How a Solo Engineer Rewrote 530K Lines of Zig to Rust with Claude and Launched Bun 1.4

Author "智见君" describes a solo, 11‑day effort that used up to 64 concurrent Claude agents to rewrite more than 530 k lines of Zig code into Rust, culminating in the official release of Bun 1.4.

The migration was driven not by trend‑following but by recurring low‑level issues in the original Zig implementation: JavaScriptCore garbage‑collection objects, manual memory handling, and mixed C/C++ library boundaries caused memory leaks, use‑after‑free bugs, and crashes. Rust’s compile‑time checks could prevent many of these problems, yet the official estimate for a team‑size rewrite was a year.

Instead, founder Jarred Sumner chose an AI‑assisted path. He ran roughly 50 dynamic Claude workflows continuously for 11 days. At peak, four workflow shards each scheduled 16 Claude instances, handling file translation, compilation errors, test failures, and code reviews. Each implementation was fed to Claude for adversarial review in its specific context, and the feedback loop was used to iteratively fix the code. The process added over 1 M lines of production‑grade code, and the existing test suite passed before merging.

Before the public launch, the rewritten code had already been stress‑tested in production: Claude‑generated Rust Bun ran for months, and Prisma Compute deployed it in a live environment, effectively using the rewrite as a trial before promotion.

Performance Impact

CPU usage : Claude‑generated code’s p99 CPU dropped from 24% to 10% and p50 from 5.8% to 2.5%.

HTTP server memory : Fastify fell from 233 MB to 120 MB; Express from 169 MB to 92 MB.

Linux cold start : hello.js startup time fell from 10.9 ms to 5.1 ms, with peak memory decreasing from 33 MB to 14.6 MB.

Windows cold start : Time reduced from 39 ms to 15.5 ms (≈2.5× faster).

Binary size : Linux and Windows binaries shrank up to 17%; macOS binaries grew by ~1 MB.

CI build times also improved dramatically: the average of 12 build channels dropped from 20 minutes to 5 minutes 54 seconds, a 3.4× overall speed‑up, with Linux channels seeing roughly 5.4× gains.

Beyond Rust’s inherent advantages, Bun 1.4 also unified its memory allocator, reduced futex calls, and applied extensive optimizations to JavaScriptCore and runtime hotspots, contributing to the observed gains.

Compatibility Improvements

Bun 1.4 passed 1 517 Node.js test cases, the largest compatibility jump since Bun 1.0. Core modules such as node:http, node:fs, node:cluster, node:timers, node:zlib, node:vm, node:stream, node:events, node:trace_events, and node:sqlite now achieve 97‑100% test coverage.

Tooling support has expanded: Playwright runs directly on Bun, Vitest’s coverage and multi‑process mode are functional, Next.js 16.3 works with Turbopack and React Compiler, and integrations for OpenTelemetry, Datadog, TypeORM, Fastify, and Nuxt have been added. However, 100% Node.js compatibility is not yet reached; the remaining gaps affect legacy projects, native extensions, and complex networking scenarios.

New Runtime Capabilities

Bun.WebView

: Built‑in headless browser automation (navigation, clicks, scrolling, JS execution, screenshots) without needing Puppeteer or Playwright. Bun.Image: Native image decode, resize, rotate, and format conversion API similar to sharp, no native extensions required. Bun.markdown: Convert Markdown to HTML, React elements, or terminal text with GFM tables, task lists, and strikethrough support. Bun.cron(): Register OS‑level scheduled tasks or in‑process periodic jobs. Bun.Terminal: Embedded pseudo‑terminal enabling JavaScript‑driven shells, vim, htop, eliminating the need for node‑pty.

Parallel execution and testing via bun run --parallel and bun test --parallel, adding multi‑task, multi‑process CI sharding.

Package Manager Enhancements

A new global virtual store allows cache reuse across projects. In CI scenarios where node_modules is cleared, isolated installs become up to 7× faster. Commands such as bun audit fix, bun dedupe, and bun prune are now part of the toolbox.

Upgrade Considerations

Node.js ABI aligned to 26; native‑extension packages may need to rebuild against NODE_MODULE_VERSION 147.

New monorepo defaults to isolated linker; existing lockfiles retain hoisted layout.

Bun no longer auto‑loads .env when run as node; use --env-file explicitly. Bun.YAML now parses according to YAML 1.2, treating yes, no, on, off as strings. bun.lock upgraded to version 2, tightening package integrity and Git dependency path validation.

Upgrade command remains simple: bun upgrade Overall, Bun 1.4 demonstrates that a massive AI‑assisted rewrite can deliver a lighter, faster runtime with substantial compatibility gains, while also expanding the standard library to cover browser automation, image processing, markdown rendering, scheduling, terminal emulation, and parallel execution.