Open Source Maintainers Are Drowning in AI Code

AI-Generated Code Is Overwhelming Open Source, Not Replacing It

AI-generated code in open source refers to software contributions produced wholly or substantially by coding agents or large language models, which now arrive in growing volumes as pull requests on platforms like GitHub and are forcing projects to rethink contribution rules, mentoring models, and what it means to be a sustainable maintainer community. This is not a theoretical debate anymore; it is a day-to-day operational crisis. Godot Engine is rewriting its contribution policy to bar most AI-generated code from its repositories after months of discussion and an exploding review backlog filled with AI-authored pull requests. At the same time, a large study of 1,888 repositories that adopted AI coding agents shows newcomer participation has not collapsed, even as code complexity and review volume climb. The core tension is clear: AI is changing how code arrives, but humans still have to clean it up.

Inside Godot’s AI Contribution Ban: Protecting Mentors, Not Just Code

Godot’s AI contribution ban is less a purity test about AI-generated code open source and more a defense of human mentoring time. The foundation says maintainers can no longer keep pace with a growing backlog of pull requests, many of them low-effort AI-generated submissions that feel like "vibe-coded" experiments rather than serious patches. The problem is not only GitHub code quality; it is that feedback on these pull requests does not teach a model, and heavy AI users often cannot explain or adjust their own code based on review comments. Reviewing PRs has always been tedious, but it was rewarding because it invested in a person who might become a future maintainer. With a machine on the other end, the mentoring payoff disappears. In response, Godot now prohibits AI from generating any substantial piece of code and requires new contributors with three or fewer merged PRs to get explicit sign-off before attempting big features or refactors. This is a community drawing a hard line to save its people.

Zig, Ghostty, curl: A Broader Backlash Against "Vibe Coding"

Godot is not an outlier; it is part of a growing front of projects pushing back on AI-generated code open source norms. The Zig programming language adopted a similar zero-tolerance policy for AI-assisted contributions, with its community leadership arguing that the whole point of reviewing a pull request is to bet on the contributor, not just the code. If there is no learning human behind the patch, the time investment makes less sense. Other projects, including terminal emulator Ghostty and the C library curl, have restricted or shut down parts of their contribution pipelines as AI tooling increases review burden and false bug reports. This backlash is not driven by abstract fear; it is maintainer triage. When autonomous agents and "vibe-coded" PRs trigger automatic bans in Godot’s GitHub repository, it reflects a belief that permissive AI use is incompatible with sustainable open source maintainer workload. These communities are willing to sacrifice some contributions to keep their review queues human-centered.

What the Data Shows: Complexity Up, Contributors Still Showing Up

Against the gut-level anxiety about AI crowding out humans, the Peking University study offers a more sobering picture of GitHub code quality and participation. Researchers tracked 1,888 repositories that adopted agents such as Cursor and Claude Code, treating adoption as the first committed configuration file, and compared them against matched non-adopters. They found newcomer participation held steady or rose slightly, with the worst case a non-significant 1.5% dip. The striking changes are technical and operational: cyclomatic complexity ticked up 3%–4% across languages, and cognitive complexity in Python projects jumped about 11%. On the exact 128 Python repos where complexity increased, newcomer entry and retention remained stable while the active contributor base grew. Meanwhile, platform-wide merged pull requests have ballooned from about 25 million per month in early 2023 to roughly 90 million a month today, even though maintainer ranks are not growing at the same pace. The data undercuts the "AI kills juniors" story but strengthens the case that review work is exploding.

The New Trade-Off: AI Tooling vs. Human Sustainability

The biggest fear about AI in open source—that agents would seize the easy issues and choke off the pipeline of new human maintainers—has not materialized in the data set studied so far. Newcomers keep arriving; the crowding-out fear is, for now, put to rest. But a different problem is getting worse: open source maintainer workload. Pull request volume has nearly quadrupled, code is slightly more tangled, and contributors may not fully understand the AI-written changes they submit. Policies like the Godot AI contribution ban and Zig’s zero-tolerance stance are blunt instruments aimed at restoring a workable mentoring environment. Other projects throttling their contribution pipelines show the same tension between embracing AI tooling and protecting human capacity. The Godot Foundation already describes its stance as conservative and expects to revisit it as tools change. That is the right framing: these bans are not the final word on AI, but temporary shields while communities figure out how to keep maintainers from burning out under the new flood of machine-written code.