The creator of the Zig programming language has formalized a ban on AI-assisted contributions, calling them "invariably garbage" , and his argument goes well beyond code quality into the economics of open-source review labor and the philosophy of mentorship.
Andrew Kelley does not mince words. In a conversation on the JetBrains podcast that has since ricocheted across developer communities, the creator and president of the Zig Software Foundation said AI-assisted contributions have "negative value" , not zero, negative. His reasoning: they consume scarce reviewer time without adding anything that couldn't be done better by a human who actually understood what they were submitting. The ban, which is now embedded in Zig's code of conduct, covers not just code but LLM-generated comments, issue tracker translations, and anything brainstormed or debugged with an AI tool. It is one of the most comprehensive and explicitly justified rejections of AI tooling by any major language maintainer on record.
The Register published a wide-ranging interview with Kelley on May 28, in which he reiterated the position while discussing Zig's broader trajectory. The project is currently at version 0.16, still pre-1.0, and Kelley has made clear that reaching a stable release is a matter of achieving what he calls "uncompromising perfection" rather than hitting an arbitrary calendar date. In that context, the AI ban is not an isolated policy quirk , it fits a project culture that is fundamentally hostile to shortcuts.
What makes Kelley's case technically credible is that he frames the problem as one of resource allocation, not ideology. The Zig project has roughly 200 open pull requests at any given time, and a small number of core team members responsible for reviewing them. Every AI-generated submission that enters that queue displaces a human contribution that could have been mentored into something valuable. Kelley's phrase for the dynamic , "contributor poker" , captures the underlying frustration: AI tooling has removed the effort-based friction that once acted as a natural quality filter on pull request volume. When generating a passable-looking patch costs nothing, the signal-to-noise ratio in maintainer inboxes collapses.
There is also a mentorship argument that is easy to overlook. Zig's stated mission includes developing contributors over time, turning occasional submitters into reliable team members. AI-assisted contributors, Kelley argues, are effectively unteachable within that model , they are not learning the codebase, not absorbing feedback, and not accumulating the judgment that makes a long-term contributor worth investing in. The person submitting an AI patch is not helped by receiving review notes on code they did not write.
Zig is not alone, but the company is small
A survey of 112 major open-source projects conducted in March 2026 found that only four had implemented outright bans on AI contributions: Zig, NetBSD, GIMP, and QEMU , just 3.5% of the sample. QEMU's policy, committed directly to its repository, declines contributions where AI use is known or suspected, citing GitHub Copilot, ChatGPT, and similar tools by name. NetBSD requires prior written approval from its Core Team for any AI-generated code. The reasoning across all four is consistent: legal ambiguity around AI-generated copyright, the Developer Certificate of Origin problem (which requires contributors to assert they created the code themselves), and the review economics that Kelley has articulated most publicly. A broader informal index on Codeberg , the "slopfree software" list , tracks a growing number of smaller projects taking similar stances, suggesting the movement is widening even if the named projects remain a small fraction of the ecosystem.
Separately, the Zig project has migrated its primary repository from GitHub to Codeberg, a German nonprofit. Kelley cited frustrations with GitHub Actions and discomfort with GitHub's increasing AI integration as motivating factors. The move is consistent with a project that takes its infrastructure politics seriously, and Codeberg's European nonprofit structure aligns with Zig's preference for institutional independence. The foundation itself reported $670,000 in funding, enough to sustain core development without chasing venture capital or platform partnerships.
What this means for AI developer tools
The commercial case for GitHub Copilot, Cursor, Devin, and their competitors rests heavily on developer adoption , on the claim that AI coding tools are becoming the default interface for professional software work. That story has been under pressure in 2026. GitHub paused new Copilot individual sign-ups amid surging compute costs from agentic workflows. Microsoft was forced to revert a VS Code update that automatically added "Co-authored-by: Copilot" to commits without user consent. GitHub's own policy change to train on user interaction data by default sparked significant backlash. Against that backdrop, Kelley's formal policy lands in a context where developer trust is already running thin.
The deeper challenge for AI coding tool vendors is that Zig is exactly the kind of high-stakes codebase where their products' weaknesses are most exposed. Systems-level programming , the domain Zig occupies alongside C and Rust , requires correctness tolerances that leave no room for plausible-sounding but subtly wrong suggestions. When Kelley says AI contributions are "invariably garbage," he is not speaking from a general philosophical objection to automation. He is speaking from the perspective of someone who reviews code where a misunderstanding of memory ownership is not a code smell but a security vulnerability. That grounding is what gives his position weight that a blanket cultural complaint would not have.
Whether other major open-source projects follow Zig's lead will depend less on ideology than on practical reviewer fatigue. If the volume of AI-assisted pull requests continues to grow alongside the proliferation of agentic coding tools, maintainers who have not yet formalized a policy will increasingly find themselves managing the same bottleneck Kelley describes. The four projects that have banned AI contributions so far represent a small minority , but they are, notably, among the most technically rigorous projects in the ecosystem. That is probably not a coincidence.
Also read: The Biggest Tell That Something Was Written by AI | Look closely and you'll see that every part of the text is not quite • MiniMax chose Hong Kong over Wall Street and its shareholders are glad it did • Samsung's last-minute bonus deal signals that AI's productivity windfall is becoming a labor negotiation rather than a corporate perk
