On Sat, Nov 14, 2026 at 11:32:31PM +0000, Tom Smyth wrote:
> Folks,
> to be fair there was a request for a review into the mailing list, the
> poster was transparent it was AI,
> Fair,
> I dont think the OP intended to demand anyone specifically  to review the 
> code.
> 
> but we all do have to be mindful of human developers time, and as a
> dear friend called "some
> LLM outputs a wall of text" it is mentally draining...  that is Fair.
> 
> my personal view is that any AI that is posted had better been tested
> by a person and reviewed as thoroughly
> as the ability of the person proposing it can give...
> I have used AI tools (with very mixed results (albeit improving) but
> before posting anything review it
> test it...
> there are a lot of volunteers on this project and their time is valuable...
> there are only so many skilled developers who can  fully understand
> the code and its implications to the project
> and its community...it is important not to overwhelm them (even if
> well intentioned) ...  (mae culpa) sorry

I will reply to both of your posts here despite only quoting the latter
(I am lazy).

OpenBSD's selling point is being a fully functional operating system,
while trying to remain as simple and auditable as possible. For the past
few decades, it has been largely successful at this, and many OpenBSD
projects are used on other platforms, such as openssh and opensmtpd.

Take the recent development of got, I recently picked this up after
seeing the talk at FOSDEM and I was amazed, everything I need in a VCS
is there, I read the well written man page and almost all my questions
are answered. Apart from some minor hiccups on unexpected behaviour
which was quickly solved on IRC, it has been a flawless experience, dare
I say, a better experience than using git. Oh how my experience would
have differed if it was all written by claude?

That is what OpenBSD offers. Now you slap AI into the problem, it
generates highly verbose code, which is hard to audit and maintain. It
has no concept of licencing, it will violate copyright and Hansteen has
pointed out, such a lawsuit could be devastating for an open source
project, and that's before we even touch on the inherit security issues
AI brings, yet alone the fact that AI products usually fall short on
expectation.

There is no doubting there has been substantial success with AI in some
industries, it is far better than search engines at replying with the
right information, and it is surprisingly good at finding common
programming mistakes which lead to RCEs, hence the vulnapocalypse.
However, the fatal flaw of AI is its too confident, its like an arrogant
15 year old who believes itself to be flawless, and like a arrogant 15
year old, it in reality does everything wrong and knows nothing, in many
cases AI has been known to make stuff up if it can't properly find the
answer to your question, and you are suggesting we trust code written by
such model to handle our emails? It would be like trusting a kid with a
litre of petrol and a box of matches while you go to the supermarket.

By adopting AI into your workflow, you are trading short term laziness
and productivity with long term security and maintainability. You are
sacrificing the future of your project to make your immediately workload
easier.

A large percentage of open source projects have fallen to AI, Linux
notably has adopted it fully, mainly because it is developed by people
who run AI firms or companies big on AI.

NetBSD has declared a hard ban on AI, OpenBSD to my knowledge hasn't
made any official ruling on it yet, ditto FreeBSD, but in the case of
FreeBSD as I pointed out in my previous post, it seems very likely they
will rule in favour of AI.

I will be blunt, OpenBSD has already merged one or two patches which
were AI generated, small diffs, nothing to lose sleep over. The
implication of a new daemon within OpenBSD (if the team chooses to merge
it) would be the death of everything OpenBSD stands for, and that is no
exaggeration.

I am sure many people here know the common workflow when updating
OpenBSD servers, wait a few weeks to see if others devices break when
running the latest release before committing to it. The same approach
should be taken with AI. OpenBSD is not missing out on anything by not
adopting AI, it has existed for decades, it will continue to exist with
or without AI.

However, if the AI hype ends out being as catastrophic as so many of us
fear, OpenBSD adoption of AI would be the end of it, whereas it would still
survive if it didn't.

Many companies are already realising the AI hype is exactly that, hype.
Investors are beginning to bail out, knowing they have lost their money,
Nvidia's stocks before the recent scandal were already shaky, OpenAI has
been struggling with funding, big companies have been making
redundancies to plough more money into the AI black hole. Government has
now stepped in to throw money at AI, mainly in the interest of mass
surveillance which is good for none of us. 

The fact that so much of the open source community is pro-AI is
astounding to me, and I fear, it will be the end of open source as we
know it.

Take care,
-- 
Polarian
Website: https://polarian.dev
Jabber/XMPP: [email protected]

Reply via email to