<h2>What happened</h2>
<p>Bun 1.3.14 shipped as the last Zig-based release. Six days later, a pull request adding over one million lines of Rust landed on the main branch. The port was written almost entirely by Anthropic’s Claude Code, guided by a <code>porting.md</code> document containing roughly 300 rules for AI agents to follow — a line-by-line translation from Zig rather than an idiomatic Rust rewrite.</p>
<p>A second PR removing 600,000+ lines of Zig was automatically flagged by GitHub as “AI slop” and closed, which is a funny footnote to an already unusual story.</p>
<h2>Why Rust, why now</h2>
<p>Two things pushed this along. First, Zig is still pre-1.0 and has been breaking things with each release — the Bun team had accumulated real pain from that. Second, and more directly: Zig’s project has a strict no-AI contribution policy, which conflicts with how Anthropic (which acquired Bun’s parent company Oven in late 2025) wants to work. An AI-first team maintaining a codebase in a language that bans AI tooling is an awkward fit.</p>
<p>Jarred Sumner had committed a Zig-to-Rust porting guide to GitHub before the PR landed, which sparked speculation. He described it as exploratory at the time. Then it shipped.</p>
<h2>The results so far</h2>
<p>The port passes Bun’s full test suite across all platforms. Binary size drops 3–8 MB. Benchmarks are neutral to faster — the Three.js bundling benchmark (10 runs with minification and sourcemaps) came in 10% faster than the Zig version. Several memory leaks are fixed, and the team expects the rewrite to close around 200 open GitHub issues. One entertaining test failure: the TOML and YAML parser stack overflow tests failed because the Rust version could recurse deeper than the tests expected.</p>
<p>Sumner’s take: “Bun in Rust is better than the original, and it’s going to keep getting better.”</p>
<h2>The legitimate concerns</h2>
<p>The Rust code has roughly 10,000 <code>unsafe</code> blocks. That’s not surprising given it’s a direct port from Zig with FFI bridges to C++ libraries, but it does mean the memory safety story is more complicated than “we switched to Rust, problem solved.”</p>
<p>The deeper concern is that no human has fully read this codebase. It was written by Claude and reviewed by Claude. Tests validate known paths, but they don’t catch subtle cross-function invariants, edge cases nobody thought to test, or concurrency bugs that only surface under specific conditions. Passing a test suite is necessary but not sufficient — and a million lines of unread code is a lot of surface area for surprises.</p>
<p>This isn’t an argument that the rewrite was wrong. It’s just an honest accounting of what “AI-assisted port” means in practice: you get speed and you trade some depth of understanding.</p>
<h2>The yt-dlp fallout</h2>
<p>yt-dlp has already responded by deprecating Bun support. Supported versions are now capped at 1.2.11–1.3.14 — the last Zig-based release. Their stated reasons: versions below 1.2.11 ignore the EJS lockfile (a real security issue given recent npm supply chain attacks), and the AI-assisted Rust rewrite introduces compatibility and reliability uncertainty they don’t want to absorb. The announcement sparked debate in their issue tracker, but the decision stands for now.</p>
<h2>The broader point</h2>
<p>What’s actually interesting here isn’t the Zig-vs-Rust debate. It’s that a team ported one of the largest Zig codebases ever written to a different language in less than a week. Language choice used to feel permanent — you picked your stack and lived with it. If AI-assisted migration can compress a multi-year rewrite into days, that changes the calculus. Teams can experiment more, reverse decisions, or chase compiler guarantees they couldn’t afford to pursue before.</p>
<p>Whether that’s good depends entirely on what happens next. The real question is whether AI-generated, minimally-reviewed code can be maintained and debugged in production over years. Bun is the test case.</p>


Collections

Bun 1.3.14 was the last Zig-based release. Six days after opening a branch, a Claude Code-generated pull request adding over one million lines of Rust landed on main, effectively porting the entire runtime. The port was guided by a 300-rule porting document and passes Bun's full test suite, with binary size reductions of 3–8 MB and a 10% speedup on the Three.js bundling benchmark. However, the codebase contains roughly 10,000 unsafe blocks and has not been fully read by any human. yt-dlp has already deprecated Bun support beyond 1.3.14 citing reliability uncertainty. The broader implication: AI-assisted migration may compress what used to be multi-year rewrites into days, but the long-term maintainability of minimally-reviewed AI-generated code remains an open question.

Bun is being rewritten in Rust, moving away from Zig

<p>Makes sense given Bun was purchased by one of the biggest AI companies. The initial merge was filled to the brim with AI slop. I wonder how good the project will be when the migration is over.</p>


<p>Claude, rewrite Bun in Rust. Make no mistakes.</p>


<p>Last 5 years world is change :) Who is the imagine this :)</p>
<p><img src="https://media.daily.dev/image/upload/s--09DQHJgR--/f_auto/v1778828554/ugc/content_4014bd57-1e98-43ed-92c1-4545a7504db8?_a=BAMAMiWQ0" alt="Screenshot 2026-05-15 at 10.02.30.png"></p>


<p>Actually, this is just Sumner’s branch - <a href="https://x.com/jarredsumner/status/2051595933704761618/photo/1" target="_blank" rel="noopener nofollow">https://x.com/jarredsumner/status/2051595933704761618/photo/1 :D</a></p>


<p>Makes sense, but let’s wait for the results to arrive</p>
