Had a really interesting maintainer experience today that I think perfectly captures the nuance of the current AI coding situation.
I got an AI-generated PR on Paperback from a new contributor with a very small GitHub profile. My default stance isn't anti-AI and I've even used Claude on Paperback's codebase as recently as yesterday, but open source is getting flooded with automated slop, so I left a polite comment asking for a bit of human-written justification for the feature and confirming if they had actually tested it manually, and I received a well-thought-out, genuine response. The contributor is a long-time computer scientist who doesn't know Rust, using Claude Code to build an accessibility feature to make Paperback optionally match NVDA's browse mode navigation that directly improves their daily workflow. They tested it on their end, I double-checked the diff and edge cases on mine, everything looked great, and it got merged.
To me, this hits the exact middle ground of how AI in OSS should work: A knowledgeable developer used a tool to bridge a gap and ship a genuinely valuable accessibility feature to an app they use. If I had a strict "no AI" policy, this contribution would have been shut down immediately, and both the project and the user would be worse off for it. But still, human oversight is mandatory. If maintainers just let everything slip through without pushback, repositories quickly get overrun with untested slop PRs.