DEVCLASS

Bun creator Jarred Sumner rewrote Bun's ~500,000 lines of Zig code into Rust in just 11 days using ~50 parallel Claude Code agents, at a cost of roughly $165,000. The AI-generated Rust code passed Bun's full test suite of over one million assertions. Zig creator Andrew Kelley publicly criticized the move, calling the AI-generated code 'unreviewed slop' and arguing that Bun's bugs stemmed from poor programming practices rather than Zig's limitations. Kelley also noted that Zig's project policy rejects AI-generated contributions due to quality concerns. HashiCorp co-founder Mitchell Hashimoto praised the speed of the AI-assisted rewrite. The controversy raises broader questions about code quality, engineering oversight, and the viability of large-scale AI-driven rewrites.

Zig creator calls Bun's Claude Rust rewrite 'unreviewed slop'

I have to agree with Kelley on this. No test has 100% coverage and no human oversight on the code should be concerning to those who use Bun.

Both sides have a point—shipping in 11 days is wild, but readability and long-term maintenance matter just as much.

We waste so much time on a) “If I didn’t do it my way it is rubbish”. b) Overuse of the words “AI Slop”, when we mean, AI did it 80% correct in 1% of the time it took me. Now I can spend some time improving the slop and adding new capabilities faster.

Passing the old test suite is the wrong thing to feel safe about, since a 500k-line port is only as reviewed as those tests are complete, and legacy suites encode plenty of behavior nobody re-checks. The “slop” claim is really about untested edge coverage, not Rust or the agents. Would a differential run against the old Zig binary settle it faster than arguing about review?