Fundamentals set you free.

6 libraries, 3 tools, one C11 monorepo: bcduplicate finds duplicates at 758k files/s — 18× faster than rmlint on the same thread count.

82 % of developers use an AI weekly. A randomized study measured −19 % productivity on experienced devs. 80 % of companies see no gain. Six C libraries, four CLI tools, and a full intranet — built in four months for $1,000 in AI usage. What actually worked, and what I still refuse to delegate.

Why find | xargs -P plateaus at scale, and how three phases separated by a strict barrier — parallel walk, sequential merge, parallel process — take a recursive hash from 12s to 7s on 650k files.

Find duplicates among a million files in 3.7 seconds: a size → fast-hash → full-hash pipeline, io_uring, and adaptive dispatch. Numbers, architecture, and what the hash choice actually changes.

malloc calls brk for small allocations and mmap for large ones. What really happens under the hood, syscall by syscall.