Hacker News

Jarred Sumner

Jack Herrington

Last month, one engineer using Claude Code got more of Node.js’s test suite passing in Bun than the entire Bun team — including founder Jarred Sumner — managed across nine months of 1.2-to-1.3 development. Sumner posted this himself, which is either admirably honest or a very confident flex, depending on how you read it.
That’s the headline number, but the context makes it weirder. Bun’s entire JavaScript runtime was ported from Zig to Rust in 11 days in May 2025, using a multi-agent Claude Code setup: one main agent, adversarial review agents checking its work, and a fixer agent cleaning up the mess. Fifty dynamic workflows, multiple work trees, the whole thing.
The estimated token cost at API prices: around $165,000. Which is only viable because Bun is owned by Anthropic. That detail is doing a lot of work here — this isn’t a workflow you can replicate on a personal API key.
The resulting Rust code leans heavily on <code>unsafe</code>. That’s the honest consequence of mechanical translation from Zig rather than idiomatic Rust — the code works, but it’s not what a Rust developer would write from scratch. The team has been cleaning it up since, but the Zig creator publicly criticized the move, and the community drama followed predictably.
Here’s what’s actually interesting beneath the discourse: AI-assisted language-to-language porting of large codebases is a genuinely compelling use case, and probably the most defensible one in this whole conversation. If you have a solid test suite to validate correctness — and Bun does — you can mechanically translate a codebase and let the tests tell you when you’ve broken something. That’s a real unlock for legacy modernization projects that have been sitting in the “too expensive to touch” pile for years.
The one-engineer-beats-the-whole-team stat is the part that should make people sit with their feelings for a minute. It’s not that the Bun team was slow. It’s that the leverage ratio has shifted in a way that’s hard to reason about cleanly.

Trends

Jarred Sumner, Bun's creator, shared that a single engineer using Claude Code passed more of Node.js's test suite in one month than the entire Bun team did over nine months. The context: Bun's runtime was ported from Zig to Rust in 11 days using a multi-agent Claude Code setup with one main agent, adversarial review agents, a fixer agent, and 50 dynamic workflows. The estimated API cost was ~$165,000, only feasible because Bun is owned by Anthropic. The Rust output relied heavily on `unsafe` due to mechanical translation rather than idiomatic Rust, drawing criticism from the Zig creator. The broader takeaway is that language-to-language codebase porting is emerging as a strong AI use case — test suites provide objective verification, making it a cleaner problem than open-ended generation. The legacy modernization angle (C, COBOL, older C++) is particularly significant for enterprises with large untouchable codebases.

One engineer with Claude Code outpaced the whole Bun team. The numbers are hard to argue with.

I cannot wait to do the react/vite to react native AI translation for under 1000bucks. ;)

Wow, I had no idea Bun is owned by Anthropic.

Building is easy… making it work and maintaining is hard !

the cost number is a red herring honestly, what matters is they had node’s test suite as a spec. that’s the pattern I’d bet on for old COBOL/C migrations too, verification already exists so you’re not asking the model to invent correctness.

Now show me the cost without the test suite being available. I say that because public access to those are going bye-bye for future software.