Six weeks have passed since Bun merged the most ambitious AI-assisted rewrite in open-source history — 535,000 lines of Zig ported to Rust in 11 days using 64 Claude agents at a cost of $165,000. As of today, July 27, 2026, Bun has gone 11 consecutive weeks without a stable release, the longest gap since the project launched. Tom Lockwood’s analysis of why is trending on Hacker News right now with 260 points and 175 comments, and it contains a number that complicates the “completed in 11 days” story: 2,475. That’s how many PRs from “robobun” — the Claude agent behind the Bun Rust rewrite — are still sitting in the queue, waiting for CI/CD to process them.
The Backlog Nobody Talked About
The $165,000 API cost and 11-day timeline were the headline. What Jarred Sumner didn’t lead with: robobun didn’t stop after the merge. On July 9, there were 1,277 open PRs from the Claude agent. By July 27, that number nearly doubled to 2,475. At the current CI/CD throughput of 40-90 minutes per merge, Lockwood’s analysis projects approximately 86 continuous pipeline-days to clear the queue. The rewrite isn’t a completed sprint. It’s an ongoing operation.
This is why there’s no Bun v1.4.0 stable release yet. The Rust code is in the codebase. The canary builds exist. However, the codebase is still evolving at machine speed, and shipping a stable release while a Claude agent is actively submitting thousands of PRs would be like freezing a river mid-current. The honest framing: the AI completed the port, then kept going. That’s not a failure — it’s just not what the announcement implied.
Prisma’s Data Settles the “Slop” Debate
On July 14, Zig creator Andrew Kelley called the rewrite “unreviewed slop,” arguing that no human fully read the million lines of machine-generated Rust and that diverging value systems — speed vs. quality — made the result suspect. It’s a fair concern. However, Prisma published production data that makes the abstract debate concrete. Their engineering team deployed the Rust canary against the same failure modes they’d been hitting in stable Bun: an S3 memory leak that pushed stable Bun past 900 MB peaked at just 118 MB on the canary — a 7.5x reduction. A SQL connection pool that hung permanently after scale-to-zero operations on stable Bun recovered correctly on the canary.
According to Prisma’s production report, “The specific failure we kept hitting in Compute went away on the canary.” That’s not a benchmark — it’s a real production workload. Kelley is right that no human reviewed all million lines. He’s wrong if the conclusion is that unreviewed therefore unreliable. The Rust compiler and a comprehensive test suite caught what manual review would have. Production data from Prisma cuts against the strongest version of his critique.
Related: Bun Rewrote 535K Lines from Zig to Rust Using Claude — the original announcement breakdown
Claude Code Already Proved It in Production
The framing of “canary only, not for production” is technically accurate for the public release. In practice, it understates what’s already deployed. Simon Willison inspected the Claude Code binary on July 19 and found it embedding “Bun v1.4.0 (macOS arm64)” with Rust source file paths like src/runtime/bake/dev_server/mod.rs. Claude Code has millions of installs. Rust Bun is already in production at scale — it just didn’t need a press release to get there. Willison’s verdict: “Boring is good. The rewrite is largely invisible to users which means stability.” Invisible is exactly what a successful runtime rewrite looks like.
Moreover, Anthropic shipping their own flagship product on the Rust canary is a meaningful signal. When the team that built the AI that wrote the code stakes their own tool on it, “production-ready” becomes less of an open question and more of a demonstrated fact.
What Developers Should Do Now
The stable v1.4.0 has no announced release date. The 2,475-PR backlog makes a near-term timeline uncertain. If your workload resembles Prisma’s — memory-intensive S3 operations, scale-to-zero databases, high connection churn — the canary is worth testing now. For everyone else, the risk/benefit math still favors waiting for stable. The 19 documented regressions from the initial merge were all fixed, but an actively-evolving AI-generated codebase carries inherent unpredictability. Test on staging before touching production.
# Try the Rust-based canary build
bun upgrade --canary
# Verify you're running the Rust build
bun --version # Should show v1.4.0-canary.X
# Roll back to stable anytime
bun upgrade --stable
Key Takeaways
- Bun’s Rust rewrite is not finished — 2,475 pending Claude PRs and an estimated 86-day CI/CD backlog mean the AI is still actively evolving the codebase
- Prisma’s production data is the most concrete counter to the “unreviewed slop” critique: 900 MB → 118 MB memory reduction and a fixed SQL hang are hard results to dismiss
- Claude Code already ships Rust Bun to millions of users — the “canary-only” framing understates how deployed this already is in the real world
- Test the canary if your workload matches Prisma’s profile; wait for stable v1.4.0 if it doesn’t
- The real lesson from the Bun Rust rewrite: AI can port a codebase in days, but codebase evolution continues for months
