This is a special post for quick takes by Simon DeDeo. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
I'm involved with a small open-source project, and the maintainer contacted me about an unusual series of pull requests.
In particular, the maintainer and I had used Claude to find some cheap performance improvements; I had filed these as "issues", not wanting to just AI slop a nice project. However, at the same time as that was happening, a new contributor showed up and produced clean pull requests for the main issues raised. The person seems to be using AI-generated text 100% of the time for communication.
We looked into his history, and while he claims a Microsoft affiliation, it's not particularly plausible; he's also apparently been trying pull requests at larger projects. None of them cluster around an identifiable set of interests (e.g., one project was a "festival events management" project).
One possibility is that it's a real person using AI to karma farm, building credibility either for a job application, or something stranger.
Another is that it's actually an AI doing the same.
Has anyone noticed this? The project maintainer and I are both not particularly hooked in to large-scale open source dynamics these days, and I'd appreciate any thoughts people have.
Yes, all open-source repos are like this now. Virtually all PRs are agent-written, and after a while you recognize the issues that code agents are attracted to the scent of: Any post that diagnoses an issue, suggests a fix, but does not immediately implement it will cause an agent to jump in and implement it ahead of you. The agents are often very over-trusting of the suggested solution.
I don't think there's any grand conspiracy here, just a mix of autonomous agents plus coding beginners and CS undergrads or similar. Some genuinely want to help, some are just farming contributions for a resume.
Some genuinely want to help, some are just farming contributions for a resume.
If I had criminal intentions, I would be doing exactly the same thing: farming contributions so that other people will trust me when they see my edit history. And when the moment comes, I would deliver the important backdoor, and the maintainers would just click approve because they see a friendly contributor with a long history.
Thank you, yes — it seemed a little like the automated text we get as professors; it seems like a weird thing to do (or, rather, a thing that will only work for another few months, until the people who read resumés start to catch on.)