. Idk if this is Microsoft/GitHub though, even though they are so well poised, but I doubt it.