AI Code Contributions: Speed Without Understanding
AI code contributions are software changes generated wholly or mostly by machine-learning models or coding agents, submitted to open source repositories by humans or autonomous bots, and they raise hard questions about code quality, maintainer responsibility, review sustainability, and whether rapid contribution velocity can replace deep human understanding of the underlying logic in complex codebases. The Godot engine team has decided the answer, for now, is no. They announced that they are rewriting their contribution policy to prohibit almost all use of AI from contributors after an overwhelming wave of pull requests that appeared to be machine-written. Maintainers say they can no longer keep pace with a growing backlog of AI-authored changes, and that reviewing them no longer feels like investing in people or the project’s future. This is not a rejection of tools; it is a rejection of unaccountable authorship.
Godot’s Ban: Quality Control and Demoralized Maintainers
Godot is rewriting its rules to bar most AI-generated code from its repositories, and the stated reasons are blunt: maintainers find AI code demoralizing, low quality, and hard to fix. They report that heavy AI users often cannot explain or repair the code they submit, leading to the stark judgment that “AI cannot take responsibility, and we can’t trust heavy users of AI to understand their code enough to fix it.” For an open source project that depends on volunteer time, this is not a minor complaint; it is maintainer burnout in plain sight. Reviewing pull requests has always been tedious, but it used to be “rewarding because reviewers generally feel that their efforts are contributing to educating a new contributor — who may become a future maintainer.” Once the author is effectively a model, that reward disappears, and the review process collapses into unpaid triage of AI slop.
Guardrails Over Firehoses: New Rules and Tools
Rather than accepting the AI firehose, maintainers are building code review guardrails. Godot’s updated policy requires explicit sign-off before any new contributor with three or fewer merged pull requests can submit new features or major refactors, a deliberate filter against vibe coders and coding agents who flood the queue without understanding the engine. Contribution discussions are required to stay human-to-human, with AI outputs banned in review threads except for translation. Other projects have tightened their contribution pipelines for similar reasons, citing review burden and false bug reports as quality control risks. On the tooling side, dupehound offers deterministic guardrails against AI code duplication, scanning about 1.5 million lines per second to detect cloned blocks and failing continuous integration when a change duplicates existing code, pointing contributors back to the original function. This is a clear pattern: if AI is involved, it must be boxed in by rules, checks, and disclosure.
Duplication Slop and the Case for Deterministic Checks
AI coding agents do not only create buggy code; they create vast amounts of duplicated slop. An analysis of 211 million changed lines found that cloned code blocks have roughly quadrupled since 2022, a direct symptom of models that cannot see an entire repository and therefore rewrite functions that already exist. Predictably, asking the same models to spot duplicates fails for the same reason: limited context. Dupehound responds with a deterministic pipeline: parsing files into functions with tree-sitter, normalizing identifiers and literals, fingerprinting structure with winnowing, and matching candidate pairs without an all-pairs scan. In practical terms, it offers two commands: Scan, which reports duplicate clusters and a repository-level slop score, and Check, which fails CI when a new change copies existing code and names the function to reuse instead. Projects can even instruct agents in configuration files to run "dupehound check ." before committing, baking quality control into the AI loop.
Mentorship, Not Velocity, Is the Real Metric
The deeper message in these policies is that open source values mentorship and knowledge transfer more than raw contribution velocity from AI agents. Godot’s foundation argues that reviewing code has always been partly about training future maintainers, and that once the counterpart is a machine, review work no longer builds human capacity. Zig has already adopted a zero-tolerance stance on AI-assisted contributions, with its community leaders describing code review as an investment in the person behind the pull request rather than the code alone. This matches the idea that in “contributor poker, you bet on the contributor, not on the contents of their first PR.” When AI dominates submissions, that bet cannot be placed; there is no mentee to grow into a maintainer. So projects are drawing a line: AI may help with small tasks like completion, regex, or find-and-replace if disclosed, but substantial changes must be written and defended by humans. The future of open source will not be decided by how fast code can be generated, but by who is willing to stand behind it.
