zig A field note by Vikrant Sharma
MicroZig bans LLM-generated code contributions
An embedded Zig project updated its code of conduct to explicitly forbid AI-generated pull requests. The reasoning is sharper than I expected.
The MicroZig project updated its code of conduct yesterday to ban LLM-generated code and documentation. No GitHub Copilot suggestions. No Claude-written docstrings. Human contributors only. The reasoning is simple. Embedded systems work requires understanding memory layout, timing constraints, register access patterns. An LLM does not understand why you mask bit 3 before writing to UART_CR. It pattern-matches from training data where half the embedded C code is quietly broken in ways that only surface when the watchdog timer fires on hardware you do not own. The policy draws a line between using an LLM as a search engine and using it to generate contributions. Asking Claude to explain an ARM Cortex-M memory barrier is fine. Pasting its generated code into a pull request is not. The project wants contributors who can debug the code they write, not people who ship plausible-looking nonsense and disappear. This is interesting because it is not about code quality gates or review process. It is about contributor legibility. If you cannot explain why the code works, you cannot fix it when it breaks. Embedded work has no margin for “I just tried what the AI suggested and it compiled”. The Zig embedded community is small enough that this might actually hold. Maintainers know the regular contributors. A pull request from someone who suddenly writes in a different register or submits perfect code with zero prior context is obvious. The policy gives maintainers permission to reject it without debate. I wonder how long until this spreads to other low-level systems projects. The kernel people already ignore drive-by patches. This makes the quiet part loud.
