Summary
Peter Yang demonstrates how to build a personal AI life and career advisor inside OpenAI Codex (also compatible with Claude Code) using nothing more than a folder of structured markdown files. The system is built around four documents: skill.md (advisor behavior rules), plan.md (personal context including goals, principles, and business details), learnings.md (insights that accumulate across conversations), and an eval checklist the AI runs before delivering advice.
Yang shares real examples from his own advisor, including advice it gave him over three months while he was deciding whether to leave his job — guidance around psychological readiness, family priorities, and income targets. The tutorial walks through each document in detail, showing exactly what sections to include in plan.md (goal statement, principles, energy sources, customer profile, financial context) and how to configure skill.md so the AI adopts the role of a trusted friend rather than a generic assistant. An optional Mercury MCP integration lets the advisor pull live bank data to track progress against financial goals.
The broader argument Yang makes is that the quality of AI advice is almost entirely determined by the specificity of the context you provide — and that writing a clear one-page plan document is worth doing even if you never build the AI system around it. Useful for founders, independent creators, and anyone who has found AI advice frustratingly generic.
📺 Source: Peter Yang · Published June 17, 2026
🏷️ Format: Tutorial Demo
