There would be many more programmers in the world if not for the fact that programming is hard, and you need to spend a lot of time learning, practicing, and testing your code.
Then "vibe" coding arrived, and suddenly you could just have an AI write code for you, whether you understood what the code meant or not.
Vibe coding meets open source
Good intentions or not, trouble is brewing
I've written before that vibe coding is risky , but when you think about how open-source software development works, it becomes an even more concerning trend. Many open-source projects are open for anyone to contribute code, which is then reviewed by the project maintainer, who will either accept or reject the contribution.
Even if one particular project won't let you contribute, nothing stops you from taking the open source code and making your own fork where you are the maintainer, and then integrating your code changes.
This worked pretty well when the only people who could contribute code were those who could write that code, because that's naturally a small percentage of people. However, if every person can just generate code using AI, then you can see where this is going.
The flood of AI-generated pull requests and issues
Too much of a "good" thing?
Project maintainers of some open source code repositories have already raised the alarm when it come to the flood of AI code contributions they have to deal with. There's a good piece by JP Capara on Medium (paywalled, sadly) that lays out a timeline showing tldraw closing external pull requests , curl closing its bug bounty , and RedMonk coining the term " Slopaggedon ."
Meanwhile, a preprint academic article from Cornell proclaims that Vibe Coding Kills Open Source after looking at how this phenomenon impacts OSS on multiple levels. The short version is that the way open-source software development has operated until now is absolutely not geared for the advent of AI vibe coding.
Quality, context, and the review burden
No quality or control
All of these pull requests aren't just people vibe coding and then trying to submit that code. People are also using agentic AI to find or invent problems to exploit bug bounties, or to simply interfere with a project by spamming it.
Github now lets maintainers disable pull requests , and the fact is human reviewers simply can't go over the mass of code or requests in any sort of practical way.
Of course, GitHub would love if you used its AI models to combat the AI models of spammers, but I don't personally like that idea, because you're still removing humans from the quality control part of the process. It doesn't solve anything.
Community friction and cultural strain
Open source is built on human community
As was detailed in a video by Jeff Geerling I embedded above, AI agents are sowing chaos in the open source community, even doing truly unhinged things like writing a hit piece about a maintainer after he rejected the agent's code.
The big problem with all of this, in my opinion, is that the only viable option for open-source projects is to become insular. Now, only people who are trusted or who have been vetted can contribute or even be part of the discussion around that software. The pool of contributors and testers will have to shrink to maintain integrity.
Just by doing that, you're stripping away most of the actual benefits of open-source software development. The security and quality benefits of having thousands of coders and testers contribute to your project is what makes this software possible. Development will slow to a crawl, quality will drop, and the sense of community and shared vision will fade.
Can this become sustainable?
Is there a balance with the machine?
The harsh truth is that vibe coding isn't going anywhere, and neither is human coding. In fact, learning to code in the age of vibe coding is probably a smart move. This technology is powerful and useful in the hands of a skilled coder, but outright dangerous in the hands of someone that doesn't understand the code coming out of the LLM.
However, the genie is out of the bottle, which means these projects will have to adjust in a way that strikes a balance between being open and preventing AI slop from destroying everything. There's a need for trust systems, ways to triage what should be reviewed first or at all, and how to integrate human oversight without banning AI entirely.
I won't lie, I think this is going to be a tough nut to crack. Perhaps there's a future where all code is created by AI, and it doesn't really matter. Maybe every person will just have their own personalized apps, but that day may never come. AI coding tools, and LLMs in general are hitting a plateau and aren't good enough to be used blindly by unskilled users. Finding a sustainable solution for open source is literally an existential crisis, and we should all be worried about that.
