We Are Losing Open Contribution

Projects like Ghostty and tldraw have begun blocking external PRs more strongly. As AI-assisted PRs increase, contributions that pass tests but lack context and follow-through are wearing maintainers out.

The core of the problem is not code generation itself, but the entire contribution process, including understanding, responding to reviews, and maintenance. The author points out that unlike the Hacktoberfest spam of the past, which was filtered out immediately, AI PRs that look fine on the surface but lack substance are now a bigger burden.

On the other hand, there are good examples too. A user who reported a bug to Ghostty transparently disclosed that they were not familiar with Zig or macOS, and used AI to create a crash analysis script, providing clues that resolved 4 actual issues. The key is transparency, communication with people, and an attitude of receiving feedback and iterating.

The author believes that in the long term, line-level attribution that can trace which prompt and context each diff came from, along with cryptographic attestation from AI providers, will be needed. But until that future arrives, the author argues that as a minimal solution available today, AI usage should be honestly disclosed in PRs and commits, and tools like Leeroy that automate this are needed.