Forbidding AI comments: one dev's rules for slop-free Claude Code
seatedro · x · 2026-10-29
Game dev Karl Zylinski found Claude (Aug 2026) good enough to write correct Odin code on the first try — but while building his Karl2D library, he slipped into accepting AI code he no longer properly reviewed, drifting toward 'slop'. After auditing and fixing everything, he imposed three strict rules: no comments from the AI, imperative/procedural code only, and never commit or push unless explicitly told. He argues AI tools are worth using only with guardrails on how you review their output, and explores the junior-senior paradox of overestimating your grip on AI-written code.
More from coding & agent
- Addy Osmani: You're overprompting Claude — one goal plus a check beats prompt bloat — addyosmani · 2026-10-29
- Teach Hermes OPNsense docs with /learn, get an optimized firewall in 15 minutes — Teknium · 2026-10-29
- Interactive ghost story built on Gemma 4 stress-tested with 478 adversarial prompts — Extreme_Ad5709 · 2026-10-29
- Redditor recruiting 20-50 AI agents for a weekly virtual farmers market — fishdogcatman · 2026-10-29
- MCP servers with real shape-library access fix AI diagram generation — Jaig5970 · 2026-10-29
- Augmenting Human Intellect: a developer's case against the pure-agency narrative — ryolu_ · 2026-10-29