You and OP both work for the same "High Performance AI Inference" company, you might want to disclose that.
EDIT: and while you're at it, you might also want to work on your attitude. "you idiot", "get lost" and "you need to touch grass" are not helping any HN discussions
Well, in that case I'm curious... Why did you think hijacking OP's stance of "Claude did rewrote lots of my original messy code" with your own opposing position of "the project itself is not AI", and getting quite offensive about it, would benefit any discussion about this 13h old project?
It's a personal project of your dear ex-colleague, mind you!
Sure, the author just happened to one-shot a project and a landing page on new year's eve. And their writing style is just coincidentally very similar to Claude's.
The repo being a single commit doesn't mean it's AI. It is quite common to first develop on a private repo and then clean up the commit history for the first public release.
Its an extract of two weeks of work. And yes Claude did the website and rewrote my code that was absolutely without comments and a gigantic mess. It's an extract of the fourth attempt actually. Src4/ was the original folder. But my goal my to test the architecture applied to nats not to say I've done it without ai?
I'm also building a network server with thread-per-core and io_uring, except it's a web server, it's written in Rust, and io_uring is provided by a fork of Monoio runtime (I forked it to make it work with Windows and FreeBSD).
Unusual tones all around in the thread here. My initial observations before reading the comments here:
* "wow, OSS projects are starting to have some pretty wild landing pages, guess it's not just AI logos at the top of the README anymore"
* "wow, all in one commit. was it vibe-one-shotted, curated private work that was squashed, or something in between"
* "wow, Zig is kind easy to read although I really don't want to learn another language in 2026 although I already started learning some to use libghostty"
* "wow, is Zig really this much performant than Golang at the tails"
* "weird it uses Bazel, doesn't Zig have it's own build system like Golang"
* "so who is the author? I see they made an GitHub org for this. Are they going to keep doing stuff after the commit and should I keep this in my messaging queue neurons? Is this some company or person I should follow"
* "the README has a misalignment, do I PR that?"
* "oh cool, it lets you tune memory and the dispatcher"
---
I never thought of exactly how it manifested, except about the single commit. I have started "vibe coding" much more as the capabilities really improved in the last few months, so that isn't intrinsically a trash approach.
But the "who" and the "how" and the "why" do matter, in terms of whether one should look at it for education or infotainment or as a potential tool.
Disclosure of the intention and method would be courteous to the community when we create and share these things. Otherwise we'll all have high cognitive burden with the amount of projects we'll be seeing in 2026!
That’s fair, I should have framed it more clearly upfront. Thanks for the feedback.
I was excited about the results. The intent was to talk about performance and architecture, not to imply this was a quick or effortless project. There’s been a lot of iteration and experimentation behind it, and I should have communicated that context better as well as the use of AI for the help.
I totally get it and received the offering. =) Love seeing more use of io_uring too and interesting to see how that's done in Zig. Happy New Year: All the best on this and other projects.
Upvote for Bazel. I think these days I place a lot more value on how well an ecosystem slots into Bazel/friends because monorepos are increasingly more useful and relevant.
So nice to see there are good rules for Zig and that folks are using them.
Also ironically I think starting with Bazel/Buck/whatever your poison of choice is almost always a good move even if people tell you it's overkill. The easiest time to do it as at the beginning, all times after that is too hard and the marginal cost of building with it from the start is minimal.
People are free to knock themselves out with Bazel if they’re into that kind of masochism, but having it as the ONLY way to build your OSS project is a big no.
The problem with "the language tooling is already a build system" is that cross-language dependency chains are a thing. The moment you need a Rust or Zig file to be regenerated and recompiled when a JSON schema or .proto file is updated, you're outside what most of those language-specific toolchains can support. This is where Bazel absolutely shines.
If all of your dependencies need to use the same build system as your project then your build system/process is defect anyway. It should be possible to invoke a foreign build system as part of your build.
> I am assuming the message durability guarantees lean towards YOLO rather than ACID?
"Core" nats doesn't have durability. Nats jetstream is the api built on top of nats that in the main nats-server impl provides durability. Jepsen tested Nats Jetstream.
Also from your link:
> Regular NATS streams offer only best-effort delivery, but a subsystem, called JetStream, guarantees messages are delivered at least once.
The project linked here does not implement the nats jetstream api, just normal nats.
So yes, it seems its same (documented, understood) "yolo" as normal nats.
Putting aside the whole presenting slop-coded things as cromulent projects, is anyone else tired of this constant myopic focus on performance? I don't need a message queue that can "saturate the bandwidth of the next generation of hardware". Rather I want middleware that is easy to use - simple to set up (not hundreds of configuration knobs for optimizing performance), reliable (can run clustered on a few instances on its own, not using k666s or anything, and handle instances going away for a few weeks if one dies or I'm reconfiguring things), has good semantics that won't encourage Heisenbugs down the line (eg look at MQTT's actual semantics versus how it's incorrectly used by Home Assistant generic MQTT endpoints), and so on. I get that there's no surveillance industry money backing projects aimed at individual users, but it's still pretty sad that individuals creating projects in their spare time are still focusing on features desired by the surveillance industry.
Oh god no. Just having fun with zig and being a little over enthusiast I guess. I'm a big fan of nats, and really wanted to see how far you can push the idea if you do it differently. I was not expecting that tbh but, hpn too!
Claude did rewrote lots of my original messy code. No shame in that? But in the end the interest was in the underlying architecture, applied to nats protocol. Anyway.
Which is totally fine, I use Claude Code a bunch myself. I said nothing about shame, just that one single commit plus a website that looks vibe coded together has all the hallmarks of AI-driven code.
There may have been less pushback if this had been expressed up front. But also, what is it ? Is it to "test the architecture applied to nats" or is it to be a fully fledged NATs replacement (as per the impression given by table at the bottom of the website) - which becomes much harder if AI has significantly re-written the authors original code (and commented it badly).
Fair point. And as what it is, not a nats replacement, certainly dont have the time to maintain that this way, a test/tech demo/fun side project that yielded super interesting results is probably the answer. As usual I'm probably way too enthusiast when I see some nice results like that and the goal here was to talk about that, but it shifted super fast. So yes Claude rewrote lots of parts, and that's what I love about it. Testing an idea happens in way less time than before, and I find that super cool.
Nothing individually is a good indicator of slop in itself, a human could also have written this readme full of Claud-isms and a borked ASCII schema or the code littered with idiosyncratic comments.
It's the convergent set of clues that makes the case.
I’d invite you to reconsider the kind of language you’re using to interact with other forum members here.
Dropping to profanities is not conducive to maintaining an environment that’s facilitating dialogue between its members.
I’ve seen you at least twice call other members here “you idiot”, “get lost”. Etc. Have a hard think as to whether you could rephrase that without the name calling, and if at worse you can’t manage to, you can always ask an LLM to do it for you.
Is the only zig community that Andrew Kelley "anti-endorses"[1]:
> I don’t have the time or energy to evaluate most Zig communities so I can neither endorse nor anti-endorse them, however, the Zig subreddit is an exception.
> It’s an awful place and I stand by my decision to permanently close it. I am unhappy that it was reopened against my will.
I did a similar thing few days back just not with NATS protocol (Made it pure websocket based), and with rust. Couple of questions:
- Where did you get the machine to test your server on?
- Why did you end up going with zig?
Anyone can buy a 9950x on Amazon or any tech store, it's consumer hardware.
Given that this entire project is a single[1] vibe-coded commit, I really doubt the author bothered buying hardware to test it.
[1]: https://github.com/bustermq/bustermq/commits/master/
Who cares whether it’s vibe coded ? As long as it’s good and well maintained over time of course.
Maybe there is a niche market for artisanal software engineering where real humans make holes in punchcards, but I would not bet.
"Who cares whether it’s vibe coded"
Oh sweet summer child.
> Who cares whether it’s vibe coded ? As long as it’s good and well maintained over time of course
Maybe 12 hours after the first commit is a bit early to be confident about that…
> Maybe there is a niche market for artisanal software engineering where real humans make holes in punchcards, but I would not bet
Or maybe there exist a world between punchcards and evening AI slop “projects”, who knows.
he’s been working on it for 2 weeks, as he said somewhere else
And he later said that he doesn't intent to maintain it:
> And as what it is, not a nats replacement, certainly dont have the time to maintain that this way
[flagged]
You and OP both work for the same "High Performance AI Inference" company, you might want to disclose that.
EDIT: and while you're at it, you might also want to work on your attitude. "you idiot", "get lost" and "you need to touch grass" are not helping any HN discussions
sure, i hereby disclose that he left in September
i already said in another reply that i knew him personally
Well, in that case I'm curious... Why did you think hijacking OP's stance of "Claude did rewrote lots of my original messy code" with your own opposing position of "the project itself is not AI", and getting quite offensive about it, would benefit any discussion about this 13h old project?
It's a personal project of your dear ex-colleague, mind you!
because he was been sending me updates about it for the last two weeks
because we've been discussing architecture, optimisations and benchmarks
because it's not a 13h old project
Sure, the author just happened to one-shot a project and a landing page on new year's eve. And their writing style is just coincidentally very similar to Claude's.
Who's going to buy that?
The repo being a single commit doesn't mean it's AI. It is quite common to first develop on a private repo and then clean up the commit history for the first public release.
Its an extract of two weeks of work. And yes Claude did the website and rewrote my code that was absolutely without comments and a gigantic mess. It's an extract of the fourth attempt actually. Src4/ was the original folder. But my goal my to test the architecture applied to nats not to say I've done it without ai?
My personal rig and Zig because I worked with it for a little more than a year. It was a fun test to do.
I'm also building a network server with thread-per-core and io_uring, except it's a web server, it's written in Rust, and io_uring is provided by a fork of Monoio runtime (I forked it to make it work with Windows and FreeBSD).
You should at least try and align the ascii flowchart in the readme on the repo.
One day Claude will do it correctly but today is not that day.
Unusual tones all around in the thread here. My initial observations before reading the comments here:
* "wow, OSS projects are starting to have some pretty wild landing pages, guess it's not just AI logos at the top of the README anymore"
* "wow, all in one commit. was it vibe-one-shotted, curated private work that was squashed, or something in between"
* "wow, Zig is kind easy to read although I really don't want to learn another language in 2026 although I already started learning some to use libghostty"
* "wow, is Zig really this much performant than Golang at the tails"
* "weird it uses Bazel, doesn't Zig have it's own build system like Golang"
* "so who is the author? I see they made an GitHub org for this. Are they going to keep doing stuff after the commit and should I keep this in my messaging queue neurons? Is this some company or person I should follow"
* "the README has a misalignment, do I PR that?"
* "oh cool, it lets you tune memory and the dispatcher"
---
I never thought of exactly how it manifested, except about the single commit. I have started "vibe coding" much more as the capabilities really improved in the last few months, so that isn't intrinsically a trash approach.
But the "who" and the "how" and the "why" do matter, in terms of whether one should look at it for education or infotainment or as a potential tool.
Disclosure of the intention and method would be courteous to the community when we create and share these things. Otherwise we'll all have high cognitive burden with the amount of projects we'll be seeing in 2026!
That’s fair, I should have framed it more clearly upfront. Thanks for the feedback.
I was excited about the results. The intent was to talk about performance and architecture, not to imply this was a quick or effortless project. There’s been a lot of iteration and experimentation behind it, and I should have communicated that context better as well as the use of AI for the help.
I totally get it and received the offering. =) Love seeing more use of io_uring too and interesting to see how that's done in Zig. Happy New Year: All the best on this and other projects.
Is it bad if I work in private and then squash?
No, that’s very common and appreciated, especially when curated. All my comments there were musings, not value judgements.
Why use Basel instead of Zig build tools, as it’s all written in Zig anyway?
Probably by "mistake". Vibe code jerry AI just spewed a bazel build step, and author was clueless what is is, so he just went with it.
That was in the prompt.
Upvote for Bazel. I think these days I place a lot more value on how well an ecosystem slots into Bazel/friends because monorepos are increasingly more useful and relevant.
So nice to see there are good rules for Zig and that folks are using them.
Also ironically I think starting with Bazel/Buck/whatever your poison of choice is almost always a good move even if people tell you it's overkill. The easiest time to do it as at the beginning, all times after that is too hard and the marginal cost of building with it from the start is minimal.
agreed, love Bazel + BuildBuddy
Downvote for this web site is a horror movie billboard and zig already has a build system which is zig and that's one of it's neat features.
People are free to knock themselves out with Bazel if they’re into that kind of masochism, but having it as the ONLY way to build your OSS project is a big no.
Yeah I'm never touching Bazel again. I value my sanity.
well, bazel is by far the most reliable one so I'm not sure why you're complaining
The problem with "the language tooling is already a build system" is that cross-language dependency chains are a thing. The moment you need a Rust or Zig file to be regenerated and recompiled when a JSON schema or .proto file is updated, you're outside what most of those language-specific toolchains can support. This is where Bazel absolutely shines.
Zig build system can do all that just fine though
If all of your dependencies need to use the same build system as your project then your build system/process is defect anyway. It should be possible to invoke a foreign build system as part of your build.
and it would be terrible for hermeticity and reproducibility, nix tries very hard and gets mediocre results
perhaps, just perhaps, why people go through the trouble not because they are idiots but for actual engineering reasons
Rust build system can do all that just fine though
This looks fairly cool. If I had the production need for this, I’d definitely consider this.
I paired with Claude and simply added nats.c to the zig buildup system for my zig project at work. It works like a charm.
I am assuming the message durability guarantees lean towards YOLO rather than ACID? See also https://news.ycombinator.com/item?id=46196105
> I am assuming the message durability guarantees lean towards YOLO rather than ACID?
"Core" nats doesn't have durability. Nats jetstream is the api built on top of nats that in the main nats-server impl provides durability. Jepsen tested Nats Jetstream.
Also from your link:
> Regular NATS streams offer only best-effort delivery, but a subsystem, called JetStream, guarantees messages are delivered at least once.
The project linked here does not implement the nats jetstream api, just normal nats.
So yes, it seems its same (documented, understood) "yolo" as normal nats.
jepsen on nats still gives me anxiety
Comparison/benchmark to other alternatives?
Does it have a similar system to Jetstream? If yes, does it address the reliability issues Jetstream has been criticized about lately?
Absolutely not and will never have.
Wow! This entire project seem to be a big ball of AI slop.
your lack of skills doesn't mean it is (it's not)
Who would believe someone that writes something like this online?
Not a single serious dev would use crap like this near something that resembles production. Why even bother sharing a vibed spaghetti ball like this?
Putting aside the whole presenting slop-coded things as cromulent projects, is anyone else tired of this constant myopic focus on performance? I don't need a message queue that can "saturate the bandwidth of the next generation of hardware". Rather I want middleware that is easy to use - simple to set up (not hundreds of configuration knobs for optimizing performance), reliable (can run clustered on a few instances on its own, not using k666s or anything, and handle instances going away for a few weeks if one dies or I'm reconfiguring things), has good semantics that won't encourage Heisenbugs down the line (eg look at MQTT's actual semantics versus how it's incorrectly used by Home Assistant generic MQTT endpoints), and so on. I get that there's no surveillance industry money backing projects aimed at individual users, but it's still pretty sad that individuals creating projects in their spare time are still focusing on features desired by the surveillance industry.
[dead]
[flagged]
Oh god no. Just having fun with zig and being a little over enthusiast I guess. I'm a big fan of nats, and really wanted to see how far you can push the idea if you do it differently. I was not expecting that tbh but, hpn too!
nobody cares about the website being done with AI because the code of the project itself is not AI
you need to touch grass
The code of the project absolutely does look like it was done with AI lol, it’s a single commit…
Claude did rewrote lots of my original messy code. No shame in that? But in the end the interest was in the underlying architecture, applied to nats protocol. Anyway.
Which is totally fine, I use Claude Code a bunch myself. I said nothing about shame, just that one single commit plus a website that looks vibe coded together has all the hallmarks of AI-driven code.
Dude, when I move projects to GitHub I also often collapse everything into a single commit.
I do this to avoid having to check e-mail addresses and names in commits - maybe I mistakenly made a commit from my work account etc.
After the “initial” commit making it all public, I start to work “in the open”. I see many others doing it the same way.
That is NOT a reliable indicator of slop!
And the author has admitted at least some assistance here:
https://news.ycombinator.com/item?id=46452907 https://news.ycombinator.com/item?id=46452841
There may have been less pushback if this had been expressed up front. But also, what is it ? Is it to "test the architecture applied to nats" or is it to be a fully fledged NATs replacement (as per the impression given by table at the bottom of the website) - which becomes much harder if AI has significantly re-written the authors original code (and commented it badly).
The website being AI coded I can take or leave.
Fair point. And as what it is, not a nats replacement, certainly dont have the time to maintain that this way, a test/tech demo/fun side project that yielded super interesting results is probably the answer. As usual I'm probably way too enthusiast when I see some nice results like that and the goal here was to talk about that, but it shifted super fast. So yes Claude rewrote lots of parts, and that's what I love about it. Testing an idea happens in way less time than before, and I find that super cool.
It is super cool. People just want others to be honest about it in my experience, for better or worse.
Good thing it's not the only signal then.
Nothing individually is a good indicator of slop in itself, a human could also have written this readme full of Claud-isms and a borked ASCII schema or the code littered with idiosyncratic comments.
It's the convergent set of clues that makes the case.
[flagged]
I’d invite you to reconsider the kind of language you’re using to interact with other forum members here.
Dropping to profanities is not conducive to maintaining an environment that’s facilitating dialogue between its members.
I’ve seen you at least twice call other members here “you idiot”, “get lost”. Etc. Have a hard think as to whether you could rephrase that without the name calling, and if at worse you can’t manage to, you can always ask an LLM to do it for you.
the way he came barging through the door doesn’t invite sympathy
https://news.ycombinator.com/item?id=46453333
Unlike you, I won't insult you despite you being wrong.
That’s not an excuse for your behavior. Everyone here is expected to exercise a certain level of decorum.
> 1/ AIs are terrible at Zig
Claude isn't, and the zig subreddit is actually full of AI slop projects unfortunately.
And yesterday's front page Zig project was also vibe coded (though in that case the author acknowledged it: https://news.ycombinator.com/item?id=46442792)
> zig subreddit
Is the only zig community that Andrew Kelley "anti-endorses"[1]:
> I don’t have the time or energy to evaluate most Zig communities so I can neither endorse nor anti-endorse them, however, the Zig subreddit is an exception.
> It’s an awful place and I stand by my decision to permanently close it. I am unhappy that it was reopened against my will.
[1]: https://ziglang.org/community/
That doesn't change the fact that Claude code is definitely good enough at Zig to make vibe-coded projects.