AI coding shouldn't be infinitely fast: review and testing are the real constraints, argues Allen Holub
nodosenlared · x · 2026-10-03
Allen Holub pushes back on the notion that AI-assisted coding should be frictionlessly fast. In a thought experiment: if all bottlenecks vanished and you shipped eight months' worth of work every eight hours, with 10% of the program changing each cycle, users would constantly find previously working features broken and relearn chunks of the app—rendering it effectively unusable. The 'slowness' of review and testing isn't waste to eliminate but a necessary constraint keeping software coherent.
More from coding & agent
- Vibe coding is driving major projects to rewrite in Rust over cheaper vCPU costs — evilsocket · 2026-10-03
- Users slam Dots: permission blocks, dead sub-agents, and "unlimited Astra" quietly burning local Codex limits — brandon_galang · 2026-10-03
- The foo[count:-count] gotcha: -0 silently empties your array slice — theshawwn · 2026-10-03
- Dev open-sources an approval-queue tool that stops AI agents from sending anything until you sign off — paulofilip3 · 2026-10-03
- Princeton-backed Choir open-sources a protocol for multi-agent autoformalization on GitHub — burny_tech · 2026-10-03
- Running 256k-context open models on 2x RTX 3090 for months: a home server LLM retrospective — knighty1981 · 2026-10-03