Interesting to see this when the current top post on HN is someone worrying about Bun as it was acquired by Anthropic. The top comment there describes “Anthropic does experiments on their own codebase, the Bun team is not gonna do the same vibe coding experiments”.
Yet here we are, what looks like a massive undertaking for vibe coding.
Time will tell how this will turn out. Would be nice if the Bun maintainers could give some clarification about what they’re doing here, and why they’re doing this.
They recently tried to upstream an improvement to zig, but were prevented from doing so because zig has a hard and fast "no AI code" rule. Whether you think this response is trying to put pressure on zig or whether they're just moving for practical reasons is up to you.
Makes me wonder why zig announced the strict LLM rule recently. I'm afraid one reason could be that zig doesn't want to accept code from the bun fork in the first place (because of LLM usage, deviation and other reasons)
One non-obvious reason is that an important aspect of their community is to shepherd new contributors [1]. LLMs crushing everything would reduce that. More obvious is all the toil for maintainers dealing with LLM PRs (broadly it’s an issue). The Zig maintainers prefer to put their energy into improving people and fostering those relationship.
That's a solid reason to keep LLMs away from the kind of tasks that help with onboarding. But a patch series from a competent team that changes 3000 lines should probably be evaluated on its own merits. Or at least, the collaboration-based reasons to reject AI don't apply and the real reason would be something else.
(Though I don't know if this particular patch series would get accepted on its own merits.)
The recent article explained the bun patch would have been refused on technical merits as it's intrinsically incorrect, to be able to work properly it required some language changes.
Well said! I don't think either party is really at fault here, but if Anthropic wanted to contribute non-negligible amounts of code over time then it's an absolute dealbreaker.
Sucks for people who were invested in contributing to Bun and don't like working with AI tools to be sure, but I think the writing was on the wall for them pretty much immediately post-acquisition. You must admit, it's hard to predict that 100% of source lines will be written by AI if you're not walking the walk!
Yeah, I remember when the lazy bastards started writing programs using compilers instead of learning assembly language. Now I don’t have a single colleague who can write assembly. There’s whole generations now who can’t code assembly. Most don’t even know what a register is. Hope Zig holds against this latest attempt to make everyone stupid.
To add to the other commenters, loads of people don’t know assembly, which speaks to the quality of the average developer. The ones that still understand assembly to this day tend to be better developers, writing faster and more efficient code.
There are other reasons why a project like Zig might not want to accept LLM generated contributions.
Zig, as programming language, has a multiplier codebase. A bug may affect a significant larger portion of users than most libraries or binaries will, as it's a fundamental building block of everything that uses Zig. Just that could be worth the extra scrutiny on every individual commit.
There's also the usual arguments: copyright ethics, environmental ethics and maintainer burden.
It's a combination of pragmatism (not wanting to wade through slop, not wanting to shove out newbie developers) and politics (usual contemporary techie progressive stuff that's now oddly anti-technology).
Wow. That xkcd was written in 2007, and part of the dialog is "didn't that [meme] die like five years ago?" Which means All Your Base, as a meme, was already getting somewhat stale by around 2002. It's hard to believe it's been that long.
Rust is a significantly more mature language. Adoption of zig has to be done on the assumption that the language will significantly improve as your project evolves, and if those improvements don't agree with your project's goals you're in something of a lurch. Rust is basically finished and adopting it has to be done on the assumption it won't change very much. I don't know what their initial logic for adopting zig was, but I think porting to a more mature language was inevitable, unless by some miracle zig happened to rapidly mature in exactly the direction they wanted,
I was hoping bash because why not. It's AI that has to work and maintain anyway and Anthropic employees aren't limited by 5 hour 7 days limits anyway I suppose.
Rust is legit one of the best languages to "vibe code" in.
The emitted AST has a lower defect rate since it incorporates strong types and in-built error handling. Other pros include native code and portability, but downside is the compile time.
This could be a subjective feeling with no real data to back it up.
People say same about Go as well that it's type system and limited feature set makes it the best AI friendly language but there too, it just seems like a hunch rather than a proven fact.
The thing is that this argument doesn't work with Go because its type system (and the whole language, really) is much less expressive and compiler gives a lot less feedback to the LLM. So it tends to have to write more unit tests and do more cycles of testing (and spend more tokens) to get it right.
The argument about type system is absurd anyway. The types in a program aren't a universal vocabulary that the LLM would already know about like the words of English language. They are unique to each program and domain so an LLM can't be better at it.
Let me elaborate further - it's like the proficiency of LLMs in writing English vs writing Sawahili or Kurdish.
The types of a program are like Swahili or Kurdish etc even worse because those languages still have sizeable chuck on the Internet and digital archives but types of a program are very specific to it.
Studies have shown that natural human languages are all more or less equally expressive in terms of bits per second while speaking. There's lots of different ways they can be structured but they tend to follow common rules that have been well-characterized by linguists. They can be used to describe formal mathematical statements, but are not rigorously formal languages themselves.
Programming languages, in contrast, are constructed and vary much more in their designs. They are formal languages, making them closer to math than spoken language. LLMs being able to describe concepts more thoroughly and precisely through more expressive semantics obviously makes some languages more suitable than others.
The type system of a language is just one aspect of it that allows the language to provide guarantees to the LLM (and the user) about correctness of the code it's writing.
I am not speaking about specific types in specific programs. I am talking about the ability to describe complex constraints that LLMs (and humans) end up using to make writing correct code easier and more productive. Some programming languages absolutely are more effective at this than others, and that's always been true even before LLMs.
As a downside, the compile time is somewhat offset once you're using agents (and especially parallel agents) anyway. Since all of your edits cost a round-trip API call to a third party server, you can accept a slightly slower compile step.
Read the previous discussions on the topic. Your summary is a sensationalist lie, since their change was apparently a smoking pile of hot garbage, and Zig already had similar performance gains in a newer release.
> what looks like a massive undertaking for vibe coding
fwiw, I suspect it's less of an undertaking than you may think. I've been playing with AI to rewrite Postgres in Rust[0] over the past couple of weeks and I found the AI to be exceptional at doing rewrites. Having an existing codebase you can reference prevents a lot of the problems you have with vibecoding. You have an existing architecture that works well and have a test suite that you can test against
Over the course of a month I've gone from nothing to passing over 95% of the Postgres test suite. Given Jarred built Bun, I bet he'll be able to go much faster
> I suspect it's less of an undertaking than you may think... having an existing codebase you can reference prevents a lot of the problems you have with vibecoding.
Yeah, it's a distinction worth making, and the language for making it kind of sucks. Vibe coding means "AI does the whole thing", or "I use tab autocomplete" depending on who you ask. It's not a very useful term anymore, we need better ones.
My benchmark is basically, "are you letting the AI drive."
In this case, an AI appears to have written the migration guide...
It was and is a perfectly good term, but people started using it without regard for its definition. I don't know why people wouldn't misuse a "better" term the same way.
In this case I think the current zeitgeist (at least among zoomers and younger millennials) really loves the word "vibe". Once they hear of the term "vibe coding", they just want to be able to say it, even if what they're doing isn't really vibe coding.
And then that leaks outside their social and age groups, because other people hear the incorrect usage, get confused, and incorporate that confusion into their own use of the term.
I do not know if there's any overlap between these teams, but it seems like Anthropic itself is fairly invested in the Rust ecosystem.
They recently proposed some of their internal tools to be the official Rust implementation[0] of Connect RPC[1]. As a protobuf based library set, this includes a new Rust-based protobuf compiler, Buffa[2].
Zig is a moving target. 0.15 -> 0.16 includes some massive structural changes concerning IO and async/threading.
Claude has absolutely no idea what it's doing with bleeding edge zig unless you feed it source and guide it closely (in which case it's useful for focused work) - I'm building a game engine & tcp/udp servers with it and it requires a hands-on approach and actually understanding what's being built.
I imagine these are not really concerns with rust at this point.
In my ideal world the team behind bun would be putting in the work to keep up with modern zig, but it's starting to look like they are running mostly on vibes in which case rust might be a better choice.
> it requires a hands-on approach and actually understanding what's being built.
I think this is true regardless of what language you’re using.
I’ve built a lot in Zig and there’s no difference between vibing stuff in it versus TypeScript/React. Claude can “one-shot” them both, and will mimic existing code or grep the standard library to figure everything out.
Which isn't particularly difficult - the language docs and std source come with the installation, so all you need to do is tell Claude where those directories are in your skill/plugin/CLAUDE.md.
> and guide it closely (in which case it's useful for focused work)
It does struggle sometimes with writing code that compiles and uses the APIs correctly. My approach to that so far has been to write test blocks describing the desired interface + semantics, and asking Claude to (`zig test` -> fix errors) in a loop until all the tests pass.
You're already at a disadvantage having to stuff the context and spend extra tokens coercing the model in the correct direction compared to it already knowing what to do (rust, ts, go, etc.)
Here, I just did a quick test with claude.
1. "make a simple tcp echo server that uses rust"
compiles and runs - took a few seconds to generate.
2. "make a simple tcp echo server that uses zig"
result: compile error, took literal minutes of spinning and thinking to generate
response: "ziglang.org isn't in the allowed domains. Let me check if there's another way, or just verify the code compiles conceptually and present it clean."
/opt/homebrew/Cellar/zig/0.15.2/lib/zig/std/Io/Writer.zig:1200:9: error: ambiguous format string; specify {f} to call format method, or {any} to skip it
@compileError("ambiguous format string; specify {f} to call format method, or {any} to skip it");
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3. "make a simple tcp echo server that uses zig 0.16"
result: compile error:
zig build-exe main.zig
main.zig:30:21: error: no field named 'io' in struct 'process.Init.Minimal'
const io = init.io;
^~
4. "make a simple tcp echo server that uses zig 0.15"
result: compile error
zig build-exe main.zig
/nix/store/as1zlvrrwwh69ii56xg6yd7f6xyjx8mv-zig-0.15.2/lib/std/Io/Writer.zig:1200:9: error: ambiguous format string; specify {f} to call format method, or {any} to skip it
@compileError("ambiguous format string; specify {f} to call format method, or {any} to skip it");
Rust took seconds and just works. Zig examples took minutes and don't work out of the box. The DX & velocity isn't even close.
i mean, if zig is doing its best (inadvertently) at shooing off slop jockeys, then i already have more confidence that:
1. the language and stdlib are written by people who know what they're doing
2. packages in the ecosystem, at the barest level, are written by those who didn't leave after a few compile errors they couldn't reason about
The agents will churn their way through the errors. The new users whose learning material is out of date, as well as the existing users that have an insurmountable task in updating their code, will give up instead.
I think the changes are improvements, but there's a real cost to language churn, and every time it happens, the graveyard of projects grows just that little bit larger.
I wouldn't call any port "prudent". In general, taking mature software and doing any major rewrite is one of the riskiest thing you can do. It is a large scale attempt to fix what isn't broken.
Sometimes it is worth it, but it may also kill projects. A risky move. And AI doesn't help its cause. AI can save a lot of time when making ports, it is one of the things it does best, but it doesn't protect from regressions.
I am not using Bun in production, but if I was, I would consider it a risk. Not because of Rust vs Zig, but for changing things that work.
There is like 1,713 open PR's on the Bun repo. I'm assuming all are from Claude or robobun?. I guess this gives us an insight on what the claude-code workflow look likes. Crazy times.
What a weird take. I do a ton of OSS, and the act of writing code is what makes it fun for me. If I were forced to use an LLM to write all my OSS code, I would just not do it anymore.
> I expect OSS to go the opposite direction: no human contribution allowed.
How is it an incorrect interpretation? Jared is indeed pitching/suggesting/predicting that human contribution will not be allowed in the near future, i.e. banned.
"Pitching" generally means that the person making the pitch is endorsing and pushing for it. (This might also be a regional word meaning/usage difference type thing.)
The person upthread should have said "predicting".
Anthropic makes claude, claude can write Rust like a champ and struggles at Zig. It's a straightforward "training data" argument.
I think there are even longer term plays that Anthropic should be looking at, in this space, but it seems like they've decided rust is the right thing, so fair play. I would be (am!) thinking about making an LLM optimized high level language that you can generate / train on intensively because you control the language spec.
Claude doesn’t write Rust like a champ. It’s still miles ahead at js and python than it is at rust. It can do macros and single file optimizations but its gotten really stuck in type hell and tried to dyn everything on multiple occasions for me.
Claude struggling at Rust: not getting types correct, using the wrong abstractions, not implementing things correctly
Claude struggling at Zig: the above + memory safety issues if you run “fast” mode.
It is generally true that Rust code tends to be written in a way that the compiler catches the issue at compile time. The same is not as true for Zig, Python or JS
I'm reminded of the old joke "how to shoot yourself in the foot in 25 different languages". The first one was "C - you shoot yourself in the foot." Zig remains very close to that philosophy.
So the difference is not in writing new stuff but in maintaining the existing codebase. Rust's rigidity makes it potentially harder to break stuff compared to Zig's general flexibility. As a project grows and matures, different types of contributors naturally come in and it's unreasonable to expect everyone to learn about historical footguns that may have accumulated.
100%.
For many people, Bun is the only reason they've even heard of Zig. I'm not in a position to comment intelligently on comparative language features per se, but when it comes to mindshare and community size, Rust is a clear winner.
I would expect all LLMs are going to be better at Rust than Zig - a strong, thorough compiler will simply prevent more mistakes, and the benefits of a "simple" language decreases the larger the code base gets. The more abstractions exist, the less valuable "no hidden control flow" or "no hidden allocations" from the standard library get, and that's before you add the mother of all abstractions of vibe coding.
They do work well. But I still see the occasional type related issue or bug from refactoring that claude will introduce into javascript and python code. It seems to be happening less and less frequently as the models get better. But, the rust compiler catches real bugs in LLM code. I consider that a win.
Has anyone made any cross language benchmarks for LLMs? I wonder if rust's conceptual complexity makes it harder for LLMs to write? If all you care about is working software, which language is best for LLMs? Python, because there's more example code? Go or Java, because they're simpler languages? Ruby because its terse? Rust because of the compiler? I'd love to see a comparison!
But why should they? This just seems like the groundwork for an initial refactor and moving from one language to another. They haven't actually committed to switching from Zig to Rust yet. I mean, I get if you are an investor and you want to see if they are using their time effectively, but why would it matter to anyone else?
They’re not required to do so, but like I said, it would be nice, because it removes a lot of speculation. And development is in the open, so people notice what they’re doing.
Lots of people, me included, heavily invested their time and expertise into Bun, using it as a daily driver, to bundle production code or even using it in production as a JS/TS runtime. Of course, we are interested in Bun to stay a useful tool. The Anthropic acquisition was worrying enough on its own.
But there isn't any change in someone's expertise in Bun though, currently, just in development. Why would they have to dive you into a daily stand-up about their development process?
Honestly, this kind of thing seems to work quite well with vibe coding. If I remember correctly, the Ladybird JS engine was "vibe-ported" to Rust as well, and it passed 100% of the original test suite, in addition to new Rust tests.
The definition is at https://x.com/karpathy/status/1886192184808149383 and no that does not match what is in the branch. Systemically migrating a code base using an LLM does not match the defintion of vibe coding.
> I’m seeing people apply the term “vibe coding” to all forms of code written with the assistance of AI. I think that both dilutes the term and gives a false impression of what’s possible with responsible AI-assisted programming.
Then "vibe coding" is a useless term, if it just means "LLM-assisted coding". We might as well just say "LLM-assisted coding".
As much as I find the word "vibe" generally annoying (in all contexts), I actually really like "vibe coding" as "LLM did everything and I didn't even look at it". It's a succinct, useful way to describe that mode of doing things. Diluting it down to "LLM-assisted coding" makes it useless.
All language is "coined terms". The point is that if you dilute the definition of a term, you make the term useless. Evolution of a term isn't done automatically. Correcting terms such as these pushed the evolution in a more useful way. Also, evolution of language is not a magic spell that automatically forgives people on making language mistakes.
I think the definition of vibe coding is a bit fluid, in this case I just meant it to be “code fully generated by AI, possibly not fully reviewed by human eyes”. I agree that this definitely not “coding based purely off vibes”, and the approach looks legit.
The question isn't whether or not you'd get the same line count with a non-LLM tool. The question of whether or not it's vibe-coded depends on whether or not the committer actually reviewed and understood the new code. And with a 75k line difference, that seems unlikely.
It depends on what you mean by "vibe coding". Is AI coding based on an existing implementation vibe coding? What about only from a natural-language spec? How does manual reviewing affect whether or not it's vibe coding?
In practice all use of AI rapidly becomes vibe coding. Even if someone says they're going to carefully manually review everything that's generated, within a couple of days they get bored and just click approve.
This is just a matter of priorities - I use LLMs to write code every day and I have never put a single line of code up for review that I didn’t read and understand.
Porting from one typed language to another seems like a perfect use for LLMs. I can see the appeal of both languages and why to consider such an action (e.g., rust is a mainstream PL vs zig's cult status (no slight intended)).
I think the big difficulty here is that Rust's ownership model in particular tends to require certain kinds of control flow to avoid a bunch of weird churning/copying, which makes it not as straightforward of a port target from other imperative languages.
Like maybe you get the LLM to try _really hard_ to churn through everything, but this feels like a big case of "perils of the lack of laziness".
Of course if you have a good idea for how to deal with allocations etc "idiomatically" already maybe that works out well. And to the credit of the port guide writer bun seems to have its explicit allocations that are already mapping pretty well to Rust.
This is all wild conjecture, but I'd assume that teaching the LLM to do that mapping is an achievable goal and then it get's close to automatic -- effectively slurp the source AST into a rust AST and render.
My only experience with ports so far is Python to Go, and it's been near flawless (just enough stupid shit to make me feel justified to be in the loop).
It really isn't if you don't have the right abstractions.
Especially for memory management the right and wrong abstractions in Rust can lead to a factor of 5 or 10 extra amount of difficulty. The right memory management abstraction and your code can be a straight line port (or even cleaner!), the wrong one and you're going to just be spending a lot of tokens to have a machine spin around in circles trying to untie itself
GC'd languages don't have this problem, though obviously you can still generate stupid amount of pain for yourself by doing something wrong
I'm porting a large-ish delphi application to c sharp. It's been pretty hands-off except for converting to async and some language capability mismatch.
Interesting how times have changed. Back in 2015, the entire Go runtime (already a mature codebase) was rewritten from C to Go semi-automatically: one of the maintainers wrote a C-to-Go conversion tool (for a subset of C they used) so that it compiled and produced identical output, and then the resulting code was manually refactored to make the Go code more idiomatic and optimized. And now you can just ask a language model.
The big difference here is that the C-to-Go tool was presumably deterministic: running it over and over again should produce the exact same result. You can trust that result because the human wrote the conversion tool, understood it, tested it, and worked the bugs out.
The LLM is non-deterministic. You could have it independently do the conversion 10 times, and you'd get 10 different results, and some of them might even be wildly different. There's no way to validate that without reviewing it fully, in its entirety, each time.
That's not to say the human-written deterministic conversion tool is going to be perfect or infallible. But you can certainly build much more confidence with it than you can with the LLM.
Picking a pre 1.0 language to build your product always seemed like a bad choice to me. Purely on that basis and ignoring the recent drama this seems like a reasonable idea for tech debt pay down to me. Assuming automated conversion can work without making things worse, which is not exactly a given.
Linked commit is probably not the most convincing for this tagline. Here's a branch[0] of Claude mass rewriting Zig code into Rust which is currently at 773,950 additions and 151 deletions:
The problem with vibe coded re-writes is that you basically sign off on understanding the generated codebase at that point. Any historical knowledge of the codebase is gone.
the rust they've written (so far) is highly unidiomatic (and with a ton of unsafe). I can't speak to the zig part, but it seems plausible to me it is line-by-line, horrendous rust.
Whether or not they can clean it up is an interesting question.
It makes the git history a bit more confusing to follow if you want to see old changes, but I'm sure a simple wrapper to check for the zig equivalent files as well wouldn't be very difficult.
I wonder if a successful, albeit slower, approach would be to walk the git commit history in lockstep, applying the behavioral intent behind each commit. If they did this, I would be interested in knowing if they were able to skip certain bug fix commits because the Rust implementation sidestepped the problem.
this is an interesting idea and i might try it with something smaller. there are more than 15,000 commits to bun, so you’d have to have some sort of way to operate on groups of commits in one prompt to get that done without thousands and thousands of api requests
I want zig to succeed but given that zig is not yet 1.x I'd imagine a large code base like bun would have difficulties addressing major breaking changes. Also given the fact that bun is using a fork of zig https://x.com/bunjavascript/status/2048427636414923250?s=20
So I can't tell if the linked commit is an actual attempt or just an experiment but it did always strike me as odd to make a JS runtime in Zig when my impression was there were a lot of work-stopping compiler bugs at the time.
I'll be very interested in how this AI port turns out. I am involved in a number of active projects that are being held back by the language / framework is holding back the project, but where a rewrite would be too big of a project to undertake by using only human power.
I've had more success vibe coding Rust than I have in more dynamic languages. I suspect the strictness of the Rust compiler forces the AI agent to produce better code. Not sure. It could be just that I am less familiar with Rust so it feels like it's doing a better job.
Having written a JavaScript runtime in Rust in the past - Rust is an excellent choice. Not just due to the development experience, but also for embedders who want to consume the project as a a library (rather than a binary, e.g. node).
Not sure about vibe-coding it. While they aren't using v8, LLMs made it easier to understand v8 quirks and update v8 as they make weird changes every now and then. It couldn't write the runtime without help though.
That seems totally reasonable but I wonder if there was some head butting in non-public channels given Bun is one of the biggest players in Zig and planned to push through a change like that on their own.
Zig is a moving target that has breaking changes in every release (which is fine as they are sub-1.0). But that means that AI tools have been trained on outdated syntax/etc. Zig isn't that common, so there is even less training data to begin with.
Rust on the other hand is pretty established by now and has less breaking changes. It also has more compile-time safety-guarantees that makes vibe-coding a bit more confident.
In top of that, Zig has rejected their upstream contributions. So they'd have to maintain their own compiler in the long run, which is probably just technical debt to maintain.
Most of my vibe coding is in zig, and it has been my experience that Claude and Codex both keep up with zig changes just fine. Every now and then I catch them writing outdated code that they burn some tokens on, but my experience says your local codebases’s idioms will influence what gets generated enough to stop this from being a problem.
Probably an experiment due to Bun's PRs to Zig being rejected (Zig does not allow AI use). If Rust works well enough, and the alternative is maintaining a fork of Zig, I'd guess they'd go with Rust.
The anti-AI policy had nothing to do with Bun's PRs being rejected. This post[0] by a core zig maintainer explains why the PRs were low quality and subsequently rejected.
It seems there was an issue where the image API ignored the ICC Profile.(now fixed)
Any developer with experience implementing image formats would almost certainly avoid this mistake. This is a problem that cannot be solved with vibe coding. In this situation, the user is merely a guinea pig for bug fixes.
If nothing, it'll be good marketing material targeted at non-technical enterprise executives so that they pressurize their engineering teams in meetings that look people are porting such complicated things from one different language to totally different language then why are we not using AI effectively?!
When I first heard that bun was written in zig, I thought that was an odd choice for such a large project, mostly because the language is "unstable" and is still making significant breaking changes.
I would guess dealing with breaking changes is a big motivation for this.
The only Bun shipped product I've used in anger is OpenCode and I regularly run into segfaults on it. I doubt this is the reason for migration but every time it happens, it reminds me the real cost of unsafe code. That being said, Zig is an absolute pleasure to write and I can't wait until it has a real library ecosystem, Rust's greatest boon.
Bun is the largest project written in zig. And it isn't close. Bun is bigger than zig itself. Seems like zig isn't mature enough to handle Bun's needs, so I don't blame them at all for looking for off ramps. Only time will tell if rigidity from the zig team is worth the cost of losing Bun. It might be.
Zig won't be affected by Bun potentially moving to Rust, the language has been growing rapidly and one of the main proposals of Zig is "maintain it with Zig". It's ability to integrate with existing C code bases, as well as be a drop-in build replacement, has widespread use.
In addition, the link in the comment you replied to explains why the PRs Bun opened to Zig would have lowered the quality of the compiler and how Zig has achieved even greater speedups, with more widely applicable features like incremental compilation and the self-hosted backend.
I am not a fan of AI but my limited experience with running local small LLM's did show me that rewriting some scripts into a different language worked really well. So my guess is this will just turn out fine.
This feels more like a reaction to Zig's anti-LLM policy than anything. Anthropic would probably like to contribute something back to Zig at some point, but I doubt anyone would ever believe their PRs were not written by Claude.
Exactly, this is a direct response to Zig refusing to accept pull requests from Bun (and Anthropic). That situation forced Bun to maintain a fork of Zig, and it makes sense in the long term that they'd rather port their entire project to Rust.
I've really enjoyed Bun the past year or so, but the acquisition by Anthropic, Bun's codebase and documentation increasingly becoming AI slop, and this impulsive complete rewrite - all of it has ruined it for me and I'm actively moving off of Bun. I don't feel comfortable relying on it any longer.
April 26th - Bun announces they used AI to fork Zig so they could make an optimization for a 4x improvement
April 27th - Zig contributor mlugg clarifies why the specific optimizations Bun did were ill advised and wouldn't have been accepted in Zig, regardless of AI use [1]
May 4 - Bun is looking into Rust as an alternative.
This, to me, seems like total whiplash. Has anyone at Bun made a statement on why they're making such dramatic changes? It seems like the lesson to internalize from mlugg is not "switch to Rust"
Interesting. When I thought of Zig, I thought of Bun. In my mind it was the flagship application for that language. Is there another? I wonder how the Zig team feels about this. To me it seems like Rust has definitively won now.
It's not really shunned - it's the standard solution for async in Rust - but it's not the right solution for every project, especially if you have specific requirements for how your project's computation should be scheduled. I would guess that Bun is one of those projects, especially as it needs to be able to schedule JS async work itself.
The answer is in the next sentence: "Bun owns its event loop and syscalls." They clearly want to manage their use of threads explicitly, which is not _unusual_ for systems programming but probably less common. Note that `rayon` is different from most of these in that it has nothing to do with async Rust - it's a tool for spreading computation over a thread pool, very popular in non-async projects, but it would also go against their goals here.
tokio is great and it's pretty performant, but you pay an allocation for every future unless you do some complex organization of your futures.
Source: I worked on Deno, competed directly with Bun on HTTP performance (and won on some metrics).
Edit: and of course I typed future instead of task (aka "spawned future"). Thanks, child commenters below. Much of Deno was built on spawning futures that mapped to promises and doing it as fast as possible. I spent ages writing a future arena to optimize this stuff..
You only allocate on box futures, which are much more rare than naked futures - generally only used where object safety (essentially dyn support) is required. Even then some workarounds exist.
It's an async runtime. The whole async-await flow removes a little bit of scheduling control and adds some forced memory management in order to give you some nicer code in an application case, but if you're trying to build a runtime yourself I think you'd much rather retain control in this case. It's just hard to reason about.
You much rather have this runtime you're building manage task scheduling and allocation and all that. It's the most natural design choice to make.
Tokio is a general purpose async runtime. Much the same could probably be said for async-std (except IIRC they do have a barebones reactor for you to build your own on). In general, a general-purpose async runtime will do worse for highly specific tasks than a purpose-built one (especially e.g. NUMA).
I think avoiding async entirely might be a mistake, and I'm not entirely convinced anything better than a general-purpose async runtime might exist for a JS runtime (it itself is general purpose after all).
Avoiding std::fs is fucking bizarre to me: it's completely sync and is a really lightweight abstraction over syscalls.
my guess is they want to do AI/O as part of their event loop explicitly, and blocking a thread in a syscall waiting for an IOP (ala std::fs) isn't the vibe.
You shouldn't have to pull in big complex dependencies to do what should be primitive things. Zig is putting a strong and thought-out effort into getting async & parallelism "right" inside the stdlib. I'm honestly not up to speed with where rust is at with it at the moment, but last time I checked it was a bit of a mess.
Async is much harder to work with than sync+threading is. And while threads have more overhead in theory, in practice almost nobody is writing applications at such a scale where that overhead actually matters. So I don't blame them for eschewing async, there's likely no benefit for the project in it.
Interesting to see this when the current top post on HN is someone worrying about Bun as it was acquired by Anthropic. The top comment there describes “Anthropic does experiments on their own codebase, the Bun team is not gonna do the same vibe coding experiments”.
Yet here we are, what looks like a massive undertaking for vibe coding.
Time will tell how this will turn out. Would be nice if the Bun maintainers could give some clarification about what they’re doing here, and why they’re doing this.
They recently tried to upstream an improvement to zig, but were prevented from doing so because zig has a hard and fast "no AI code" rule. Whether you think this response is trying to put pressure on zig or whether they're just moving for practical reasons is up to you.
It's probably a bit of both.
Makes me wonder why zig announced the strict LLM rule recently. I'm afraid one reason could be that zig doesn't want to accept code from the bun fork in the first place (because of LLM usage, deviation and other reasons)
One non-obvious reason is that an important aspect of their community is to shepherd new contributors [1]. LLMs crushing everything would reduce that. More obvious is all the toil for maintainers dealing with LLM PRs (broadly it’s an issue). The Zig maintainers prefer to put their energy into improving people and fostering those relationship.
[1] https://kristoff.it/blog/contributor-poker-and-ai/
That's a solid reason to keep LLMs away from the kind of tasks that help with onboarding. But a patch series from a competent team that changes 3000 lines should probably be evaluated on its own merits. Or at least, the collaboration-based reasons to reject AI don't apply and the real reason would be something else.
(Though I don't know if this particular patch series would get accepted on its own merits.)
The recent article explained the bun patch would have been refused on technical merits as it's intrinsically incorrect, to be able to work properly it required some language changes.
Well said! I don't think either party is really at fault here, but if Anthropic wanted to contribute non-negligible amounts of code over time then it's an absolute dealbreaker.
Sucks for people who were invested in contributing to Bun and don't like working with AI tools to be sure, but I think the writing was on the wall for them pretty much immediately post-acquisition. You must admit, it's hard to predict that 100% of source lines will be written by AI if you're not walking the walk!
Yeah, I remember when the lazy bastards started writing programs using compilers instead of learning assembly language. Now I don’t have a single colleague who can write assembly. There’s whole generations now who can’t code assembly. Most don’t even know what a register is. Hope Zig holds against this latest attempt to make everyone stupid.
To add to the other commenters, loads of people don’t know assembly, which speaks to the quality of the average developer. The ones that still understand assembly to this day tend to be better developers, writing faster and more efficient code.
Generating AI code/PR is not the same as using compilers because of at least two things:
- the scale of how much and how fast you can generate code with AI vs how fast can you write code for compiler
- the mental model of what is being generated and how much the contributor understands and owns the generated code
Using an LLM isn't analogous to using a higher level language.
Your analogy falls apart because the "lazy bastards" still knew how to program and understood the code they were working on.
Vide-coders often don't read, let alone understand, the code they send for PRs.
There are other reasons why a project like Zig might not want to accept LLM generated contributions.
Zig, as programming language, has a multiplier codebase. A bug may affect a significant larger portion of users than most libraries or binaries will, as it's a fundamental building block of everything that uses Zig. Just that could be worth the extra scrutiny on every individual commit.
There's also the usual arguments: copyright ethics, environmental ethics and maintainer burden.
> has a multiplier codebase. A bug may affect a significant larger portion of users than most libraries or binaries will
Couldn't you say exactly the same about bun?
It's a combination of pragmatism (not wanting to wade through slop, not wanting to shove out newbie developers) and politics (usual contemporary techie progressive stuff that's now oddly anti-technology).
I like your username.
Possibly, but the Zig creator is active on Lobste.rs, where he's been vocally anti-LLM for a year now, so the timing could just be a coincidence.
>They recently tried to upstream an improvement to zig, but were prevented from doing so because zig has a hard and fast "no AI code" rule.
And will Rust team accept their vibe coded patches?
Anthropic just needs to buy Zig! Problem solved.
Perfect A/B experiment opportunity. Fork Zig, call the fork Zag.
Lock the syntax/api together for a couple of years. Allow AI code in Zag.
Review after a few years, see which is better.
...and rewrite it in rs.
Take off every Zig
It's time!
https://xkcd.com/286/
Wow. That xkcd was written in 2007, and part of the dialog is "didn't that [meme] die like five years ago?" Which means All Your Base, as a meme, was already getting somewhat stale by around 2002. It's hard to believe it's been that long.
Relevant XKCDs:
* https://xkcd.com/647/
* https://xkcd.com/1477/
So if tomorrow Rust denied the "improvement" to upstream Rust then what's the next language they plan to vibe code it in?
Rust is a significantly more mature language. Adoption of zig has to be done on the assumption that the language will significantly improve as your project evolves, and if those improvements don't agree with your project's goals you're in something of a lurch. Rust is basically finished and adopting it has to be done on the assumption it won't change very much. I don't know what their initial logic for adopting zig was, but I think porting to a more mature language was inevitable, unless by some miracle zig happened to rapidly mature in exactly the direction they wanted,
C obviously.
I was hoping bash because why not. It's AI that has to work and maintain anyway and Anthropic employees aren't limited by 5 hour 7 days limits anyway I suppose.
Perl
Rust is legit one of the best languages to "vibe code" in.
The emitted AST has a lower defect rate since it incorporates strong types and in-built error handling. Other pros include native code and portability, but downside is the compile time.
This could be a subjective feeling with no real data to back it up.
People say same about Go as well that it's type system and limited feature set makes it the best AI friendly language but there too, it just seems like a hunch rather than a proven fact.
The thing is that this argument doesn't work with Go because its type system (and the whole language, really) is much less expressive and compiler gives a lot less feedback to the LLM. So it tends to have to write more unit tests and do more cycles of testing (and spend more tokens) to get it right.
The argument about type system is absurd anyway. The types in a program aren't a universal vocabulary that the LLM would already know about like the words of English language. They are unique to each program and domain so an LLM can't be better at it.
Let me elaborate further - it's like the proficiency of LLMs in writing English vs writing Sawahili or Kurdish.
The types of a program are like Swahili or Kurdish etc even worse because those languages still have sizeable chuck on the Internet and digital archives but types of a program are very specific to it.
Studies have shown that natural human languages are all more or less equally expressive in terms of bits per second while speaking. There's lots of different ways they can be structured but they tend to follow common rules that have been well-characterized by linguists. They can be used to describe formal mathematical statements, but are not rigorously formal languages themselves.
Programming languages, in contrast, are constructed and vary much more in their designs. They are formal languages, making them closer to math than spoken language. LLMs being able to describe concepts more thoroughly and precisely through more expressive semantics obviously makes some languages more suitable than others.
The type system of a language is just one aspect of it that allows the language to provide guarantees to the LLM (and the user) about correctness of the code it's writing.
I am not speaking about specific types in specific programs. I am talking about the ability to describe complex constraints that LLMs (and humans) end up using to make writing correct code easier and more productive. Some programming languages absolutely are more effective at this than others, and that's always been true even before LLMs.
If we are gonna go down that rabbit hole, then the natural conclusion is Haskell.
Which seems pretty reasonable tbh. Claude Code is amazing with Elm in my experience.
Excellent comment.
As a downside, the compile time is somewhat offset once you're using agents (and especially parallel agents) anyway. Since all of your edits cost a round-trip API call to a third party server, you can accept a slightly slower compile step.
Even if AI had not been used, the changes would not have been upstreamed, see https://ziggit.dev/t/bun-s-zig-fork-got-4x-faster-compilatio... tl;dr the supposed improvements are not sound and the zig compiler has already gotten a whole lot faster
This should be the top comment in the whole thread. AI is not the point, the PR is just not of a good quality.
That is a devastating comment. I will now be extremely skeptical of bun.
Thanks, that is the answer.
What a sober, detailed forum post.
Read the previous discussions on the topic. Your summary is a sensationalist lie, since their change was apparently a smoking pile of hot garbage, and Zig already had similar performance gains in a newer release.
seems easier to fork zig
Then that becomes an ongoing effort. The rewrite is once. (Good idea or not)
good, more reason to stay away from zig
Stay away. Everyone wins.
> what looks like a massive undertaking for vibe coding
fwiw, I suspect it's less of an undertaking than you may think. I've been playing with AI to rewrite Postgres in Rust[0] over the past couple of weeks and I found the AI to be exceptional at doing rewrites. Having an existing codebase you can reference prevents a lot of the problems you have with vibecoding. You have an existing architecture that works well and have a test suite that you can test against
Over the course of a month I've gone from nothing to passing over 95% of the Postgres test suite. Given Jarred built Bun, I bet he'll be able to go much faster
[0] https://github.com/malisper/pgrust
> I suspect it's less of an undertaking than you may think... having an existing codebase you can reference prevents a lot of the problems you have with vibecoding.
That's because it's not vibe coding - stingraycharles doesn't seem to understand what vibe coding is. Vibe coding was defined here https://x.com/karpathy/status/1886192184808149383
> There's a new kind of coding I call “vibe coding”, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.
This is very far from Anthropic's migration plans.
Yeah, it's a distinction worth making, and the language for making it kind of sucks. Vibe coding means "AI does the whole thing", or "I use tab autocomplete" depending on who you ask. It's not a very useful term anymore, we need better ones.
My benchmark is basically, "are you letting the AI drive."
In this case, an AI appears to have written the migration guide...
It was and is a perfectly good term, but people started using it without regard for its definition. I don't know why people wouldn't misuse a "better" term the same way.
In this case I think the current zeitgeist (at least among zoomers and younger millennials) really loves the word "vibe". Once they hear of the term "vibe coding", they just want to be able to say it, even if what they're doing isn't really vibe coding.
And then that leaks outside their social and age groups, because other people hear the incorrect usage, get confused, and incorporate that confusion into their own use of the term.
Waiting until they decide to call non-assisted programming ‘unc coding’
i mean AI docs are usually the result of collabs between users and AI using /plan
with superpowers, i see a lot of specs -> impl plan -> execute plan
"Vibe coding" = "let Dario take the wheel" as ThePrimeagen puts it.
I do not know if there's any overlap between these teams, but it seems like Anthropic itself is fairly invested in the Rust ecosystem.
They recently proposed some of their internal tools to be the official Rust implementation[0] of Connect RPC[1]. As a protobuf based library set, this includes a new Rust-based protobuf compiler, Buffa[2].
[0]: https://github.com/orgs/connectrpc/discussions/7#discussionc...
[1]: https://connectrpc.com/
[2]: https://github.com/anthropics/buffa
I imagine claude is better at Rust than Zig?
Zig is a moving target. 0.15 -> 0.16 includes some massive structural changes concerning IO and async/threading.
Claude has absolutely no idea what it's doing with bleeding edge zig unless you feed it source and guide it closely (in which case it's useful for focused work) - I'm building a game engine & tcp/udp servers with it and it requires a hands-on approach and actually understanding what's being built.
I imagine these are not really concerns with rust at this point.
In my ideal world the team behind bun would be putting in the work to keep up with modern zig, but it's starting to look like they are running mostly on vibes in which case rust might be a better choice.
> it requires a hands-on approach and actually understanding what's being built.
I think this is true regardless of what language you’re using.
I’ve built a lot in Zig and there’s no difference between vibing stuff in it versus TypeScript/React. Claude can “one-shot” them both, and will mimic existing code or grep the standard library to figure everything out.
> unless you feed it source
Which isn't particularly difficult - the language docs and std source come with the installation, so all you need to do is tell Claude where those directories are in your skill/plugin/CLAUDE.md.
> and guide it closely (in which case it's useful for focused work)
It does struggle sometimes with writing code that compiles and uses the APIs correctly. My approach to that so far has been to write test blocks describing the desired interface + semantics, and asking Claude to (`zig test` -> fix errors) in a loop until all the tests pass.
You're already at a disadvantage having to stuff the context and spend extra tokens coercing the model in the correct direction compared to it already knowing what to do (rust, ts, go, etc.)
Here, I just did a quick test with claude.
1. "make a simple tcp echo server that uses rust"
compiles and runs - took a few seconds to generate.
2. "make a simple tcp echo server that uses zig"
result: compile error, took literal minutes of spinning and thinking to generate
response: "ziglang.org isn't in the allowed domains. Let me check if there's another way, or just verify the code compiles conceptually and present it clean."
/opt/homebrew/Cellar/zig/0.15.2/lib/zig/std/Io/Writer.zig:1200:9: error: ambiguous format string; specify {f} to call format method, or {any} to skip it @compileError("ambiguous format string; specify {f} to call format method, or {any} to skip it"); ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3. "make a simple tcp echo server that uses zig 0.16"
result: compile error:
zig build-exe main.zig main.zig:30:21: error: no field named 'io' in struct 'process.Init.Minimal' const io = init.io; ^~
4. "make a simple tcp echo server that uses zig 0.15"
result: compile error
zig build-exe main.zig /nix/store/as1zlvrrwwh69ii56xg6yd7f6xyjx8mv-zig-0.15.2/lib/std/Io/Writer.zig:1200:9: error: ambiguous format string; specify {f} to call format method, or {any} to skip it @compileError("ambiguous format string; specify {f} to call format method, or {any} to skip it");
Rust took seconds and just works. Zig examples took minutes and don't work out of the box. The DX & velocity isn't even close.
i mean, if zig is doing its best (inadvertently) at shooing off slop jockeys, then i already have more confidence that:
1. the language and stdlib are written by people who know what they're doing 2. packages in the ecosystem, at the barest level, are written by those who didn't leave after a few compile errors they couldn't reason about
The agents will churn their way through the errors. The new users whose learning material is out of date, as well as the existing users that have an insurmountable task in updating their code, will give up instead.
I think the changes are improvements, but there's a real cost to language churn, and every time it happens, the graveyard of projects grows just that little bit larger.
Contributors and maintainers will also be easier to find in Rust than Zig.
Zig is a great language and I want to see it succeed, but this is a prudent move for Bun.
I wouldn't call any port "prudent". In general, taking mature software and doing any major rewrite is one of the riskiest thing you can do. It is a large scale attempt to fix what isn't broken.
Sometimes it is worth it, but it may also kill projects. A risky move. And AI doesn't help its cause. AI can save a lot of time when making ports, it is one of the things it does best, but it doesn't protect from regressions.
I am not using Bun in production, but if I was, I would consider it a risk. Not because of Rust vs Zig, but for changing things that work.
This is likely irrelevant given bun has stopped taking community PR's entirely and Jarred is pitching that human contributors should be banned.
There is like 1,713 open PR's on the Bun repo. I'm assuming all are from Claude or robobun?. I guess this gives us an insight on what the claude-code workflow look likes. Crazy times.
Where is a source for either of these extraordinary claims?
https://x.com/jarredsumner/status/2048434628248359284
Wow, didn't realize how bad the situation was. Completely lost any respect and trust I had in the Bun project and its lead dev.
What a weird take. I do a ton of OSS, and the act of writing code is what makes it fun for me. If I were forced to use an LLM to write all my OSS code, I would just not do it anymore.
The gp's interpretation of that tweet is such a completely incorrect reading as to make one think it's likely disingenuous.
> I expect OSS to go the opposite direction: no human contribution allowed.
How is it an incorrect interpretation? Jared is indeed pitching/suggesting/predicting that human contribution will not be allowed in the near future, i.e. banned.
"Pitching" generally means that the person making the pitch is endorsing and pushing for it. (This might also be a regional word meaning/usage difference type thing.)
The person upthread should have said "predicting".
A prediction is not a policy.
Why didn't they use Rust in the first place then ? All this was true before AI
Anthropic only acquired Bun in December of last year. They weren't there in the first place, to make the decision.
I don't think Zig is different enough from rust or any other systems language for it to matter. If you can write rust you can write Zig.
Anthropic makes claude, claude can write Rust like a champ and struggles at Zig. It's a straightforward "training data" argument.
I think there are even longer term plays that Anthropic should be looking at, in this space, but it seems like they've decided rust is the right thing, so fair play. I would be (am!) thinking about making an LLM optimized high level language that you can generate / train on intensively because you control the language spec.
Claude doesn’t write Rust like a champ. It’s still miles ahead at js and python than it is at rust. It can do macros and single file optimizations but its gotten really stuck in type hell and tried to dyn everything on multiple occasions for me.
Claude struggling at Rust: not getting types correct, using the wrong abstractions, not implementing things correctly
Claude struggling at Zig: the above + memory safety issues if you run “fast” mode.
It is generally true that Rust code tends to be written in a way that the compiler catches the issue at compile time. The same is not as true for Zig, Python or JS
claude does not struggle with zig? not in my hands anyways.
I'm reminded of the old joke "how to shoot yourself in the foot in 25 different languages". The first one was "C - you shoot yourself in the foot." Zig remains very close to that philosophy.
So the difference is not in writing new stuff but in maintaining the existing codebase. Rust's rigidity makes it potentially harder to break stuff compared to Zig's general flexibility. As a project grows and matures, different types of contributors naturally come in and it's unreasonable to expect everyone to learn about historical footguns that may have accumulated.
100%. For many people, Bun is the only reason they've even heard of Zig. I'm not in a position to comment intelligently on comparative language features per se, but when it comes to mindshare and community size, Rust is a clear winner.
fwiw before today I'd heard of Zig and not Bun :D
something JS-adjacent could certainly be more known than an obscure language but are that many people using drop-in node replacements?
fwiw I knew about both but I had no idea Bun was written in Zig.
I would expect all LLMs are going to be better at Rust than Zig - a strong, thorough compiler will simply prevent more mistakes, and the benefits of a "simple" language decreases the larger the code base gets. The more abstractions exist, the less valuable "no hidden control flow" or "no hidden allocations" from the standard library get, and that's before you add the mother of all abstractions of vibe coding.
I have no doubt that LLMs are good at Rust.
But I can’t reconcile the reasoning about “strong, thorough compiler” with the fact that LLMs are also fantastic at Ruby.
They also write really great posix shell (including very sophisticated scripts) and python.
Something more subtle is going on.
They do work well. But I still see the occasional type related issue or bug from refactoring that claude will introduce into javascript and python code. It seems to be happening less and less frequently as the models get better. But, the rust compiler catches real bugs in LLM code. I consider that a win.
Has anyone made any cross language benchmarks for LLMs? I wonder if rust's conceptual complexity makes it harder for LLMs to write? If all you care about is working software, which language is best for LLMs? Python, because there's more example code? Go or Java, because they're simpler languages? Ruby because its terse? Rust because of the compiler? I'd love to see a comparison!
I think itnis ok to use or build vibe coded tools if it is built by experts in the domain and they take the ownership.
The industry does not shape bases on HN top posts, nor media buzz. Remember youtube birth. Necessity, available tech, fresh talent.
I believe now we have all but we fail at choosing.
But why should they? This just seems like the groundwork for an initial refactor and moving from one language to another. They haven't actually committed to switching from Zig to Rust yet. I mean, I get if you are an investor and you want to see if they are using their time effectively, but why would it matter to anyone else?
They’re not required to do so, but like I said, it would be nice, because it removes a lot of speculation. And development is in the open, so people notice what they’re doing.
Lots of people, me included, heavily invested their time and expertise into Bun, using it as a daily driver, to bundle production code or even using it in production as a JS/TS runtime. Of course, we are interested in Bun to stay a useful tool. The Anthropic acquisition was worrying enough on its own.
But there isn't any change in someone's expertise in Bun though, currently, just in development. Why would they have to dive you into a daily stand-up about their development process?
Honestly, this kind of thing seems to work quite well with vibe coding. If I remember correctly, the Ladybird JS engine was "vibe-ported" to Rust as well, and it passed 100% of the original test suite, in addition to new Rust tests.
> what looks like a massive undertaking for vibe coding
It doesn’t look like that at all. Do you think that all use of AI is vibe coding?
Did you look at the branch? This is vibed, even with the most liberal definition
https://github.com/oven-sh/bun/compare/claude/phase-a-port
This single commit is 65k lines of additions
https://github.com/oven-sh/bun/commit/ffa6ce211a0267161ae48b...
The definition is at https://x.com/karpathy/status/1886192184808149383 and no that does not match what is in the branch. Systemically migrating a code base using an LLM does not match the defintion of vibe coding.
There's a decent article by Simon Willison that talks about this: https://simonwillison.net/2025/Mar/19/vibe-coding/
> I’m seeing people apply the term “vibe coding” to all forms of code written with the assistance of AI. I think that both dilutes the term and gives a false impression of what’s possible with responsible AI-assisted programming.
You're right, all 750k lines of code added in a single day - definitely reviewed and completely understood.
The dilution of the term is a real problem sometimes.
But pointing your AI at an entire codebase to transpile pretty much entirely by itself? Yeah vibe coding is a fitting term.
Even if you wrote it a small essay on how to Rust. That improves the situation but doesn't change the core autonomy/hope of the task.
This is just a coined term; definitions evolve over time based on usage
Then "vibe coding" is a useless term, if it just means "LLM-assisted coding". We might as well just say "LLM-assisted coding".
As much as I find the word "vibe" generally annoying (in all contexts), I actually really like "vibe coding" as "LLM did everything and I didn't even look at it". It's a succinct, useful way to describe that mode of doing things. Diluting it down to "LLM-assisted coding" makes it useless.
All language is "coined terms". The point is that if you dilute the definition of a term, you make the term useless. Evolution of a term isn't done automatically. Correcting terms such as these pushed the evolution in a more useful way. Also, evolution of language is not a magic spell that automatically forgives people on making language mistakes.
I think the definition of vibe coding is a bit fluid, in this case I just meant it to be “code fully generated by AI, possibly not fully reviewed by human eyes”. I agree that this definitely not “coding based purely off vibes”, and the approach looks legit.
what would you call a fully uncommented commit with
"+27,939Lines changed: 27939 additions & 0 deletions"
of new rust code
The commit would look exactly like that if it was a 100% deterministic transpilation (like Golang did with their original C implementation?).
This is obviously very different from that, but the way the commit looks doesn't make it so.
The question isn't whether or not you'd get the same line count with a non-LLM tool. The question of whether or not it's vibe-coded depends on whether or not the committer actually reviewed and understood the new code. And with a 75k line difference, that seems unlikely.
Just another Monday in 2026.
The blind leading the blind.
I'm sure it will be called Systems Programing . Because Rust.
It depends on what you mean by "vibe coding". Is AI coding based on an existing implementation vibe coding? What about only from a natural-language spec? How does manual reviewing affect whether or not it's vibe coding?
In practice all use of AI rapidly becomes vibe coding. Even if someone says they're going to carefully manually review everything that's generated, within a couple of days they get bored and just click approve.
While I'm sure you're speaking for many, this is definitely not true across the board.
This is just a matter of priorities - I use LLMs to write code every day and I have never put a single line of code up for review that I didn’t read and understand.
Not to mention that manually writing code is itself a process of understanding. It cannot be replicated by reading code, no matter how carefully.
Porting from one typed language to another seems like a perfect use for LLMs. I can see the appeal of both languages and why to consider such an action (e.g., rust is a mainstream PL vs zig's cult status (no slight intended)).
I think the big difficulty here is that Rust's ownership model in particular tends to require certain kinds of control flow to avoid a bunch of weird churning/copying, which makes it not as straightforward of a port target from other imperative languages.
Like maybe you get the LLM to try _really hard_ to churn through everything, but this feels like a big case of "perils of the lack of laziness".
Of course if you have a good idea for how to deal with allocations etc "idiomatically" already maybe that works out well. And to the credit of the port guide writer bun seems to have its explicit allocations that are already mapping pretty well to Rust.
This is all wild conjecture, but I'd assume that teaching the LLM to do that mapping is an achievable goal and then it get's close to automatic -- effectively slurp the source AST into a rust AST and render.
My only experience with ports so far is Python to Go, and it's been near flawless (just enough stupid shit to make me feel justified to be in the loop).
It really isn't if you don't have the right abstractions.
Especially for memory management the right and wrong abstractions in Rust can lead to a factor of 5 or 10 extra amount of difficulty. The right memory management abstraction and your code can be a straight line port (or even cleaner!), the wrong one and you're going to just be spending a lot of tokens to have a machine spin around in circles trying to untie itself
GC'd languages don't have this problem, though obviously you can still generate stupid amount of pain for yourself by doing something wrong
I'm porting a large-ish delphi application to c sharp. It's been pretty hands-off except for converting to async and some language capability mismatch.
Interesting how times have changed. Back in 2015, the entire Go runtime (already a mature codebase) was rewritten from C to Go semi-automatically: one of the maintainers wrote a C-to-Go conversion tool (for a subset of C they used) so that it compiled and produced identical output, and then the resulting code was manually refactored to make the Go code more idiomatic and optimized. And now you can just ask a language model.
The slides: https://go.dev/talks/2015/gogo.slide#3
An interesting similarity:
>We had our own C compiler just to compile the runtime.
The Bun team maintain their own fork of Zig too
The big difference here is that the C-to-Go tool was presumably deterministic: running it over and over again should produce the exact same result. You can trust that result because the human wrote the conversion tool, understood it, tested it, and worked the bugs out.
The LLM is non-deterministic. You could have it independently do the conversion 10 times, and you'd get 10 different results, and some of them might even be wildly different. There's no way to validate that without reviewing it fully, in its entirety, each time.
That's not to say the human-written deterministic conversion tool is going to be perfect or infallible. But you can certainly build much more confidence with it than you can with the LLM.
Picking a pre 1.0 language to build your product always seemed like a bad choice to me. Purely on that basis and ignoring the recent drama this seems like a reasonable idea for tech debt pay down to me. Assuming automated conversion can work without making things worse, which is not exactly a given.
Linked commit is probably not the most convincing for this tagline. Here's a branch[0] of Claude mass rewriting Zig code into Rust which is currently at 773,950 additions and 151 deletions:
[0]: https://github.com/oven-sh/bun/compare/claude/phase-a-port
I was curious how much work this would be. Here are the top five from cloc:
The problem with vibe coded re-writes is that you basically sign off on understanding the generated codebase at that point. Any historical knowledge of the codebase is gone.
This prompt defines the translation as a file for file, line for line port. Seems like historical knowledge will be fine.
Having dabbled with both Zig and Rust, they do things so fundamentally differently, it isn’t possible to do exact lines like that.
the rust they've written (so far) is highly unidiomatic (and with a ton of unsafe). I can't speak to the zig part, but it seems plausible to me it is line-by-line, horrendous rust.
Whether or not they can clean it up is an interesting question.
If anyone can do it, it's Anthropic. The question is more how long it will take and how many tokens it will burn/how much groundwater.
care to attempt a top 3 differences that someone doing this kind of rewrite should know?
(would teach me a little about Zig, about which i know 0)
It makes the git history a bit more confusing to follow if you want to see old changes, but I'm sure a simple wrapper to check for the zig equivalent files as well wouldn't be very difficult.
I wonder if a successful, albeit slower, approach would be to walk the git commit history in lockstep, applying the behavioral intent behind each commit. If they did this, I would be interested in knowing if they were able to skip certain bug fix commits because the Rust implementation sidestepped the problem.
Many segfaults in Bun issue tracker. I bet it would sidestep many.
Well…there would still be panics.
this is an interesting idea and i might try it with something smaller. there are more than 15,000 commits to bun, so you’d have to have some sort of way to operate on groups of commits in one prompt to get that done without thousands and thousands of api requests
Interesting idea
I want zig to succeed but given that zig is not yet 1.x I'd imagine a large code base like bun would have difficulties addressing major breaking changes. Also given the fact that bun is using a fork of zig https://x.com/bunjavascript/status/2048427636414923250?s=20
Its never been easier to rewrite X in Rust than today.
Will everything eventually be rewritten in Rust and we finally achieve utopia?
So I can't tell if the linked commit is an actual attempt or just an experiment but it did always strike me as odd to make a JS runtime in Zig when my impression was there were a lot of work-stopping compiler bugs at the time.
I'll be very interested in how this AI port turns out. I am involved in a number of active projects that are being held back by the language / framework is holding back the project, but where a rewrite would be too big of a project to undertake by using only human power.
I've had more success vibe coding Rust than I have in more dynamic languages. I suspect the strictness of the Rust compiler forces the AI agent to produce better code. Not sure. It could be just that I am less familiar with Rust so it feels like it's doing a better job.
Yes it generates trash Rust code.
> Not sure. It could be just that I am less familiar with Rust so it feels like it's doing a better job.
Ya think?
Doy!
Given they have "unlimited" AI usage, do we expect the port to be complete tomorrow?
Having written a JavaScript runtime in Rust in the past - Rust is an excellent choice. Not just due to the development experience, but also for embedders who want to consume the project as a a library (rather than a binary, e.g. node).
Not sure about vibe-coding it. While they aren't using v8, LLMs made it easier to understand v8 quirks and update v8 as they make weird changes every now and then. It couldn't write the runtime without help though.
For those curious: https://github.com/alshdavid/ion
Why? Are there particular reasons that the maintainers of Bun feel the need to attempt to migrate from Zig to Rust?
Possibly related to https://simonwillison.net/2026/Apr/30/zig-anti-ai/ where the Bun team wanted to upstream work to Zig that was rejected by a blanket anti-LLM contribution policy.
Code origin was not even a factor https://ziggit.dev/t/bun-s-zig-fork-got-4x-faster-compilatio...
That seems totally reasonable but I wonder if there was some head butting in non-public channels given Bun is one of the biggest players in Zig and planned to push through a change like that on their own.
Even if there was anything in private channels, the reasons stated in that forum post are alone more than enough to reject Bun's Zig changes.
I wonder if they didn’t consider the problems of their changes in Zig what else do they not consider in Bun
Zig is a moving target that has breaking changes in every release (which is fine as they are sub-1.0). But that means that AI tools have been trained on outdated syntax/etc. Zig isn't that common, so there is even less training data to begin with.
Rust on the other hand is pretty established by now and has less breaking changes. It also has more compile-time safety-guarantees that makes vibe-coding a bit more confident.
In top of that, Zig has rejected their upstream contributions. So they'd have to maintain their own compiler in the long run, which is probably just technical debt to maintain.
Most of my vibe coding is in zig, and it has been my experience that Claude and Codex both keep up with zig changes just fine. Every now and then I catch them writing outdated code that they burn some tokens on, but my experience says your local codebases’s idioms will influence what gets generated enough to stop this from being a problem.
Probably an experiment due to Bun's PRs to Zig being rejected (Zig does not allow AI use). If Rust works well enough, and the alternative is maintaining a fork of Zig, I'd guess they'd go with Rust.
The anti-AI policy had nothing to do with Bun's PRs being rejected. This post[0] by a core zig maintainer explains why the PRs were low quality and subsequently rejected.
[0] https://ziggit.dev/t/bun-s-zig-fork-got-4x-faster-compilatio...
Also, if Zig itself doesn’t accept AI contributions, it’s probably NGMI unless somebody is willing to maintain that fork.
If the computer can do it for them, then why not?
Interesting. What are the main trade-offs they expect from the switch?
https://github.com/oven-sh/bun/issues/30197
It seems there was an issue where the image API ignored the ICC Profile.(now fixed) Any developer with experience implementing image formats would almost certainly avoid this mistake. This is a problem that cannot be solved with vibe coding. In this situation, the user is merely a guinea pig for bug fixes.
... and that bug was spotted in the canary release, reported and fixed.
Sounds like responsible open source software development to me. That's what pre-releases are for.
For better or for worse, at least Bun is open source, and the world is not lacking a NodeJS alternative.
What is the most interesting here for me is:
- a big, clear outcome and acceptance criteria, vibe coding project on
- a public, working, high performance, full featured, production codebase by
- the leading LLM model maker known for the strongest coding ability
A good example no matter if it successes or not.
If nothing, it'll be good marketing material targeted at non-technical enterprise executives so that they pressurize their engineering teams in meetings that look people are porting such complicated things from one different language to totally different language then why are we not using AI effectively?!
Comparing this claude/phase-a-port branch with main: “Showing 1,646 changed files with 773,950 additions and 151 deletions.”
When I first heard that bun was written in zig, I thought that was an odd choice for such a large project, mostly because the language is "unstable" and is still making significant breaking changes.
I would guess dealing with breaking changes is a big motivation for this.
The only Bun shipped product I've used in anger is OpenCode and I regularly run into segfaults on it. I doubt this is the reason for migration but every time it happens, it reminds me the real cost of unsafe code. That being said, Zig is an absolute pleasure to write and I can't wait until it has a real library ecosystem, Rust's greatest boon.
the rust port (at least currently) heavily uses unsafe as well
https://github.com/oven-sh/bun/compare/claude/phase-a-port#d...
that isn't particularly surprising, but the point is I would expect getting things more stable than the zig version would take a bit.
This is a huge loss for the zig language and community.
As a fan of the language, I hope it leads to some reflection on things that might need to change moving forward.
I think it reflects more on Bun. [1].
[1] https://ziggit.dev/t/bun-s-zig-fork-got-4x-faster-compilatio...
Bun is the largest project written in zig. And it isn't close. Bun is bigger than zig itself. Seems like zig isn't mature enough to handle Bun's needs, so I don't blame them at all for looking for off ramps. Only time will tell if rigidity from the zig team is worth the cost of losing Bun. It might be.
Zig won't be affected by Bun potentially moving to Rust, the language has been growing rapidly and one of the main proposals of Zig is "maintain it with Zig". It's ability to integrate with existing C code bases, as well as be a drop-in build replacement, has widespread use.
In addition, the link in the comment you replied to explains why the PRs Bun opened to Zig would have lowered the quality of the compiler and how Zig has achieved even greater speedups, with more widely applicable features like incremental compilation and the self-hosted backend.
Didn't they write a whole blog post on why they chose Zig over Rust?
I think they are simply experimenting to fully exploit Claude's models' powerful capabilities.
I suspect that an experiment is being run. In any case, that'll be a hell of a story!
this isn't vibe coding. this is vibe rewriting. ~500k lines of code. nobody is reading those diffs line by line. nobody.
https://x.com/bunjavascript/status/1966806250827714736
Haha, is it really okay not to retract that that the official account previously posted a caricature criticizing Rust?
Yes, it's quite ok to not "retract" a goofy image from months ago. It's harmless fun.
Could just be an experiment or something. It's Monday, the week is young
I am not a fan of AI but my limited experience with running local small LLM's did show me that rewriting some scripts into a different language worked really well. So my guess is this will just turn out fine.
what a win
Probably a good thing for the project even if the only net positive ends up being the Bun team stops maintaining a fork of Zig.
maybe anthropic should‘ve just acquired deno
Which makes one think, why they did not buy deno at first place then?
If they did, I guess they would rewrite deno in C++
Here we go again ...
Company A buys company B. A's management decrees the henceforth B's aqcuihired team must comply with company A's standards.
Second system effect kicks in. Bugs multiply.
Half of original company B devs leave.
I'm investigating whether future projects should revert to using Deno.
> Read this whole document before writing any code.
Hm does that actually work?
oh for christ’s sake
Aside from Zig's anti-AI stance and maintaining their own Zig fork, I think this port will showcase that Anthropic can re-engineer a massive codebase.
As an aside, I've been bitten by Zig's breaking changes on my own projects as well. It's taken the shine off of Zig and I'm looking at alternatives.
Any confirmation that a genuine port is underway? This might just be an experiment.
How well does that long translation prompt work?
This feels more like a reaction to Zig's anti-LLM policy than anything. Anthropic would probably like to contribute something back to Zig at some point, but I doubt anyone would ever believe their PRs were not written by Claude.
Exactly, this is a direct response to Zig refusing to accept pull requests from Bun (and Anthropic). That situation forced Bun to maintain a fork of Zig, and it makes sense in the long term that they'd rather port their entire project to Rust.
I've really enjoyed Bun the past year or so, but the acquisition by Anthropic, Bun's codebase and documentation increasingly becoming AI slop, and this impulsive complete rewrite - all of it has ruined it for me and I'm actively moving off of Bun. I don't feel comfortable relying on it any longer.
Zig said they wouldn't have accepted the changes without AI either.
You're right, I was just now reading the full explanation. https://ziggit.dev/t/bun-s-zig-fork-got-4x-faster-compilatio...
This makes me respect Zig team's stance more, that it's a technical decision more than an ideological one.
the days are not far when golang will be ported to rust.
Watch your mouth.
I hope they ship and use this. It’ll be a super interesting case study in a few years.
"Claude, migrate bun to Rust, make no mistakes"
I fully support this decision
April 26th - Bun announces they used AI to fork Zig so they could make an optimization for a 4x improvement
April 27th - Zig contributor mlugg clarifies why the specific optimizations Bun did were ill advised and wouldn't have been accepted in Zig, regardless of AI use [1]
May 4 - Bun is looking into Rust as an alternative.
This, to me, seems like total whiplash. Has anyone at Bun made a statement on why they're making such dramatic changes? It seems like the lesson to internalize from mlugg is not "switch to Rust"
[1] https://lobste.rs/s/ifcyr1/contributor_poker_zig_s_ai_ban#c_...
Interesting. When I thought of Zig, I thought of Bun. In my mind it was the flagship application for that language. Is there another? I wonder how the Zig team feels about this. To me it seems like Rust has definitively won now.
Ghostty is mainly Zig aside from the UI parts.
That TigerBeetle database I think.
>*No `tokio`, `rayon`, `hyper`, `async-trait`, `futures`.* No `std::fs`,
I'm not a rust dev but even I kind of notice that tokio is kind of shunned in most projects. Why is that? Is it just bad or what?
It's not really shunned - it's the standard solution for async in Rust - but it's not the right solution for every project, especially if you have specific requirements for how your project's computation should be scheduled. I would guess that Bun is one of those projects, especially as it needs to be able to schedule JS async work itself.
The answer is in the next sentence: "Bun owns its event loop and syscalls." They clearly want to manage their use of threads explicitly, which is not _unusual_ for systems programming but probably less common. Note that `rayon` is different from most of these in that it has nothing to do with async Rust - it's a tool for spreading computation over a thread pool, very popular in non-async projects, but it would also go against their goals here.
tokio is great and it's pretty performant, but you pay an allocation for every future unless you do some complex organization of your futures.
Source: I worked on Deno, competed directly with Bun on HTTP performance (and won on some metrics).
Edit: and of course I typed future instead of task (aka "spawned future"). Thanks, child commenters below. Much of Deno was built on spawning futures that mapped to promises and doing it as fast as possible. I spent ages writing a future arena to optimize this stuff..
You only allocate on box futures, which are much more rare than naked futures - generally only used where object safety (essentially dyn support) is required. Even then some workarounds exist.
Edit: and tasks.
Do you mean allocate on every task?
It's an async runtime. The whole async-await flow removes a little bit of scheduling control and adds some forced memory management in order to give you some nicer code in an application case, but if you're trying to build a runtime yourself I think you'd much rather retain control in this case. It's just hard to reason about.
You much rather have this runtime you're building manage task scheduling and allocation and all that. It's the most natural design choice to make.
In pretty much every bit of code I've written both professionally and leisurely I have always used tokio.
However, there are reasons why you might not want to use it:
- You don't need async at all
- You want to own the async execution polling completely
- You want some alternative futures executor like io uring (even though tokio-uring is a thing)
Tokio is a general purpose async runtime. Much the same could probably be said for async-std (except IIRC they do have a barebones reactor for you to build your own on). In general, a general-purpose async runtime will do worse for highly specific tasks than a purpose-built one (especially e.g. NUMA).
I think avoiding async entirely might be a mistake, and I'm not entirely convinced anything better than a general-purpose async runtime might exist for a JS runtime (it itself is general purpose after all).
Avoiding std::fs is fucking bizarre to me: it's completely sync and is a really lightweight abstraction over syscalls.
my guess is they want to do AI/O as part of their event loop explicitly, and blocking a thread in a syscall waiting for an IOP (ala std::fs) isn't the vibe.
You shouldn't have to pull in big complex dependencies to do what should be primitive things. Zig is putting a strong and thought-out effort into getting async & parallelism "right" inside the stdlib. I'm honestly not up to speed with where rust is at with it at the moment, but last time I checked it was a bit of a mess.
Async is much harder to work with than sync+threading is. And while threads have more overhead in theory, in practice almost nobody is writing applications at such a scale where that overhead actually matters. So I don't blame them for eschewing async, there's likely no benefit for the project in it.
You try to use it you'll get it. Otherwise it's just words. Like these: rust failed at async.
Async is an anti-pattern but sometimes inexperienced developers don't realize that and will infect your codebase with it.
Please explain.
it will make it more portable.
you can use both zig and rust in a single project, duh
instead of writing it once in C++
Makes sense on merit. There really isn’t room for Zig when Rust exists, is more ergonomic, and also safe.
hahaha eat your heart out "don't port it to rust" gang
I don't think problem ever is Rust, Rust is by far the best systems programming language.
Problem is fanboys like YOU.
People are asking why they would switch from zig to rust. I wonder the opposite: why would anyone would use zig over rust?