So the "Bun being a vibe coded rewrite from Zig to Rust" is a different story than I think it's being sold as, from everything I can see.
-
@cwebber that I'm now paranoid that by "doing research on competing implementations" he means "copycatting Kat's work" and now I'm glad that I've stopped working on orogene in public.
-
-
-
So the "Bun being a vibe coded rewrite from Zig to Rust" is a different story than I think it's being sold as, from everything I can see. It's worth seeing several events laid out back to back.
- Zig has a no AI contributions policy. They've explained their rationale: https://kristoff.it/blog/contributor-poker-and-ai/
- Bun has a fork of Zig with AI generated changes. Zig's developers explain why they aren't merging them, and that they have their own path to relevant improvements https://ziggit.dev/t/bun-s-zig-fork-got-4x-faster-compilation-times/15183/19
- Anthropic acquires Bun https://www.anthropic.com/news/anthropic-acquires-bun-as-claude-code-reaches-usd1b-milestone
- Bun is working on an experimental rewrite from Zig to Rust using Claude driven by the Bun lead, now working for Anthropic https://news.ycombinator.com/item?id=48016880What does Bun's lead have to say about AIgen contributions to FOSS?
> I expect OSS to go the opposite direction: no human contribution allowed. Slop will be a nostalgic relic of 2025 & 2026.
Jarred Sumner (@jarredsumner)
I expect OSS to go the opposite direction: no human contribution allowed. Slop will be a nostalgic relic of 2025 & 2026.
Nitter (xcancel.com)
What can we make of this?
The thing I laugh about in all of this, the thing that is absolutely the funniest, is that while these AI generated or "rewritten" projects may be Open Source, it will never be free.
It was created using a sizable investment.
It can only be maintained by additional investment.
No one, external, will want or be able to contribute, meaning no prospect of continuance.There will absolutely be licensing fees, because money has to come from somewhere, which absolutely undercuts whatever value they might think they're offering. That kind of renders these projects as instant abandon-ware, and any systems architect worth their salt is absolutely going to spot that red flag.
Truly, the learning opportunities are endless.
-
@cthos @janl @cwebber to be clear: I easily share my code with folks. I don't mind the "spirit" of foss. I take issue with having my work thrown into a soulless machine devoid of any of the community I would otherwise be building. I think it's healthy for us to look at each others' implementations and compare notes and have done a lot of that, and a lot of sharing of my own.
but honestly, fuck Jarred in particular. He's always been kind of a dickbag.
-
@cthos @janl @cwebber to be clear: I easily share my code with folks. I don't mind the "spirit" of foss. I take issue with having my work thrown into a soulless machine devoid of any of the community I would otherwise be building. I think it's healthy for us to look at each others' implementations and compare notes and have done a lot of that, and a lot of sharing of my own.
but honestly, fuck Jarred in particular. He's always been kind of a dickbag.
-
-
Note that this rewrite has a lot of resources at its disposal. The porting.md file has a ton of direction, more than historically an LLM can keep track of: https://github.com/oven-sh/bun/blob/46d3bc29f270fa881dd5730ef1549e88407701a5/docs/PORTING.md
But Anthropic acquired Bun, so I imagine this project has effectively as many tokens to use as it wants.
If this rewrite succeeds, this will effectively not only be a "fuck you Zig for not accepting my AIgen patches", it'll be a big headline and marketing story for Anthropic: look! Claude could rewrite Bun from Zig to Rust!
It's hard for me to read this then as anything but an experiment to see if Anthropic can get a big marketing boost in favor of vibecoding from.
@cwebber hard for an LLM, but in 2026 these kinds of things are not done just by an LLM. Multiple parallel and subagents using different models, that can reference multiple files and go back and check on these instructions and correct mistakes, investigating changes from git history, running tests, creating new tests, and so on.
Of course, it is all running on LLMs in the end, and using thousands of dollars of tokens. Which he can get for free from his employer, of course.
-
@cwebber I don't understand being the lead of a project, ostensibly understanding most of the code and its history, and slopping it up into an AI translation that you don't know at all. Yes, please turn my carefully handcrafted artifact into a pile of legacy code, at scale.
@federicomena @cwebber There's also Andreas Kling/Ladybird making similar announcement ~3 months ago. https://ladybird.org/posts/adopting-rust/
You and I have diff views on the matter (I like genAI for code, within reason). I read Kling's move as: I have goals w/ ladybird, ie (1) making a new high quality browser and (2) having as much fun/enjoyment in the process as it's fair. C++ is kinda getting in the way of (1), but manual rewrite is polar opposite of (2), so let's see if genAI can help eat cake + have cake. -
@federicomena @cwebber There's also Andreas Kling/Ladybird making similar announcement ~3 months ago. https://ladybird.org/posts/adopting-rust/
You and I have diff views on the matter (I like genAI for code, within reason). I read Kling's move as: I have goals w/ ladybird, ie (1) making a new high quality browser and (2) having as much fun/enjoyment in the process as it's fair. C++ is kinda getting in the way of (1), but manual rewrite is polar opposite of (2), so let's see if genAI can help eat cake + have cake.@federicomena @cwebber But Federico's point "I don't understand being the lead ..." applies 5x to Ladybird. For Bun, Christine points out these folks know 100% what they're doing (epic marketing stunt + savage diss at Zig with its no-AI policy). Kling and Ladybird is the "Got no horse in the race, and yet I think AI-assist is the way to go."
-
-
But this won't be an ordinary vibecoding story if it succeeds, but it might *look that way* from a marketing perspective. It has the lead maintainer giving a *ton* of guidance and perspective. It has the resources of Anthropic itself at its disposal. And it has the perfect story attached to it, so there's a lot to win if it succeeds, and little to lose if it doesn't, since the failure can just be ignored and they can move on.
@cwebber build many to throw away is the slop coding way.
Like LLMs in general it can sometimes be useful, but used a lot of resources, even more if you are not careful.
Still: port from one language to another is something they are at least fairly good for.
-
Note that this rewrite has a lot of resources at its disposal. The porting.md file has a ton of direction, more than historically an LLM can keep track of: https://github.com/oven-sh/bun/blob/46d3bc29f270fa881dd5730ef1549e88407701a5/docs/PORTING.md
But Anthropic acquired Bun, so I imagine this project has effectively as many tokens to use as it wants.
If this rewrite succeeds, this will effectively not only be a "fuck you Zig for not accepting my AIgen patches", it'll be a big headline and marketing story for Anthropic: look! Claude could rewrite Bun from Zig to Rust!
It's hard for me to read this then as anything but an experiment to see if Anthropic can get a big marketing boost in favor of vibecoding from.
@cwebber porting Bun from to rust is more than possible, Deno proved that rust is capable of building a JavaScript runtime. I agree with you, it looks like a "big headline" grab in the making.
-
But this won't be an ordinary vibecoding story if it succeeds, but it might *look that way* from a marketing perspective. It has the lead maintainer giving a *ton* of guidance and perspective. It has the resources of Anthropic itself at its disposal. And it has the perfect story attached to it, so there's a lot to win if it succeeds, and little to lose if it doesn't, since the failure can just be ignored and they can move on.
Update: Merged into main already https://github.com/oven-sh/bun/pull/30412
-
Update: Merged into main already https://github.com/oven-sh/bun/pull/30412
@cwebber Jesus Christ.
-
So the "Bun being a vibe coded rewrite from Zig to Rust" is a different story than I think it's being sold as, from everything I can see. It's worth seeing several events laid out back to back.
- Zig has a no AI contributions policy. They've explained their rationale: https://kristoff.it/blog/contributor-poker-and-ai/
- Bun has a fork of Zig with AI generated changes. Zig's developers explain why they aren't merging them, and that they have their own path to relevant improvements https://ziggit.dev/t/bun-s-zig-fork-got-4x-faster-compilation-times/15183/19
- Anthropic acquires Bun https://www.anthropic.com/news/anthropic-acquires-bun-as-claude-code-reaches-usd1b-milestone
- Bun is working on an experimental rewrite from Zig to Rust using Claude driven by the Bun lead, now working for Anthropic https://news.ycombinator.com/item?id=48016880What does Bun's lead have to say about AIgen contributions to FOSS?
> I expect OSS to go the opposite direction: no human contribution allowed. Slop will be a nostalgic relic of 2025 & 2026.
Jarred Sumner (@jarredsumner)
I expect OSS to go the opposite direction: no human contribution allowed. Slop will be a nostalgic relic of 2025 & 2026.
Nitter (xcancel.com)
What can we make of this?
@cwebber It'll get interesting if/when Rust also commits to a "no AI contributions to the compiler + core" policy...
-
@Gnuxie It's been Zig as long as I've known about it
-
@Gnuxie It's been Zig as long as I've known about it
@Gnuxie Along with Tigerbeetle, was one of the flagship Zig projects
-
M mttaggart@infosec.exchange shared this topic
