Contributing to open-source software in 2026
I got into contributing to open-source (for me that was llama.cpp) at a uniquely opportune time. About a year ago, coding agents were not nearly as good as they are today, but they were good enough to answer technical questions and act as a "fancy autocomplete". That made unfamiliar code less intimidating, but you still had to do the hard yards: write it, verify it, test it by hand. The sweat is how I actually learned the codebase, and merged PRs kept the motivation up. Looking back, that window has mostly closed.
The situation today is quite different. Agents are here and they are good. They context-switch between layers of abstraction seamlessly, write ad-hoc tests to check their own work, find weird edge cases, infer vague user intent, and just keep going. This has exploded the amount of software being produced, by any measure. If you're trying to get into OSS now, there are some things you should know.
Maintainers are drowning in PRs
There has been an exponential rise in the number of PRs. Initially, around the time agents became usable, they were all slop, so you could just say no. Now they are a mix of slop and non-slop, and you can't tell which from the diff summary. That is incredibly draining for a human reviewer, because the cost of finding out is sometimes reviewing it in earnest.
AI review does help here for the first pass. For bigger features it doesn't, because those need consensus between maintainers on whether the thing should exist and how, and AI can't really drive that yet. So review is the bottleneck, and it's a human one. That means if you submit a solid PR today as newcomer, it will probably be overlooked. This should not discourage you because:
Humans are still merging code
Even in the extreme case where they're just clicking a button, a human is still merging code in most places. Email a maintainer. Most of the time they will help you out if you're sincere. This is how my journey with llama.cpp started: I emailed Johannes, he got on a call with me and explained how I could contribute meaningfully. I've since done the same for a couple of people.
Start small, no slop1
Because of all the above, this is basically an exercise in earning trust. Prove that you're thoughtful about the codebase and interested in maintaining it. Run the tests, fix the CI, be responsive, and don't use AI to respond to review comments. Do the work of seeing it through, even if you didn't write the code. You have to separate yourself from the slop.
Help is wanted
Maintainers are in demand: projects need people who can review and merge code at a faster pace than before. Pick a project that agents can't just inline, and stick around. Those projects are going to be here for some time to come, and so is the shortage of people who can review for them.
Borrowed from the LLVM AI Policy.↩