Bun's Rust Rewrite: Fast Build, Slow Ship, Big AI Bill
28 Jul 2026
The 11-day rewrite that's now 6 weeks from shipping
On July 8, 2026, Jarred Summner posted about "Rewriting Bun in Rust" — a project that, according to the timeline, was actually completed between May 3 and May 14, 2026. That's a full JavaScript runtime rewritten in under two weeks, with Anthropic employees directly involved in the effort.
The catch: as of July 27, 2026, Bun hasn't cut a new release tag in six weeks. The last one, bun-v1.3.14, dates back to May 12 — right around when the rewrite itself wrapped up.
What the numbers show
The rewrite's speed came at a real cost. Anthropic API usage for the project totaled $165,000 over 11 days, or roughly $15,000 per day. That's a meaningful operating expense for an AI-assisted migration of this scale, even before accounting for the human engineering time layered on top.
Meanwhile, the pull request backlog has grown sharply. On July 9, 2026, there were 1,277 open PRs from robobun (an automated contribution source tied to the project). By July 27, that number had nearly doubled to 2,475. The report doesn't detail what's inside these PRs or why the count spiked so fast — that's flagged as missing context.
On the infrastructure side, merging to main with Buildkite checks currently takes about 40 minutes, and sometimes up to 1.5 hours. As PR volume climbs, that merge time is a plausible early signal of CI/CD strain.
Timeline at a glance
- May 3–14, 2026: Bun rewritten in Rust
- May 12, 2026: Last release tag (bun-v1.3.14) published
- July 8, 2026: Public post about the Rust rewrite
- July 9, 2026: 1,277 open PRs from robobun
- July 27, 2026: 2,475 open PRs; still no release tag six weeks post-merge
The open questions
The report is explicit about what's unknown here, and it's a substantial list: there's no stated reason why no release has shipped since the merge, no detail on what's driving the PR backlog growth, no before/after performance comparison for the rewrite, no specifics on Anthropic's exact role, and no way to benchmark the $165,000 API spend against what a traditional engineering effort of similar scope would have cost. There's also no clarity on the current commercial relationship between Anthropic and the Bun project.
Worth noting: this isn't happening in isolation. The report mentions parallel efforts — Anthropic reportedly has a C compiler project, and Cursor has a project called FastRender, a web browser — suggesting AI-driven rewrites of foundational software tooling may be a broader pattern, not a one-off.
Why founders should care
If you're evaluating AI-assisted rewrites or migrations for your own stack, Bun's experience offers a few probabilistic signals worth weighing:
- Speed at the front, friction at the back. An 11-day rewrite completing and then sitting six weeks without a release suggests AI tooling may compress initial development dramatically while integration, testing, and release readiness likely remain bottlenecks — meaning founders should probably not assume a fast build equals a fast ship.
- Budget for compute, not just code. At $15,000/day, the API cost of an AI-assisted rewrite could plausibly rival or exceed traditional engineering costs for similar-scope work, so founders considering this path should likely model compute spend explicitly rather than treating AI assistance as a cost-saving shortcut by default.
- Review infrastructure may need to scale in parallel. A PR backlog nearly doubling in under three weeks hints that generating code with AI faster than a team can review and merge it may be a common failure mode — CI/CD and review capacity likely need proactive investment alongside any AI-driven contribution surge.
- Vendor involvement may deepen. With Anthropic employees reportedly embedded directly in Bun's development, founders working with AI vendors might increasingly encounter — and should think through — what it means when a vendor's team has hands-on involvement in core engineering decisions, not just API access.
None of this confirms Bun's rewrite is stalled for negative reasons — the report offers no explanation either way. But the gap between a headline-grabbing 11-day rewrite and a six-week release drought is a useful case study for any founder weighing similar AI-driven infrastructure bets.