Rewriting Bun in Rust

Original post: Rewriting Bun in Rust

Jarred Sumner, the author of Bun, rewrote it from Zig to Rust in 11 days. This task would have taken around a year for a small team, if AI weren’t around.

From my point of view, this is not another story about how LLMs will replace us all, and that’s why it’s worth sharing. Indeed, the strongest fact we can get out of it is crystal-clear: LLMs are incredibly powerful, but basically useless without an engineer who knows how to use them effectively.

Setting up the right environment, parallelizing the work, keeping agents from stepping on each other: all of this required rigorous engineering discipline and a person who knew what they were doing.

Beyond that, it also shows the use of coding agents at a scale most of us outside big tech aren’t used to yet. The rewrite involved 64 Claude instances across ~50 workflows and 4 worktrees, and it included different roles, like implementers, adversarial reviewers and fixers, all in different context windows.

The economic aspect is also interesting: this rewrite cost ~165k USD, which roughly equals a software engineer’s yearly salary in the USA. Yet no solo engineer could have completed this task in a single year, showing that these tools indeed can amplify what one person can do.