Rust

In eleven days, between 3 and 14 May 2026, Jarred Sumner ported Bun from Zig to Rust: 535,496 lines of Zig rewritten under a harness of roughly 50 dynamic workflows in Claude Code. The motive was a single bug class, use-after-free, double-free and leaks produced by mixing JavaScriptCore's garbage-collected memory with manually managed memory. In safe Rust those failures surface as compilation errors, and cleanup falls to `Drop` and RAII. Sumner's framing was that this is a better feedback loop than a style guide.

Kent Beck met the opposite face of the same property. Building BPlusTree3 across 110 to 130 hours, he stalled on Rust's ownership model and the compounding complexity it produced, then had the code written in Python first and transliterated into Rust by Augment's Remote Agent to get moving again. A January 2026 research report on language fit for Spec Drive Development rates Rust high friction for the same reason: the Borrow Checker collides with the probabilistic output of LLMs, and the resulting correction loops cost tokens and time.

Rust also turns up as the chosen substrate wherever coordination is the bottleneck: Home Mixer and Thunder in the X recommendation stack xAI open-sourced on 15 May 2026, and the Nostr relay behind Block's Buzz. The guarantee and the friction are one mechanism viewed from two ends. After the Bun port, about 4% of the code is `unsafe`, and 19 regressions were fixed following the merge.