Hi Vladimir, Thanks for bringing this up. I personally would prefer to not see any mentions of AI tools in the commit message, under any forms, but I believe it's very important to have it in the PR description.
Why no AI mentions in commit messages: when the commit lands in main, I assume it has all it needs (quality, clarity, respecting repo and community conventions etc.), and it's less important how we got there, including if AI was used, similarly to what IDE, linter, testing framework etc. were used. So adding it there it's just unnecessary information and noise IMO, and commit messages should have the highest SNR as possible. Why I think it's important to add it to the PR description: I look at AI-assisted PRs a little differently so I'd rather know it from the beginning and don't change strategy after a while. It's always been the case that some parts of a PR would be more "mechanical" than others and I appreciate PR authors to be frank about that so I can spend more time on those parts to fill the gap. With AI tools the chance for these mechanical parts increased, it would be great to be upfront on where attention is more needed. Possibly mentioning what model was used here could make sense if one wants to run an automatic review round with another model. Looking forward to here what the community thinks of this interesting topic. Best regards, Alessandro On Mon, Aug 31, 2026, 07:45 Vladimir Sitnikov <[email protected]> wrote: > Hi, > > I filed CALCITE-7752 and PR #5230 proposing a commit-message rule for > AI-assisted work. Julian pointed out that this is a policy decision for the > list rather than something to settle in a JIRA, and he is right, so I am > bringing it here. I will use his framing of the decisions: > > 1. Whether there should be a line in commit messages recording AI tool > use at all. > 2. If so, what that line should be. > 3. Whether to write it into AGENTS.md. > > Where Calcite is today > > Six commits on main since June 2026 name a tool in a trailer, in four > different spellings: > Co-Authored-By: Claude Opus 4.8 <[email protected]> (3 commits) > Co-Authored-By: Claude Sonnet 5 <[email protected]> (1) > Co-authored-by: Cursor <[email protected]> (1) > Generated-by: Claude Opus 5 (1M context) (1) > > Nobody chose that. Claude Code adds Co-Authored-By: unless > attribution.commit > says otherwise, and VS Code 1.118 shipped Copilot as a commit co-author by > default before reversing it. > > What the ASF says, and does not say > > The generative tooling guidance [1] recommends naming the tool in the > commit > message and offers "Generated-by:" as an example. It is a recommendation, > it > mandates no particular token, and it predates agentic tooling. > > There is an open thread on legal-discuss, "[DISCUSS] Making the > Generated-by > record mandatory for AI-assisted contributions" [2], started on 11 August > and > not concluded. The positions in it run from Greg Stein's "use whatever > annotations you wish" and Mark Thomas leaning towards no labelling at all, > to > Justin Mclean, Calvin Kirs, Alin Jerpelea, Andrew Purtell and Mark Struberg > preferring "Assisted-by:" over "Generated-by:". Justin also measured the > foundation: about 4,500 commits across 199 repositories carry an AI tag, > and > the "Generated-by" value has 481 distinct spellings. > > Elsewhere > > Projects with a written policy have converged on "Assisted-by:": the Linux > kernel [3], Zephyr, Fedora, Rocky Linux, OpenInfra and Mesa. Mesa states it > plainly: "Do not use the Co-authored-by tag as this one is reserved for > human > co-authors." > > The opposing view is real and worth weighing. Kubernetes [4] forbids AI > trailers of every kind, "Assisted-by:" included, and asks for a sentence in > the pull request description instead, on the grounds that an AI cannot > sign a > CLA. Inside the ASF, HBase and Phoenix currently recommend > "Co-Authored-By:". > > What I proposed, for reference > > One sentence in AGENTS.md: > Always end commit messages, including drafts, with > `Assisted-by: <tool> (<model-id>)`, never `Co-Authored-By:`. Keep > existing > trailers and add yours when amending someone else's commit. > > The PR explains why that wording rather than another. Since AGENTS.md is > read > by agents, I tested candidate wordings against three of them rather than > picking one that reads well. > > Disclosure: I use coding agents, so I am asking for a rule I will have to > follow myself. > > I do not want to presume the answer to question 1, but I do not think it > has a > do-nothing branch. Coding agents add Co-Authored-By: on their own, and we > cannot rely on every contributor switching that off, so "Calcite records > nothing" is also a rule, and it also has to be written where agents read > it. If > that is the consensus, I will close CALCITE-7752 and PR #5230 and open an > issue > for the opposite sentence in AGENTS.md: that an agent must not add a > trailer > naming itself. Either answer suits me. > > Vladimir >
