Mesa Loosens Rules Around AI Agents For Code Comments / Commit Messages

The Mesa 3D graphics driver project has loosened its rules laid out in their AGENTS.md for AI agents to permit AI/LLM assistance in generating code comments and commit/patch messages, as long as a human is ultimately reviewing/editing it in the end. Alyssa Rosenzweig laid out the change in finding the existing restrictions to be "unnecessarily limiting" for workflows that may involve AI-generated commit messages or comments for an initial draft and then reviewed/edited by a human before submission. Rosenzweig explained in the merge request now part of Mesa Git: "It turns out this text is unnecessarily limiting. We have policy that forbids AI-generated commit messages, for example, and this file was intended to enforce that existing policy. However, this causes problems for workflows where someone uses an agent to generate an initial draft, and then manually edits and revises. As long as the final changes submitted to us are human, we don't care about those intermediate steps. It would be an overstep for us to attempt to enforce policy about what people do "privately", and this was never the intention in the original AGENTS file. Unfortunately, it turns out that has been the case. I would like to apologize for the inconvenience this has caused. The original draft of the AGENTS file was reviewed by several AI users, but all software has bugs in its first version :-) This commit is explicitly not changing the policy for the project. The project policy was reached via a painful amount of discussion, landing on the current compromise among people on both sides of the AI issue. Personally I am just exhausted. I don't think anybody loves the current policy. It's a compromise position that (by design) satisfies nobody 100%, but I do believe a compromise is necessary given the number of strong voices on both sides of the fence. I consider the AGENTS file like software, and my intention here is to fix a bug in software I wrote. Sometimes we enable -Werror -Wextra to catch issues, and sometimes we learn later that people leave unused variables in tree when they prototype changes and the annoyance caused needing to comment out variable declarations isn't worth the payoff. So we make unused-variable a warning or silence it, even though "no unused variables" remains Mesa code style. Similarly, many parts of the tree enforce clang-format as their code style, but we do not run clang-format lints in CI because "annoying people with CI fails from clang-format problems" turned out to be worse than "annoying people with needing to re-run clang-format periodically when non-conforming code is merged". Speaking from experience - I both added asahi clang-format to CI and then reverted it later. As a community of humans I hope people can understand the original draft of AGENTS.md was a good faith attempt to capture the state of the community's compromise and improve MR signal:noise, rather than a hamfisted attempt to smuggle in stronger rules affecting Mesa developers via a backchannel. As for the policy itself: I am very much uninterested in reopening the can of worms. It was emotionally exhausting the first time it was litigated, we spent a ton of time collecting thoughts and we landed on... something. I won't say it's perfect but I'm tired and would rather write code than relitigate the same issues on new MR and new MR. My hope is that we can merge this patch to the AGENTS file to fix the regression in people's workflows, then defer conversations about the policy at least for a few months. Maybe cooler headers will prevail in spring." So the AGENTS/md update goes ahead and removes the language around AI agents contributing to GitLab directly and not generating code comments, commit messages, or GitLab comments. This is what's merged to Mesa Git.