Hi Kaxil,

I’m so sorry for the delayed response! For some reason, your message landed
in my spam folder, and I only caught it after seeing Amogh’s reply.

I wanted to say that I 100% agree with you. I think many recent
disagreements and discussions stem from the frustration that our "old ways"
no longer work in the current environment and we have not yet found a good
way to deal with it. This causes a lot of stress and strain as we
desperately try to find solutions. And our approaches to finding those
solutions often differ wildly. But I personally think if we work "together"
on it and leverage the same AI and automation, we can solve it without
forgetting that there is always a human on the other end.

Prioritizing the "why" and "why not" is such a vital part of the meaningful
conversations we should be having as a community. My goal has always been
to ensure that experienced maintainers like you, Ash, and Amogh can focus
your energy solely on those high-level insights. This is one reason we
automate routine feedback—like fixing CI, adding tests, or formatting
subjects—so you don't spend your valuable time on them.

And yes - I agree and love Ash's review when he asks "why". In many cases
it changed, improved, or redirected the proposals—including some security
issue implementations—where Ash's knowledge of the code is invaluable,
especially when the original solution didn't actually fully fix the problem
or solved a different one. And I love those. And act on those - I get to
dig deeper and understand—no matter if I use agents to respond or not and
put **my own** thinking into words.

And this has not changed - this has been the case for years - whether with
or without AI, there were often human mistakes, bad assumptions, or
misunderstandings of how things work. People sometimes call these
hallucinations now, even when humans make them - not AI—but I think it's a
really bad name. Those are what they are: mistakes, bad assumptions,
misunderstandings how things work

By leaning into automation and the triaging efforts I’ve been working on
with Magpie, I believe we can reach a point where maintainers only step in
when it’s time for those critical architectural questions. I completely
agree that holding a high bar is actually very welcoming; it helps
contributors grow. What’s truly tough for newcomers is being left in the
dark without any feedback at all.

And my focus for the next weeks will be moving from "what we want" to "how
we achieve it.". While I’ve been building some of the initial scaffolding
for this triage process, and yesterday, Magpie was finally established as a
Top-Level Project!) Magpie has finally been established as a Top-Level
Project - we could expand it to the next step: ensuring we address the
"why?questions are actually asked—currently, we have a big number of PRs
that don't even have those questions asked.

I think our efforts will only succeed if we all contribute to improving how
we use it in Airflow—turning it into an automated CI process wherever
possible, and getting maintainers focusing on the "right" reviews. However,
this must be a very collaborative process.

Simply - whenever you see a check that feels repetitive or automatable, I’d
love for us to collaborate on turning that into a permanent part of our
automated criteria. Either by submitting PRs yourself, asking in
#auto-triage-feedback or implementing new ways to handle it. Also getting
more people to understand and contribute to it - as happened with our CI. And
make sure whatever is "ready for review" - is actually ready.

But most importantly, once a PR is "ready for review," we must ensure a
human maintainer actually reviews it—one who won't need to ask about code
quality; their first question should be, "Why?" Currently many of those PRs
are not even looked at by anyone. So what kind of questions we ask doesn't
matter, because we often don't even ask any questions.

I’m more than happy to lead this process, and I’d love to have everyone’s
active involvement. If we can share the load, we can make sure every
high-quality PR gets the thoughtful "why" questions it deserves in a timely
manner.

Best,
Jarek

On Mon, Jun 15, 2026 at 4:34 PM Kaxil Naik <[email protected]> wrote:

> Hey all,
>
> While we are talking about contribution quality and ownership in various
> separate threads, I do want to say that demanding reviews are one of the
> best things this project has, and I'd hate for us to come out of these
> various conversations having quietly lowered the bar to feel more
> welcoming.
>
> The reviews I have learned the most from weren't the ones that approved my
> code. They were the ones that asked "why this way?", "what does this do to
> the person debugging it in two years?", "is this even the problem we should
> be solving?" Ash's reviews are the example I keep coming back to. We
> disagree plenty, and he's caught things in my own PRs I'd have merged
> otherwise. Those questions take real time and care, there is no metric or
> leaderboard that rewards asking them, and they are exactly the judgment
> that doesn't easily automate.
>
> After his review comments, I often asked myself why I hadn't thought of
> that approach.
>
> Bolke did the same to me in my early days (
> https://github.com/apache/airflow/pull/2810). The review stung at the time
> and made me a better engineer. Most long-time contributors here probably
> have a story like it.
>
> Holding a high bar isn't unwelcoming. It's how contributors actually grow,
> and how the codebase stays something we can all still reason about. Waving
> work through to dodge friction helps no one, least of all the contributor.
>
> To be clear, this is a stance on the bar, not on any one way of delivering
> feedback. I just don't want the two collapsed into each other, because the
> answer to a tone problem is never less rigor.
>
> Thanks to Ash, and to everyone here who puts this kind of care into
> reviews. It's a lot of unsung work, and a big part of why I still trust
> this codebase.
>
> Regards,
> Kaxil
>

Reply via email to