YouTube Summaries

← All summaries

Zig is at a crossroads

2026-05-08 Fri ⏱ 17 min @ThePrimeagen

Bun forked Zig to add ~4× faster debug builds via parallel semantic analysis, and noted they won't upstream because Zig has a strict ban on LLM-authored contributions. Prime yap-storms about whether the ban is enforceable, whether Anthropic forking an MIT project feels icky even when it's legal, and then walks it back after Zig responds.

The controversy

  • Zig's CoC bans LLMs for issues, PRs and bug-tracker comments (including translations).
  • Bun forked rather than fight the policy. Crime against humanity: they didn't call the fork "Zag."
  • Enforcement is fuzzy — tab-autocomplete is technically an LLM. The ban is more about culture than mechanism.

The vibe complaint

  • Zig has been in development ~10 years and stays pre-1.0 because Andrew Kelley moves carefully (e.g. async/await took years and landed well). Slowness here is a feature, not a bug.
  • Anthropic forking a small MIT project funded by Andrew's $150/mo donations feels greasy even though the license permits it. Prime can't articulate why beyond "Anthropic feels adversarial to devs."

Zig's response (recorded from a hotel bed)

  • "AI is entirely beside the point." Bun's parallel semantic analysis is non-deterministic — Zig has been working on a correct version that needs the rewritten type-resolution semantics to avoid spurious errors.
  • Real bottleneck is LLVM, not serial sema. Zig's experimental self-hosted backend already gives ~4× and incremental builds drop 40 s → 0.5 s — orders of magnitude beyond Bun's win.
  • Lesson: spike with an LLM if you want, but ship after you've understood the actual bottleneck.