Post

Me learning Zig while Claude ships my projects in Zig

if llm-assisted contributions are reliably high quality, the relevant skill being demonstrated is prompt/harness engineering + judgment + verification. the contributor is demonstrating they can ship correct zig code.

It makes sense, but within a year they're going to have to make a tough decision on whether to fight back hackers with AI with humans, that's not a good idea.

I have mixed feelings here. In OSS infra projects, review time is often mostly about checking hidden edge cases and long tail maintenance, and AI PRs tend to spike both unless the author can explain every tradeoff.

I guess I don't see the natural way to draw hard lines, or why. Why is a PR specifically sacred? There's definitely something human and community-oriented we want to preserve. The common purpose of open source dev or many endeavors is intrinsically human. How to retain that is the question.

As ever the essential problem in the world is finding people who can think clearly to solve problems, see the larger complex picture and cut through it, maybe even find a simpler approach it. People who cannot do that will be poor supervisors of AI effort. There's a reason fizzbuzz cuts chaff.

For a project that already has a critical mass of devs contributing and real velocity, this seems reasonable. For other projects, where the vision outstrips the coding resources, or where the end products aren't tools for developers, this seems like a less viable option.

They need self-starters, they're too thinly staffed to hold hands. Different needs apply to different projects and that's clearly their situation.

But it’s definitely not true today and there’s many steps for AI yet to take before it could be used effectively like that. I would also posit that once (or if) we reach a state like that, software dev will be so different anyways that Zig itself may just go obsolete.

This response is worded like a corporate memo carefully written to hide what they actually belive in

Death by fork is always a possibility in open source, and sometimes a necessity. The future is not preordained and will be different for different projects.

Maintainers invest real time engaging with the humans behind PRs, helping newcomers learn and feel part of the project. If the “contributor” is just a proxy for AI, that value disappears. This argument is only going to get stronger.

The translation thing struck me as a response to one of the most common objections to their blanket ban I don't like it, but I respect it as being consistent with their overall position

At face value, treating any legit issue this way is extremely idiotic—and that's after giving them the benefit of doubt reading all this codeberg.org/ziglang/zig/...

I've landed a couple of small patches to the zig stdlib. They're easily the most friendly and helpful open source project I've encountered.