GitHub's own Octoverse 2025 report puts the scale of the problem in hard numbers: developers merged nearly 45 million pull requests per month in 2025, up 23% year over year. Maintainer hours did not grow 23%. The result is a trust crisis. tldraw closed their pull requests entirely. Fastify shut down their HackerOne program. The old quality signals, clean code, fast turnaround, low complexity, used to indicate genuine investment in a codebase. AI can now produce all of them in seconds without the contributor understanding a single line.
The author, writing on the GitHub Blog, proposes a triage framework called the 3 Cs: Comprehension, Context, and Continuity. Comprehension means requiring contributors to open and discuss an issue before a pull request is accepted, a gate already in use by Codex and Gemini CLI. Context means demanding reviewable submissions: linked issues, explained trade-offs, and disclosed AI use. Projects like ROOST, Fedora, and the Processing Foundation have each landed lightweight AI disclosure policies. A fourth practical tool is AGENTS.md, a project-level instruction file for AI coding agents, already adopted by scikit-learn and Goose. These are not bans on AI. They are checkpoints that shift cost back onto the contributor.
The stakes go beyond inbox management. The article makes a specific compounding case for mentorship: two well-mentored contributors who each mentor two more every six months produce 59,049 contributors by year five. Broadcast onboarding producing 1,000 per year produces only 5,000. Burn out the mentors and you kill that multiplier permanently. GitHub has published an open RFC for community feedback on platform-level fixes, but the author is explicit that platform changes take time. The frameworks described here are what maintainers are using now. Read the full piece for the implementation details and the links to live contributing policies you can copy today.
[READ ORIGINAL →]