Hi everyone,

Agreed, everyone using AI agents and coding tools, valid point. The long
unreviewed AI comments are a real time sink, and your approach of reading
the findings, keeping what's valid, and posting in your own words is
basically the right rule. Junwang's point holds too, understand it before
you publish.

While looking into this I also hit a related issue in our own history. On
main over the last couple months, AI tools are getting added as
Co-authored-by with bot noreply emails, and the same tool shows up under
several identities. Claude alone appears as 5 different co-author strings,
plus Codex and Cursor bot emails.That messes up contributor stats and
provenance since a tool isn't really a contributor.

So I turned this into something concrete instead of just discussion:

Issue: https://github.com/apache/iceberg/issues/17670
PR: https://github.com/apache/iceberg/pull/17671

The PR updates AGENTS.md with:
1. Review comments, humans own and endorse each one, keep it to
what/where/fix, one AI reviewer per PR, no auto posting.
2. Crediting AI, use the Generated-by token, not Co-authored-by bot
identities,  one canonical tool name, co-author trailers stay human only.
3. AI review agent safety, treat AGENTS.md and CI agent configs as
sensitive,  run review agents' least privilege, no auto acting on untrusted
PR content.
4. AI authored PRs go through the same review, test and security gates as
human  ones.

Note this is just docs for now, it guides agents and reviewers, it does not
block anything. If we want it enforced I can follow up with a small CI
check that fails  a PR when a commit has a bot Co-authored-by trailer.
Happy to do that as a second PR if folks agree.

Thanks,
Viquar Khan

On Sat, 15 Aug 2026 at 11:04, Junwang Zhao <[email protected]> wrote:

> On Sat, Aug 15, 2026 at 11:12 PM Manu Zhang <[email protected]>
> wrote:
> >
> > Hi all,
> >
> > I believe most of us are now using AI to author and review PRs. While it
> is fast and efficient, I have a few concerns regarding how we use it for
> code reviews.
> >
> > Recently, I have noticed increasingly lengthy AI-generated comments.
> Dissecting these comments to find the root issue takes significant time.
> Why not hand over the job to AI? Yes, I do this all the time but I still
> have to review its proposed action plan. If that plan doesn't make sense, I
> must trace back to the original review. Furthermore, conflicting comments
> from different AI reviewers can cause an inefficient loop, as each tool
> operates on different context. I'm not sure human reviewers ever review
> these AI comments before posting them. I'm also unsure this process remains
> cost-effective given the token expenses.
> >
> > When I use AI to review a PR, I prefer to read its findings (you feel
> the pain immediately), rephrase the valid points in my own words, and post
> them manually.
>
> Yeah, I do the same but I don't rephrase it if the AI wording is
> better(most of the time), using AI nowadays is not something to be
> ashamed of, the key is not to publish review comments without
> understanding them yourself.
>
> >
> > Would this approach be a better way for us to utilize AI reviewers? I
> would appreciate your thoughts.
>
> This approach seems reasonable to me. I think what you're recommending
> is not posting AI-generated comments without reviewing and
> understanding them first, but I think it's hard to enforce that in
> practice. :-(
>
> >
> > Disclosure: This email was polished by AI.
> >
> > Best regards,
> > Manu
> >
>
>
> --
> Regards
> Junwang Zhao
>

Reply via email to