Hi Peter,

Thanks for the thoughtful reply. I want to address a couple of things.

On the point about me probably being unable to do this unassisted: I think
that's not quite right, and I suspect it comes from the same assumption I
addressed in my reply to Kooda. The picture you might have is that I
pointed an AI at the codebase and it produced the patch. The reality is
closer to me using the AI as a research and iteration tool while driving
the decisions myself.

I can explain every line in the proposal: why the FFI layer is structured
the way it is, why I chose to wrap SSL_set_fd instead of managing BIOs
directly (simplicity, mostly and it was quicker to get to a POC), etc. None
of this is opaque to me. Would it have taken me longer without AI
assistance? Absolutely. Would I have been unable to do it? No. I've done
FFI work before, I've integrated OpenSSL in other contexts, and the CHICKEN
side is a matter of learning the specific patterns (which is exactly what I
was doing).

That said, I hear you on the review burden. You're right that this is a
small community and the dynamics are different from a large project. A
medium-sized patch from a newcomer is already a lot to review, and the AI
question adds another layer of uncertainty on top of that. I should have
started smaller, and that's a fair criticism regardless of how the code was
produced.

Cheers,
Rolando

On Tue, Feb 10, 2026 at 12:45 AM Peter Bex <[email protected]> wrote:

> On Mon, Feb 09, 2026 at 11:39:39AM -0800, Rolando Abarca via
> Chicken-hackers wrote:
> > Hi Felix (and chicken-hackers),
> >
> > Thanks for bringing this to the list. I think it's an important
> > conversation to have.
>
> Hi Rolando,
>
> Thank you for engaging in a thoughtful and considerate way.  This AI
> stuff is very divisive, and it's great we can have a respectful
> conversation about it.
>
> > First, I want to be clear: I'm totally fine if this doesn't get merged.
> The
> > feedback you've already shared has been incredibly valuable, and that
> alone
> > made the exercise worthwhile. Getting insight into how CHICKEN's build
> > system works, the static vs dynamic considerations, and the cond-expand
> > approach... that's exactly the kind of learning I was hoping for.
>
> It's excellent that you're intending to learn from this, and I hope that
> you'll consider becoming a full-time contributor.  But I have to note
> that this way of learning is kinda new, and I'm not sure we have the
> resources for it at the moment.  Our community is quite small and the
> active core contributors are even fewer.
>
> Traditionally, one would start by submitting small patches to the core,
> which might get approved or rejected, but at least you'd get feedback
> and this kind of feedback would be somewhat light on the core
> contributors because we're talking about small amounts of code.
> It is also clear that the submitter is still learning, there will be
> somewhat basic mistakes indicating the level of familiarity with the
> codebase.
>
> The patch you submitted is medium-sized and somewhat of an ambitious
> change, which (no offense) you probably would have been unable to do
> unassisted.  This kind of dynamic fundamentally changes the equation
> for outside contributions.
>
> I like the change, don't get me wrong, but reviewing such a big pile
> of code is mentally hard work.  Add to that the fact that you can't
> trust a single line of code because LLMs tend to intersperse decent
> code with weird artifacts.  This means it requires an extremely thorough
> vetting, which will be very exhausting to whomever takes on the job of
> reviewing.
>
> > On the broader topic of LLM-assisted contributions: I work in BigTech,
> and
> > I can tell you things are moving faster than most people outside these
> > environments would expect.  AI-assisted development is becoming the norm,
> > not the exception. I think the CHICKEN community (and really, any open
> > source project) will need to figure out how to handle this sooner rather
> > than later.
>
> It's a big difference to working in a big company, or even a big open
> source
> project.  We're nothing like those.
>
> > One idea: what if there was a way for agents to self-check contributions
> > before submission? Something like a CONTRIBUTING_AI.md or similar
> document
> > that outlines the specific code quality concerns, style expectations, and
> > common pitfalls. An agent could review the PR against those criteria
> before
> > a human even sees it. This wouldn't replace human review, but it could
> > raise the quality bar and focus reviewer attention on what matters most.
>
> I have no idea how that would even work.  Perhaps you could make a patch
> submission so we can see what that would look like?  But perhaps first
> we want to establish some ground rules about AI usage first.  If we want
> to blanket ban any AI-assisted contributions, there's no point to
> including such a file.  If anything, the presence of such a file would
> encourage *more* AI-assisted contributions, increasing the load on our
> small team even more.
>
> > That said, if the community isn't ready to tackle this now, I completely
> > understand. This work isn't blocking anything critical. It started as a
> > personal exploration to learn CHICKEN's internals, and it served that
> > purpose well. But if there's interest, it could also be an opportunity to
> > start thinking through how the project handles this new wave of
> > contributions.
>
> I think that's a good idea, it seems our industry is headed that way,
> whether we like it or not.
>
> > Either way, I appreciate the thoughtful engagement.
>
> Likewise!
>
> Cheers,
> Peter
>

Reply via email to