My Thoughts on the Bun Rust Rewrite - Andrew Kelley

matklad July 9, 2026, 10:16am 2

On the “style guide” part, my perspective is here:

One thing about the whole ordeal is that people say it costed 160k USD to do the rewrite, but that’s not really correct in my opinion.

Usually when you say ‘rewrite’ you mean writing the final version of the code that gets you back to feature parity, but in this case Bun is not quite there yet (i.e. if they had done a traditional rewrite, by this point they would have an idiomatic Rust codebase, but they don’t have that).

So if anything it’s 160k USD to get deeper into your technical debt rabbit hole, with more money required to climb back to the outcome of a ‘proper’ rewrite.

And to be clear I’m not putting into question whatsoever whether they can do it or not. They can without a doubt do it, the point is the actual cost, and I’m my opinion we’re being lied to in that sense. The full rewrite will cost a lot more than 160k USD, and it won’t take 11 days to get there.

This is in my opinion, above all else, what to watch out for when evaluating how whether to use LLMs or not, and how to evaluate the Bun rewrite.

The feasibility question is a red herring. Provided with a tool that can even just flip a bit at a time, you can eventually achieve any outcome. LLMs are obviously better than that, and it is certainly impressive to some degree that Fable could get them to this point that quick (although not exactly cheaply), but the real question is and always has been the cost, and that’s where you should expect to find the one sleight of hand trick that truly matters.

Moortu July 9, 2026, 11:52am 4

the $160.000 is probably what it would cost if you bought the tokens.
But since they were acquired by anthropic, I think they basically got them at cost price, or for free and anthropic paid for it, which is still cost price.

Or perhaps it’s even the cost price they mention and are not telling how much it would actually have cost to port it. The non transparency makes it bad either way

vulpesx July 9, 2026, 11:59am 5

You are missing the point that this is, in part, an ad for anthropic; who’s potential customers are going to be evaluating the cost for whatever they might plan on using their LLMs for.

The nuance of how much it actually cost bun, a subsidiary of anthropic, is irrelevant.

Moortu July 9, 2026, 12:06pm 6

My point is that the number is a lie either way and at the same time its meaningless.
But yeah, poorly written.

invlpg July 9, 2026, 12:15pm 7

The direct quote from the original blog-post is this:

Pre-merge, this took 5.9 billion uncached input tokens, 690 million output tokens, and 72 billion cached input token reads — around $165,000 at API pricing

API pricing is public, so unless they’re lying about the token counts, you’re free to calculate it yourself and disprove their claims.

I think their main argument that the time cost of such a transition (which would likely mean no user facing updates for a while) is a lot greater than the cash cost of it.

I am sad to see an emotional response from zig community’s leader. Personally, the impact of the article is negative. I feel really sad.

Like Zig, I am in favor of not hiding our own “runtime”. If we have emotion, we should declare it. If we think something is bad, we should declare why we do it. I appreciate less pretending not to attack someone personally, but doing it in an implicit way.

To me, that’s not very Zig.

Bobvan July 9, 2026, 1:21pm 10

I am not Andrew and i definetly dont intend to put words in his mouth.
From my read:

• Its his personal blog, i have no issues of emotions / opinions being expressed.
• While the article is absolutely insulting of Jarred, it does not critique his character / personality, it is negative on his engineering and managment decisions and overall approach.
• Expressing someones opinion often colides with PR-talk which we all grew accustomed to.

I agree that it could have been writen more PR friendly and without emotions, however its not a ZSF official response, its a single person sharing his mind.
I personally apreciate the more human direction this article went and even laughed at some parts.

alexs July 9, 2026, 1:25pm 11

Ever since watching the JetBrains interview with Andrew, I was excited to learn Zig and I’m taking my first steps now. His humbleness and especially his stance on engineering (“no dependencies for your core product”), AI and VC culture resonate deeply with me.

This is clearly a highly emotional (human) response, and the bluntness of it feels almost refreshing to me. However, I can’t shake the feeling that it represents a missed opportunity to highlight Zig’s advantages and “ideal customer profile” instead of bashing the Bun dude. I realize that it’s not a popularity contest and getting more folks to use Zig is not the goal, but IMHO positive sentiment almost always trumps negative.

I fully agree with that.

I can’t really express it and have thought long about how to do it so here’s a try:

It’s honestly very refreshing to read something more emotional especially about a topic like this. Because we are all human and have our emotions which we need to express and also should be able to as we want to. Of course adhering to law, not harassing others, and being respectful to others. Though, I think, one should be able to criticize and sometimes even needs to do it in a harsher, emotional tone to get the point across. It takes a lot of courage to write about their feelings as openly as Andrew did, though somewhat implicitly, and this, at least for me, isn’t negative at all, as others here and on other platforms are saying.

As usual, hearing this stuff straight from Andrew is a breath of fresh air.

It’s also great to get Andrew/ZSF’s perspective of the Bun codebase. Hopefully this will reduce the amount of times people ask what the Zig community should do (“should we fork Bun?”). He also makes a great point that, a large and high profile written in Zig is going to be a sort of ambassador for Zig, but Bun didn’t exactly value the same thing that Zig does.

hkupty July 9, 2026, 2:01pm 14

I think you hit the nail on the head, Andrew, on your analysis about the root cause. For years I’ve seen companies go haywire because VC money pulls the company into the weirdest directions: shrink understaffed areas, spend unnecessarily in office spacing, cancel projects without telling the teams working on them, changing priorities out of the blue and even demanding particular tech stacks to please investors. Last I’ve heard, and I’m not joking, is that AI-usage increases company valuation. So I’m not surprised at all that a good project got tainted by ill intended board “advisors”.
What I’m positively surprised about is the position of ZSF on large scale projects, maintaining an active relationship with proactive code reviews.
As for bun itself, I don’t think there’s much more to be said. Good riddance. A good football player that piles up fouls and cards is a liability, not an asset (I’m still in world cup spirit ).

pfech July 9, 2026, 2:15pm 15

After reading the blog post the weirdest thing is actually the ‘apology’ to Jarred at the end. I now know there was bad blood involved, this probably shouldn’t have been made public.

I’m sorry to say that I feel the first third of the article read as vindictive and unprofessional, and that the article could have a better impact without it. The technical case against slop-driven development is strong enough that the personal attacks aren’t necessary.

The conclusion stating that Andrew “actually [doesn’t] have any personal criticisms of Jarred” juxtaposed with the first third of the article reads like the first third was vented out before the rest of the article was composed. Perhaps it ought to have sat a day before editing.

I really like the joy, creativity, and technical and philosophical confidence that exudes from the Zig community. I am put off by the toxicity that that confidence occasionally brings out, and it saddens me because I think it distracts from the core and important technical and social ideas that the Zig project stands for. One can build a city on the hill without getting into the mud.

ProfBof July 9, 2026, 2:31pm 17

The endgame is that you will simply press a button to get a view of assembly code in your favorite “human programming language”, so you can understand and edit it. But only assembly language will be commited to source control. So people who prefer zig can see zig code, and people who like rust can see rust. In the end, only assembly code matters. I really like that zig has a firm no-AI policy, because it means that zig people keep control, can continue to make human calls to get a good human oriented programming language. But in the future, the “transpile in my language, idiomatically” button will definitely exist. For me this is good news, because I won’t have to look at c++ or rust anymore. I might be deep in a sad AI psychosis right now, so please disregard what I said if it makes you angry…

Yeah there’s no world in which you write a page about Stinky Jarred and then say “but no hard feelings.”

That said, I don’t actually care that this sounds abrasive or however you want to characterize it. Even if Jarred/Bun didn’t personally attack Andrew, they very clearly have willfully misrepresented the Zig project over and over again. Lying is unprofessional. It’s not better just because Jarred didn’t use Bad Words. Just because Jarred lies in mostly corporate-safe language doesn’t make it any less unprofessional. Fuck that.

Tekla July 9, 2026, 2:48pm 19

Happy to read what happened behind the curtains and sad to see it end this way. I will never understand the “rewrite it in Rust” approach as a solution for badly written code in any other language, but it sure seems like the main motivator for said behaviour lately. As if good Rust code were easy to write…

I also appreciate to see a more opinionated discourse shared in public this days, and not yet another LinkedIn PR-curated longtwit. While some appear to consider it harsh or dramatic, I think it’s perfectly mild.

But hey a more complete LSP sounds good

flow July 9, 2026, 3:39pm 20

Yes, Andrew should have used a LLM for his blog post, it would have filtered out the emotions. That’s a far better world … isn’t it?