Zig doesn't want your AI-generated code. The programming language project bans LLM assistance across the board: no AI for issues, pull requests, or bug tracker comments. Even translations are excluded. It's one of the strictest anti-LLM policies in open source, and Loris Cro, VP of Community at the Zig Software Foundation, has laid out the reasoning in detail.
Zig practices what Cro calls "contributor poker." When you review pull requests, you're betting on the person, not the code. Maintainers invest time in reviews to grow new contributors into trusted, prolific community members. That effort pays off in long-term relationships. An AI-generated PR, even a flawless one, skips that entirely. As Cro puts it, you play the person, not the cards. Simon Willison, who wrote about the policy, notes a related question circulating among maintainers: if a PR was mostly written by an LLM, why should a human reviewer spend time on it instead of solving the problem themselves with their own AI tools?
The policy has real consequences. Bun, a JavaScript runtime built in Zig, was acquired by Anthropic in December 2025 and relies heavily on AI assistance. Bun runs its own fork of Zig because it can't upstream AI-authored changes. That includes a recent 4x compile-time performance gain from parallel semantic analysis and multiple codegen units. Bun's team explicitly said they have no plans to upstream that work.
This is the tension coming for every major open-source project. AI-assisted forks can ship fast and land real improvements, but they can't give back. The upstream project loses improvements and potential contributors. The fork bears the full cost of maintaining a diverging codebase. Zig is betting that long-term community health matters more than short-term velocity. Other projects, including the AI coding tool DeerFlow, are watching closely and considering similar policies of accepting AI contributions only from already-trusted developers.