I can not put into words how much this vision of yours fills me with revulsion.
The text in your report hurts to read. Don’t force your slop on to the rest of us. No, making your little machine run it through an unslopifier isn’t making it any better. Such a thing does not exist.
As the current accountant for the foundation, I can confirm, we paid a whooping $0.01 for all of our use so far!
They’re actually kindly sponsoring all of our use 
Though the $0.01 is real, we got billed for that, but it was seemingly due to a rounding error!
You do realize the irony in linking to a marketing article from a company that sells AI-powered code review tools, right? Either they’re not credible, or we should be using AI-powered code review tools.
(Like, in all seriousness, getting donated reviews from CodeRabbit might be a solve for your other objection about “who pays for this AI usage”.)
Also CodeRabbit and Gemini Code Assist kinda suck. Bring too much noise.
I think we should have a policy around straight up closing low effort contributions. If someone can’t keep up with a PR to maintain later then it shouldn’t even be accepted in. Would only bring burden to everyone. Nixpkgs has a lot of abandoned things too like a whole internal fork of factorio, or other similar game, IIRC.
It’s possible to do high effort stuff with AI, it’s just not the path of least resistancy.
A good AI setup is simpler than you guys think, there is a lot of noise out there.
The current guidelines ask reviewers to meet slop with patience, and only resort to closing a PR if there is no improvement. This is what I would label good-faith social behavior. What you instead propose is to switch to anti-social behavior as our default mode of operation to accommodate more chatbots.
II stand corrected re: the Github bill 
And yes, I saw the irony in linking to CodeRabbit; they provided the clearest explanation of the issue.
I’ve been curious about what is the amount of assistance that already landed into nixpkgs.
So i collected the commits of last year (2025-07-16 to today), filtering out merges, bots & nix-init. I can only count ai-contribution that were properly labelled as such.
| top human | commits | top ai-assisted | commits |
|---|---|---|---|
| Fabian Affolter | 6865 | Aliaksandr | 36 |
| Martin Weinelt | 4077 | Guy Chronister | 31 |
| Robert Schütz | 2675 | Markus Hauck | 26 |
| Gaetan Lepage | 2142 | eljamm | 22 |
| Stefan Frijters | 1313 | Sam Estep | 19 |
| K900 | 898 | Gerhard Schwanzer | 12 |
| Sigmanificient | 882 | Ilan Joselevich | 9 |
| Harinn | 853 | lucasew | 7 |
| Guy Chronister | 819 | Philip Taron | 7 |
| Doron Behar | 818 | Lisa Scheers | 7 |
| Sandro Jäckel | 801 | phibkro | 5 |
| Sarah Clark | 763 | Tyce Herrman | 4 |
| Wolfgang Walther | 742 | Ihar Hrachyshka | 4 |
| Bobby Rong | 683 | dfjay | 4 |
| quantenzitrone | 676 | Alexis Williams | 4 |
| Michael Daniels | 643 | Arian van Putten | 4 |
| jopejoe1 | 615 | Nova DasSarma | 4 |
| Austin Horstman | 574 | Otavio Salvador | 4 |
| pancaek | 553 | Maciej Krüger | 4 |
| sternenseemann | 528 | bitbloxhub | 4 |
| …humans | 50549 | …ai-assisted | 61 |
| total | 76352 | total | 312 |
For me personally, my biggest concern (which I am already seeing at an accelerated pace) is people stop trying to understand why code does what it does, they stop asking questions. I have seen Claude do some amazing things, but I also have at least one colleague who does not even care what it does or how it does it, only that he gets himself a nice little button on a nice little website to click. Heck, he forgot his search engine skills within weeks after we got AI-powered search results.
End users have to trust that the Nix contributors know what they are doing, that they understand it.