dotey on the AI code quality debate: black-box verification is replacing code review
dotey · x · 2026-09-18
dotey clarifies his position in the ongoing debate about how non-engineers can trust AI-generated code: the disagreement is smaller than it seems—it's not about ignoring quality, but about how quality is ensured. White-box code reading is giving way to black-box behavioral verification.
Key points: those who can't read code should verify function, performance and security externally; even engineers rely on contracts plus tests (as with compilers and libraries), not line-by-line review; perfect information before acting doesn't exist—iterate in small, verifiable, rollback-able modules; and in the AI era, plausible-looking reasoning is a worse signal than results. For safety-, money- and data-critical paths, bring in professionals when unsure.
More from coding & agent
- Experimental Jev-powered RLS linter catches Supabase policy bugs rule-based lint can't — SeanOliver · 2026-09-19
- Virtual green screen from a RealSense D436: depth-based RGB filtering, no backdrop needed — chrismatthieu · 2026-09-19
- Runway MCP turns video to 120fps with a single prompt — tlakomy · 2026-09-19
- Cost breakdown: 100 LLM calls in 8.5 seconds for roughly a quarter of a cent — narphorium · 2026-09-19
- Dev builds an interpreter for an English logic language so an LLM can reason via proof search — narphorium · 2026-09-19
- AgenticLinux Ships an Immutable Distro Preloaded With Claude, Codex and Docker Sandboxes for Agents — steipete · 2026-09-19