Bun, the JavaScript and TypeScript runtime, has rewritten hundreds of thousands of lines from Zig into Rust in an AI-assisted project that the company says fixed 128 longstanding memory-safety bugs. The result is a striking case study in using coding agents for a task usually associated with years of cautious engineering, not a long weekend and a very large API bill.
Watch Desk analysis
What happened
According to InfoQ, Bun creator Jarred Sumner said the rewrite moved between 535,000 and 1 million lines of code from Zig to Rust over a period reported as ranging from 11 days to four months. The project used 64 parallel Claude instances and about $165,000 in API credits.
The reported technical results are substantial. Bun says the migration eliminated 128 longstanding memory-safety vulnerabilities and various memory leaks, reduced peak bundling memory use by 90% and cut binary size by 20%. The figures and the account come through the InfoQ report, which also records sharp criticism from Zig creator Andrew Kelley over the methodology.
Why it matters
This is the useful version of the AI coding debate. Not whether a chatbot can produce a tidy function, but whether a carefully structured fleet of agents can help move a large, real codebase between systems languages while producing measurable engineering gains.
It also shows where the hard questions begin. A rewrite can remove one class of memory-safety problem while introducing behavioural regressions, performance surprises or less visible maintenance costs. “The tests pass” is comforting, but it is not the same as “the migration was sound”. Software has a rich tradition of passing tests while preparing a small administrative coup.
Our read
The reported results make this worth taking seriously as a production experiment, not merely a dramatic coding demo. Rust’s memory-safety guarantees are the obvious attraction, while the reported reductions in memory use and binary size give the project a practical payoff beyond fashionable language migration.
But the methodology matters as much as the headline numbers. Readers should treat the claims as an attributed account from Bun, with criticism from Andrew Kelley, rather than as an independently established benchmark. The next useful evidence is reproducible detail: the test suite, regression record, review process and the parts of the rewrite that still required conventional human engineering.
What to watch
- Whether Bun publishes a detailed comparison of the old and new implementations.
- How many regressions, rejected changes and manual interventions the AI-assisted process produced.
- Whether the reported memory and binary-size gains survive wider workloads.
- Whether other large systems projects attempt similarly broad AI-assisted rewrites, or decide that one Bun is quite enough for now.
Discussion spark: Would you trust an AI-assisted rewrite of a large production codebase if the test suite, memory figures and security results were public, or should migrations of this scale still require predominantly human-written code?
Sources and evidence
- Bun Rewrites 535K Lines of Zig into Rust in Four Months, Eliminates Numerous Memory Leaks (21 September 2026, 17:03 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.