Six Weeks After the Bun Rust Rewrite: Is It Done Yet?

On July 8, Jarred Sumner published Rewriting Bun in Rust -- the story of porting 535,496 lines of Zig to Rust in 11 days using Claude, at a cost of $165,000 in API calls. The post went viral on Hacker News (641 points, 377 comments). Developers debated whether this was the future of code migration or a case study in survivorship bias.

Six weeks later, Tom Lockwood wanted to know: how is that rewrite actually going?

What Lockwood Found#

Lockwood's investigation (July 27, 2026) digs through commit data, PR counts, and release history. His findings challenge the clean narrative:

No release in six weeks. The last Bun release tag is bun-v1.3.14 from May 12, 2026. The only comparable gap was between October and December 2022, between v0.2.2 and v0.3.0. For a project that historically shipped monthly, this silence stands out.

2,475 open PRs from robobun. The Claude Code bot that drove the rewrite has 2,475 open pull requests as of July 27. At roughly 40 minutes per Buildkite pipeline run (sometimes up to 90 minutes), merging all of them would take an estimated 86 days of continuous CI time. Lockwood notes that some PRs show signs of extensive human review, suggesting the AI output requires non-trivial human oversight.

Hidden costs. The headline $165K covers only Anthropic API tokens. Lockwood estimates that when you factor in Buildkite CI/CD costs and continued Claude usage at roughly $10K/day, the real expense is approaching $800K. Anthropic employees are also directly contributing Rust commits, which is relevant context that the original announcement did not foreground.

Two graphs worth studying. Lockwood's commit analysis shows a sharp spike in Claude-authored Rust files during the rewrite period (May 3-14), followed by a sustained plateau of Anthropic employee and robobun activity. The machine is still running.

What HN Is Saying#

The Hacker News discussion on Lockwood's post (422 points, 316 comments) is notably more measured than the original Bun rewrite thread.

simonw raised the strongest counterpoint: Lockwood's article omits the fact that "Bun-on-Rust has been live in Claude Code itself since June 17th." A Rust-powered Bun is already shipping to millions of users through Claude Code. Whether or not there is a public release tag, the software is running in production. Simon also noted the "unsafe" count in the Rust codebase has stayed constant rather than declining -- worth watching but not a crisis.

Jarred Sumner himself responded directly. He confirmed the Rust rewrite "is going well overall" and explained the release delay: "In the Bun v1.4 video, I promised a certain number of newly passing Node.js tests were added to force us to improve compatibility, and that number is not true yet. The release is delayed until it is true." He expects the release "most likely next Tuesday."

abalashov captured the ambivalence many feel: "I did suspect that the triumphalist pronouncements, and even the seemingly honest and forthright deep dive, were a little premature." The concern is that LLM-assisted rewrites look done much earlier than they actually are, because the first pass handles the easy 80% and the remaining 20% (edge cases, correctness, CI stability) takes longer than expected.

reliabilityguy pushed back on the cost comparison: "$165K is cheaper than a team of multiple engineers working on the rewrite for a year" is a flawed argument, they argue, because "the team of engineers would have produced idiomatic Rust, and it would take probably 100K+ of tokens more to make the bun in Rust idiomatic Rust." In other words, the cost savings may come with a quality tax.

losvedir offered the most pragmatic take: "I'm not sure Anthropic even cares about releasing the next version. The Rust one has been in use in Claude Code for more than a month now, used by millions of people." The open-source project's release cadence may simply not be Anthropic's priority.

What This Really Tells Us#

The Bun Rust rewrite is real. It is running in production in Claude Code. Jarred expects a public release soon. The technical achievement of porting 535K lines in 11 days is not in dispute.

What Lockwood's investigation reveals, and what the HN discussion sharpens, is that "done" means different things depending on who you ask:

• Done enough to ship internally. The Rust Bun runs Claude Code. That is real value. It does not need a v1.4 tag to be useful.
• Not done enough to release. Public release requires passing a specific Node.js compatibility bar that the team set publicly. Open-source users waiting for the tag are still waiting.
• Not done in cost terms. If the real cost is approaching $800K including CI and continued agent usage, the economics look different than the $165K headline. Whether $800K is expensive or cheap depends on what you compare it to -- a year of senior engineer salaries at Anthropic is multiples of that.

For developers evaluating whether AI-assisted rewrites make sense for their own projects, this case study offers a more honest template: the initial AI pass is fast and cheap, but the tail of correctness verification, human review, and CI churn is long and not free. Our earlier deep dive on the agent orchestration behind this rewrite covered the workflow architecture in detail. The economics of AI rewrites is the broader context for thinking about whether this pattern generalizes.

What to Watch Next#

Two things to track:

1. The v1.4 release. If it ships as Jarred promised "next Tuesday" (likely this week given the post date), the release gap becomes a six-week delay on a massive rewrite -- entirely reasonable. If it slips further, skeptics have a stronger case.
2. The open PR backlog. Whether the 2,475 open robobun PRs get merged, closed, or abandoned will tell us a lot about the sustainability of AI-first contribution models. Our analysis of over-editing in AI codebases is directly relevant here.

The rewrite that was a proof-of-concept is now a production system with a messy, visible, human-scale cleanup phase. That is not a failure. It is what real engineering looks like.

Sources#

Continue Reading#

• Bun Rewrites 535K Lines of Zig to Rust in 11 Days Using Claude -- The original announcement that started the conversation
• How Bun Coordinated 64 Concurrent Claude Agents to Port 535K Lines -- The workflow architecture behind the rewrite
• AI Rewrite Economics: When Codebase Migration Pays Off -- Broader framework for evaluating AI-assisted rewrites
• pgrust: Postgres Rewritten in Rust Passes 100% of Tests -- Another open-source Rust rewrite story, different outcomes
• Over-Editing: When AI Rewrites What Isnt Broken -- The hidden cost of AI-generated churn in codebases