Zig’s Anti‑LLM Policy and Bun’s 4x Compile Speedup
Review Zig’s anti‑LLM policy to understand constraints on LLM‑assisted contributions.
Check your own projects for LLM‑assisted contributions before submitting PRs.
Summary
Zig enforces a strict anti‑LLM policy: no LLMs for issues, pull requests, or bug‑tracker comments, and no LLM‑generated translations. The policy is justified by a “contributor poker” philosophy that values contributor growth over code quality. Bun, a JavaScript runtime written in Zig and acquired by Anthropic, added parallel semantic analysis and multiple LLVM codegen units, yielding a 4× performance boost on Bun compile. Bun’s fork of Zig implemented this change, but the project declined to upstream it because Zig’s policy forbids LLM‑authored contributions. A Zig core contributor explained that parallel semantic analysis is a long‑planned feature but has implications for the Zig language itself. The policy is designed to keep the review workload manageable while encouraging new contributors.
The policy’s core points are: - No LLMs for issues, PRs, or comments. - LLM‑generated translations are disallowed. - Contributions must be human‑written. - Bun’s performance improvement was not accepted for upstream. - The rationale is to grow contributors, not just code.
This policy affects any project that relies on Zig or its forks, including Bun, and sets a precedent for LLM‑free development.
The article also highlights the tension between AI‑assisted development and open‑source governance, illustrating how policy can shape technical progress.
Key changes
- No LLMs for issues, PRs, or bug‑tracker comments
- No LLM‑generated translations allowed
- Bun added parallel semantic analysis and multiple LLVM codegen units for 4× compile speedup
- Bun declined to upstream the change due to Zig’s policy
- Policy aims to grow contributors, not just code