Bun Publishes Rust Migration Guide While Zig’s No‑AI Policy Sparks Debate

Bun is a JavaScript runtime and toolchain written in Zig that can directly execute .ts and .tsx files like vanilla JavaScript, potentially removing the need for a separate TypeScript transpilation step. It also bundles Babel‑ and tsc‑like capabilities to convert TypeScript to JavaScript for development and debugging.

Zig enforces a strict “no AI” policy: any AI‑related contributions, pull‑request comments, or bug‑tracker discussions are rejected. This stance conflicts with Anthropic, which acquired Bun in late 2025 and intends to use it for Claude Code, creating tension between the language’s governance and the project’s new ownership.

The Bun team has forked Zig, claiming that using LLVM for parallel code generation on macOS and Linux cuts compile time by a factor of four. However, a core Zig maintainer warned that changes in Bun’s Zig branch will not be accepted upstream because the parallel semantic analysis may exhibit nondeterministic behavior and splitting the LLVM backend into multiple modules is seen as a waste of effort.

Jarred Sumner, Bun’s creator, posted a Zig‑to‑Rust migration guide on GitHub. The guide outlines “Phase A,” which captures the program’s logic even if the Rust code does not compile, and “Phase B,” which enables crate‑by‑crate compilation. Sumner emphasizes that there is no commitment to a full rewrite; he is merely curious about what a runnable Rust version would look like and how it would perform.

Community feedback is mixed: some users call building a product on a still‑experimental language “crazy,” while others praise Bun’s speed and flexibility but note persistent bugs and memory‑leak issues. Observers point out that if Bun continues the migration, AI will likely be heavily leveraged, citing examples such as Cloudflare’s one‑week AI‑driven reimplementation of most Next.js API features and Ladybird’s two‑week Rust rewrite of its JavaScript engine.

Sumner predicts that open‑source development will trend toward disallowing human contributions, with AI handling code generation, testing, and PR processing, while humans continue to discuss priorities and high‑level design.