Bun Replaces Zig with Unsafe Rust in AI-Generated Port, Sparking Systems Engineering Controversy

Bun has migrated its TypeScript runtime codebase from Zig to unsafe Rust using heavily automated AI agents under its parent company, Anthropic. The transition has triggered a sharp debate between Zig creator Andrew Kelley and Anthropic over whether the rewrite was necessitated by Zig's memory-safety limitations or driven by poor engineering practices and corporate marketing.

Technical Execution of the Bun Migration

Bun, a high-performance TypeScript runtime originally built in Zig, has transitioned its codebase to unsafe Rust. The migration was merged to the mainline branch two months prior to any public engineering explanation. Technically, the port was executed as a literal, file-by-file migration to unsafe Rust, a strategy designed to minimize transition risk while establishing a baseline for future architectural redesigns.

This approach allowed the development team to bypass the immediate safety constraints of idiomatic Rust to achieve parity with the original Zig implementation. While language transitions of this scale are typically preceded by detailed RFCs or architectural proposals, the post-merge explanation left systems engineers to analyze the structural changes after the fact.

The Ideological Divide on AI Contributions

The transition highlights a fundamental ideological split regarding the role of machine learning in systems programming. Zig maintains a strict policy of allowing 0% AI contributions in its ecosystem. Conversely, Bun—which was acquired by AI safety and research lab Anthropic—claims its new codebase consists of near 100% AI-generated contributions.

Anthropic, currently highly valued and approaching a potential IPO, has actively promoted a narrative that software engineering and manual coding are transitioning toward obsolescence. The migration was heavily automated using agentic AI workflows, specifically utilizing Anthropic’s models and Claude Code. However, critics point out that despite the marketing narrative of AI-driven efficiency, the automated tools failed to catch basic systems-level bugs, including a use-after-free vulnerability, during the translation process.

Conflicting Rationales: Language Limitations vs. Engineering Debt

The systems engineering community is divided on the true catalyst for the rewrite, with three distinct interpretations emerging:

• The Anthropic and Bun Narrative: The development team asserts that Bun’s engineering resources were overwhelmed by persistent memory bugs, concluding that Zig's language features were insufficient to maintain stability at scale.
• The Zig Creator Narrative: Andrew Kelley, creator of Zig, publicly criticized Bun's engineering leadership. He attributed the high defect rate to systematic design flaws, lack of structured code reviews, and an over-reliance on AI agents to write and review low-level systems code.
• The Corporate Marketing Narrative: Industry analysts suggest that while memory bugs were a legitimate technical challenge, the decision to execute a complete Rust rewrite was heavily influenced by corporate alignment. Anthropic's internal infrastructure relies on Rust, Zig openly rejects Anthropic's products, and the rewrite offered a highly visible marketing demonstration for Anthropic's Fable model.

Culture, Management, and Systemic Outcomes

The engineering practices behind Bun have historically favored high-intensity development. Founder Jarred Sumner previously warned recruits that working at Bun's parent entity, Oven, would require a relentless work ethic, explicitly stating the environment was not a fit for those prioritizing work-life balance.

Kelley's critique highlighted reports of poor communication, unrealistic expectations, and low empathy within the team, characterizing the project management as highly unstable. While high-crunch environments occasionally produce high-performance tools, systems engineers note that running continuously at crunch time systematically introduces architectural defects. The resulting instability is often misattributed to language-level deficiencies rather than organizational and process-driven debt.