1. Knocking Firefox for AI functionality, but not Chrome? Chrome downloads an entire 4 GB local model without even telling you yet still sends queries to Google's hosted models.

Chromium as underlying engine doesn't have that; Chrome as browser built on top of it isn't FOSS and thus not on a list of Open Slopware

2. Recommending ls as a replacement for eza is a stupid and redundant suggestion. Nobody learns to download eza first; eza is installed as a replacement for ls because it provides more features than coreutils ls.

This is probably for those who thought to use eza as replacement for ls, suggesting they should go back to ls. You'd lose out on features, but your tool would no longer be AI sponsored.

The cherry on top is that if you use a distro with uutils, your ls is also on this list.

Arguably, permissive policy and AI-assisted code review are less egregious than outright AI sponsorship, but also, yeah, might wanna check which one your distro uses.

3. Good luck to any purist with replacing curl with wget. libcurl is everywhere and in everything.

You could control which one you use, but also, this raises the question of where you draw the line. Is there some degree of utility or criticality at which you give it a pass? Or only for certain types of AI involvement?

It's obviously hard to track just how deep the AI involvement actually goes, which makes nuanced treatment of individual projects tough.

4. "Linux Kernel"

This is my example for the case that "permissive policy" is a tag where nuanced treatment according to the best understanding and conscience of the user is warranted. The policy for the Linux Kernel for instance is both explicit that a human has to sign off and be responsible for the code, and that the AI tools used are to be tagged.

Suppose the alternative: All AI contribution is banned. Some well-meaning contributors may think this is overblown, particularly if they don't know enough to understand the difference in quality and dismiss the policy as luddism. They may then try to sneak in AI code anyway, but they don't make it transparent. Maintainers now have to be vigilant about every PR and change and meticulously investigate it for any smell of AI.

At a project of that size I don't think it's feasible. To allow them under the conditions above makes it clear that you don't consider AI agents independent "full" developers, while also allowing well-meaning contributors a way to submit AI-assisted code "legally". You're still not gonna catch the dishonest ones (or at least not the smart dishonest ones) but the honest contributors aren't walled off entirely.

Of course, you're welcome to disagree. My intention here is not to make an argument about Linux, but to demonstrate what nuanced treatment can look like... and that it is entirely infeasible at the scale of that list. Worse, for people without deeper knowledge of how this whole code and signing and shit works, it's downright impossible to give it such consideration.

That brings me back to the question of where to draw the line. I personally have never used eza, so it's trivial to say "eh, duck 'em, not gonna use that". For the Linux Kernel, I find a nuanced evaluation is warranted and consider the permissive policy acceptable.

Would I react differently to that same policy on some other project? Would I give Linux a pass if it used GenAI 'Art'? Do I treat AI functionality with similar nuance?

5. Both LLVM and gdb. I guess we're going back to the days of printf debugging, boys.

The evidence for gdb is a single accepted contribution that "Claude Opus 4.5 and GLM 4.7 assisted with". Doesn't seem like an explicit policy to me, though it also means they do not categorically reject anything with AI assistance. For similar reasons as the Linux Kernel question, I think "allow, but be transparent" is the better policy. LLVM also seems to share that policy.

Once more: you're free to draw the line elsewhere (both you, pivot_root, and any other reader). My case here is that we should consider that question in the first place. I don't see a complete hardline stance as defensible (considering Codeberg offers Weblate, which also appears in that list for its permissive policy, you'd be perusing a list hosted on an AI-permissive service in the first place). Neither do I endorse bending over and spreading wide for the AI overlords.

Few things in life are absolute and without nuance (including this rule). We need to be aware of and consciously walk that space between the extremes to find the best path. The right answer won't be perfectly in the middle, so we will need to work to find it, both for ourselves and for our interaction with one another.