Carmack maps assembly decline to Musashi-era martial arts shift under AI pressure

Carmack referenced a translation of Musashi's Book of Five Rings, tracing Japanese martial arts from pragmatic combat tools to regulated sports after the post-WW2 ban. He paralleled this to programming, noting hand-assembled opcodes and low-level work now limited to niches like FFmpeg while broader skills face devaluation.

OpenAI's 2021 Codex paper documented 28.8% pass@1 on HumanEval for code generation; subsequent models reached 67% by 2023 on the same benchmark. GitHub Copilot telemetry showed 46% of code accepted in sessions tracked through 2024, confirming measurable displacement of manual entry tasks.

The pattern matches prior skill compressions: assembly yielded to C, then to frameworks. Operational teams that retain only high-level oversight without updating verification pipelines risk capability gaps against automated baselines.

Retro scenes preserve techniques for archival value, yet production velocity now tracks model iteration cycles rather than individual opcode mastery. Teams must embed AI output review into existing CI gates within 12 months to avoid divergence from deployed standards.