AI rewrites systems code, blurring marketing and engineering
Anthropic's Fable model was used to rewrite parts of Bun from Zig to Rust, and two separate threads dissected what actually happened. The 'Rewriting Bun in Rust' post drew immediate skepticism: commenters noted that Bun powers Claude Code, so the rewrite doubled as an Anthropic product demo. The follow-up post 'My Thoughts on the Bun Rust Rewrite' went further, with the Zig community arguing that Bun was never a good representative of Zig in the first place, and that the framing of the rewrite obscured poor engineering decisions at the project level.
The pattern here: AI-assisted rewrites are now large enough to generate their own PR cycles, and it's getting harder to separate genuine technical outcomes from vendor storytelling. Fable is not a general-purpose model; it's purpose-built for code migration. The fact that it was used on a high-profile open source project that its creator depends on is not a coincidence.
The counterpoint in the threads is worth noting. Several commenters said they expected the rewrite to hurt Zig and were surprised it didn't. The actual technical result seems defensible. But the discussion kept returning to trust: who benefits from the narrative, and are the benchmarks real?
So what?
If AI-assisted rewrites become a standard marketing format, founders need to read these announcements like case studies from a vendor, not neutral engineering posts. The model doing the rewrite has a commercial interest in the outcome looking good. Evaluate the claims on your own codebase before assuming the results transfer.