Rewriting Bun in Rust

2026-07-08 · Source: Simon Willison's Weblog · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Advanced, quick

Summary

Jarred Sumner successfully rewrote the Bun JavaScript runtime from Zig to Rust, a project he had been discussing since May 9th. This ambitious undertaking was primarily motivated by persistent memory management bugs, such as use-after-free and double-free errors, which Rust's inherent safety features and RAII-like automatic cleanup mechanisms effectively mitigate. A key enabler for this massive rewrite was an "agent harness" that leveraged Bun's existing TypeScript test suite as a conformance suite. This harness automated a significant portion of the initial porting process using an earlier version of the Claude LLM, requiring 5.9 billion uncached input tokens and costing an estimated \$165,000 at API pricing. The new Rust implementation, live in Claude Code since June 17th (v2.1.181), demonstrated a 10% faster startup on Linux, validating the approach and showcasing how frontier models can facilitate large-scale software rewrites.

Key takeaway

For AI Architects evaluating large-scale system rewrites, this case demonstrates that modern frontier models fundamentally alter the feasibility equation. You should now consider LLM-powered agent harnesses, especially when a comprehensive, language-independent test suite exists, to automate significant portions of the porting effort. This approach can mitigate the traditional risks and costs associated with ground-up rewrites, enabling projects previously deemed impractical.

Key insights

LLM-powered agent harnesses can enable large-scale software rewrites, overcoming traditional barriers to such projects.

Principles

• Language choice is no longer a one-way decision.
• Robust test suites enable agentic code generation.
• Fix the generation process, not just the code.

Method

Automate initial code porting using an LLM-powered agent harness with a language-independent conformance test suite, then iteratively refine the generation process.

In practice

• Use Rust for memory-safe system-level programming.
• Develop comprehensive test suites for agentic development.
• Consider LLMs for large-scale code refactoring.

Topics

• Bun JavaScript Runtime
• Rust Programming Language
• LLM Code Generation
• Agentic Engineering
• Software Rewrites
• Memory Safety

Best for: CTO, VP of Engineering/Data, Director of AI/ML, Software Engineer, AI Engineer, AI Architect

Related on AIssential

• Why Rust is the Ideal Language for Vibe-Coding — Daniel Szoke, Sentry — Rust's strict compiler constraints, though challenging for LLMs initially, enhance production code safety by preventing common errors.
• Building AI Agents in Rust — part 3 — Using traits and a registry simplifies AI agent tool dispatch and schema management in Rust.
• When Tools Fail: Benchmarking Dynamic Replanning and Anomaly Recovery in LLM Agents — LLM agents struggle with dynamic replanning and error recovery when tools fail, a problem unaddressed by model scaling.
• Harnessing Code Agents for Automatic Software Verification — A general LLM code agent with a verification harness can fully automate formal software proof generation.
• A Beginner’s Guide to Amazon Bedrock: Your First LLM App Without the Overwhelm — Amazon Bedrock simplifies LLM app development by offering a managed API gateway to various models, abstracting infrastructure.
• Vibe coding is old now — AI agents are evolving rapidly, enabling non-coders to build software and raising critical security and usability questions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.