Resurrecting Abandoned Projects with Vibe Coding

by Rian

Open source projects always had a huge caveat in its system. If it’s too rigid and far too common, it is less likely to be given enough attention for maintenance until it breaks. If it’s too early in the development and quite niche, it is also less likely to be given enough attention. The projects I usually encounter and enjoy reading are usually in the later category. Coming from that very same niche, — that’s how I ended up googling for it — I understand the desire to come up with an algorithm. These abandoned projects never come to fruition, mostly because the solution to the problem is found but implementing it to the code can be a tedious task.

To start off with some contexts, I usually don’t just import the code base from another project to be used in my DIY. I often go through the process of rewriting them in my own style, otherwise important nuances I didn’t catch in the original code maybe left behind. It’s like active reading or marginalia, only that I expect the scribbles to be algorithmically similar to the original.

With Claude Code, the process of going through the original code has shifted significantly. I can give it a reading of what I understand the script to be doing and extract the part of the codes I want. Needless to say, this isn’t the kind of vibe coding that most people would associate with the practice. It is doing, by some arguments, “assistant” behaviors. But it is, no doubt, indispensable to do a home lab project in a limited timeframe.

I believe, given enough time and lower price of entry, some of these open source projects, including niche ones, will be available in functional form then what they are now. In fact, quite many of them only have “good bones”, but simply get abandoned after the initial commit on GitHub. I suppose the spirit of being an open source project is that someone else will contribute toward finishing it. Perhaps with the LLMs, open source community will see some finalized projects as first commit, all done solo.

This also raises an interesting question about AI generated media, how misplaced our concerns are. The debates surrounding AI generated works tend to be focused on graphical contents. So I find myself thinking — where were you when AI text slops were overflowing? Well before AI industry hit the general “art” department, it had been aiming at literature of any genre, be that as it may, written in programming language. Programming language follow stringent grammar and syntaxes with well documented style of writing. If anything, they are the most affected, as one of the commonly cited principle of programming is ‘one obvious way to do it’. Finding that one code doesn’t have to be done with human labor anymore. If real artists are weathering the storm, I think the damage to the real programmers went unnoticed by wider public.