These days, the tech world is buzzing: Andrew Kelley, the founder of the programming language Zig, is furious.
The reason is that Bun, which originally relied entirely on the Zig language, was rewritten in Rust by its founder, Jarred Sumner.
Andrew Kelley openly expressed his anger, bypassing any polite formalities surrounding this phenomenon-level technological event and directly targeting Jarred Sumner’s personal engineering habits, management skills, and the business logic behind this incident.
Bun is a high-performance JavaScript/TypeScript runtime designed as a faster, modern direct replacement for Node.js, and in recent years has emerged as a major contender challenging Node.js's dominance in the frontend world.
Bun's key advantage is its blazing speed: whether it's startup time, dependency installation, or test execution, it far outperforms competitors, partly because it's written in the Zig language.
In December last year, Anthropic announced the acquisition of Bun to serve as the infrastructure powering its AI programming tools, Claude Code and Claude Agent SDK. Jarred Sumner and other members of the Bun team are now working at Anthropic.
After large-scale adoption, particularly as the underlying engine for Claude Code, the Bun team encountered persistent stability issues that were difficult to resolve.
Specifically, the Zig version of Bun contains numerous memory safety bugs—use-after-free, double-free, and forgotten memory releases along error paths. In Zig, these issues can only be prevented by coding conventions, whereas in Rust, they are directly caught as compilation errors by the borrow checker and Drop mechanism.
On the other hand, the Zig upstream community has a zero-tolerance policy toward code generated by large language models (LLMs), even rejecting optimizations unrelated to AI. The Bun team heavily relies on AI-assisted development; continuing to use Zig would require them to maintain their own compiler fork indefinitely, at a high cost.
In May this year, we witnessed a major engineering feat in the tech world: Jarred Sumner, the founder of Bun, announced that he would rewrite Bun’s million lines of code in Rust over just 11 days, using Anthropic’s then-unreleased Claude Fable 5 (Mythos-level model) and the dynamic workflow capabilities of Claude Code.
This was an epic agentic workflow test, later promoted by Anthropic as a flagship example of Dynamic Workflows, but it also sparked controversy due to "betraying its principles."
Zig's founder, Andrew Kelley, pointed out in a recent blog post that Bun's frequent bugs before its rewrite were fundamentally due to Jarred Sumner's poor engineering practices.
Even before AI became popular, Jarred had been writing messy code. Kelley said the Zig team frequently reviewed users’ codebases and felt “extremely alarmed” by Bun’s codebase, which was filled with layer upon layer of hacky patches, abused assertions, and almost no time spent fixing bugs or paying down technical debt—all in the rush to ship new features quickly.
Then there’s the million lines of code generated by Claude. Kelley retorted: “Bun’s official claim is that 1 million lines of unreviewed Rust code (written by AI) are safe because they have test cases; but if the test cases are truly comprehensive, why didn’t they catch those annoying bugs when the code was originally written in Zig?”
Now, Kelley is deeply disappointed that Jarred has transformed from an open-source developer with beginner energy into a stinky manager.
Kelley frankly said that when he learned Bun had decided to abandon Zig, he felt not betrayal or anger, but relief. He feared that Bun’s association with Zig would mislead the public and attract a crowd of users who merely copy and paste AI-generated code. He even joked that he was sipping tea, relieved that “this is finally not my problem anymore.”
Such a direct attack prompted others in the tech community to join the debate and share their opinions.
First, let’s do the math: While many say Claude’s tokens are too expensive, according to data publicly released by Jarred Sumner and Bun, the Rust-language rewritten Bun project is estimated to have consumed $165,000 in API costs. In the tech and engineering communities, this price and timeline are frighteningly cheap.
On paper, AI has reduced development costs to about one-tenth of the original and shortened the timeline from roughly a year to less than two weeks.
Second is the attitude toward the collision between open-source community culture and the AI era. After reading Andrew’s blog post, the founder of Zig, some people felt deeply unsettled, believing that his public criticism of former key users and sponsors—Bun, which had long funded Zig—demonstrated a lack of professional conduct; some even went so far as to say, “I’ve never so actively wished for a programming language to fail.”
However, some veteran programmers have come forward in support, arguing that in an era dominated by capital and AI bubbles, Andrew is simply defending pure engineering quality, embodying the spirit of Linus Torvalds of old.
Of course, what people are most concerned about is whether the project is still easy to use after all these changes.
The most contentious issue is that these one million lines of code, being directly mechanically translated from Zig by AI, lack architectural refactoring by human engineers. The new codebase still contains as many as 27,000 lines of unsafe code blocks. Many are concerned that the cognitive and debugging costs for human developers maintaining, reading, and modifying this massive "AI-generated" codebase in the future may ultimately exceed the upfront development costs saved today.
This project, which defies the historical patterns of software engineering, will either become a milestone in AI transforming programming paradigms—or turn into a volatile mountain of technical debt that’s impossible to maintain. Only time will tell.
Reference content:
https://bun.com/blog/bun-in-rust
https://andrewkelley.me/post/my-thoughts-bun-rust-rewrite.html
This article is from the WeChat public account "Machine Heart" (ID: almosthuman2014), authored by someone interested in AI.