I Built a Self-Improving AI Writing Agent (Score Went 72 → 92 in One Loop)

I built an AI system that automatically rewrites its own writing rules and improves your marketing content — no manual editing required. Here's the exact auto research loop, step by step. This is based on Andrew Karpathy's auto research concept, but applied to marketing content creation — LinkedIn posts, newsletters, Twitter/X, and more. My content scoring went from 72 to 92 in a single loop run. In this video you'll learn: ✅ The universal 3-part pattern behind any self-improving AI system ✅ How to build a judge.py scoring engine using binary checks + LLM as judge ✅ How to structure skill.md as your editable AI writing asset ✅ How to write program.md — the instruction loop that never stops ✅ How to add human feedback (feedback.md) for subjective quality control ✅ The composite scoring formula I use across 30+ content skills If you're building AI marketing automation or want an AI agent for marketing that actually gets smarter over time — this is the framework. 🔗 RESOURCES & LINKS —————————————— LinkedIn → https://linkedin.com/in/joonhyeok-ahn Newsletter → https://aitopia.substack.com AI Topia Skool Community → https://www.skool.com/ai-topia-5405 Book an AI Transformation Call → https://calendly.com/joon-getaitopia/30min 📌 CHAPTERS —————————————— 0:00 Why Your AI Content Doesn't Improve on Its Own 2:27 Karpathy's Auto Research Loop — The "Never Stop" Philosophy 7:35 Step 1 — Creating Your Skill Folder & skill.md 12:52 Running the Loop Live — Score Goes 72 → 92 14:33 Applying This to Any Skill + How It Powers AICMO #aimarketing #AIAgent #AIAutomation #MarketingAutomation #LLMasJudge