Hi Yufei Fully agree.
That said, I think most of this is already written down in the CONTRIBUTING.md. It has the "Guidelines for AI-assisted Contributions" section. It clearly states that the person opening the PR is the author, must understand the change end-to-end, and must be able to explain and justify the design and the code. AGENTS.md also goes further for coding agents. I don't think we need new rules and being strict (it can be seen as pedantic :)), instead we should point contributors to CONTRIBUTING.md and AGENTS.md, reviewers should feel free to link them when a PR clearly needs to follow them. I fully agree on small PRs. I think it can make sense to have a large draft PR to illustrate a concept and split in smaller PRs in a follow-up. It's what I did for Polaris Directories,. To Yong, for reference, CONTRIBUTING.md says "two working days is a good base to get first reviews and comments". It's a guideline to give reviewers time, not a commitment to merge within that window. It's extremely hard (and pretty risky) to announce any commitment around that. I strongly against because it's not the Apache philosophy: we are all contributing as individual, volunteer, best effort. For some of us, it's also our job, but it doesn't change the philosophy. As I said in a previous thread, a fast first response is important to keep contributors motivated, but time to merge is less important than quality and review. AI review tools are clearly a good first pass, for authors and reviewers, but they don't replace committer review. Ideally, contributors address (or answer) the bot comments before asking for a human review. Partial reviews are also very welcome. If you only review part of a PR, it's completely fine (it should be mentioned to inform the other reviewers). Regards JB On Fri, Oct 30, 2026 at 6:09 PM Yufei Gu <[email protected]> wrote: > Hey folks, > > It’s great to see more contributions coming into Polaris! At the same time, > the review load is pretty high, so it’s normal for some PRs to take longer > to get reviewed. > > I’d encourage more people to help with reviews. It is helpful even you > don’t review the whole PR and look at the parts you’re familiar with and > sharing feedback. > > For authors using AI, I think we should be the first reviewers of the code > and docs it generates. At least read and understand what you’re submitting, > and challenge its decisions. Do we really need this extra doc or a new > interface? Can we solve the problem within the existing design? A few > rounds of AI review, combined with your own judgment, can help catch issues > that a single pass might miss. Few-shot is always better than one-shot, > esp. with human’s insights. Taking responsibility for this initial review > helps reduce the back and forth with reviewers. > > As I mentioned in my earlier email [1], please keep PRs small and well > scoped. They’re easier to review thoroughly and usually have a faster > turnaround. Even if AI makes it easy to produce a large change quickly, > splitting it into focused PRs still helps everyone. > > Just my two cents. Curious what you all think. > > Thanks, > Yufei > > [1] PR review and committership, September 2 ( > https://lists.apache.org/thread/jdsc283vwbrff822kw2hh5nlb6qbw6x6) >
