What’s the future of open source?

Recently I’ve been reflecting a lot on the state of open source and online communities built around software projects. About a decade ago, I wrote an article, mostly not knowing what I was saying, about why doing doing a project “open source first” was a good idea, then from 2017 took the maintainer role on ExternalDNS and while I can’t say I am the most expert or the most prolific maintainer ever, I have my share of opinions. I’m not writing here to throw my opinions on the other side of the fence. I’m writing because, from what I am seeing with the LLM hype, I am not sure where this open source thing that we used to do is supposed to go and how as a maintainer I am supposed to go forward. Let’s see what is going on.

LLMs everywhere

“Coding is solved” they say. While today’s agent can produce results that are a working piece of code starting from prompts and spec, I don’t think “coding” is solved. In that sense though, the act of writing software has become easier and faster and the generation of small focused tools has become quicker. At the same time, we have the explosion of AI slop, with repos inundated by, frankly, bullshit PRs, so much that GitHub had to implement a feature to allow repositories to stop PR contributions. Without going too much into this (full disclosure: I am a GitHub employee, my freedom to share opinions on this subject is limited), as projects start refusing PRs, the era of contributions to open projects has changed. I don’t think this is limited to accepting other people’s contribution, rather to the essence of other people’s contribution. Code is liability, so why should open source maintainers accept other people’s code in a low trust context when they can’t trust it even less? Why should a maintainer of an open source trust that it was tested and not only LLM-generated? If code generation is so cheap, why wouldn’t a project maintainer just request a prompt and then review/test what the agent has built, removing the human part? Other than “the human part is important” or “I hate AI” (which are all valid, albeit subjective), those are all valid questions IMO.

So what’s gonna happen?

I have no idea. I know that things have changed. Code is still code, it will still be built, possibly even more. I definitely see more artisanal projects that will totally discard LLMs and that is obviously fine, maybe even desirable, especially for things that are well built and that want to stay away from a dependence on model providers. But if the majority of people and projects switch to fully generated code, the act of contributing code to a project may change forever. For now, my mind will keep buzzing with those questions and, not being able to predict the future, I’ll stay on the lookout.