Summary
IndyDevDan walks through a complete ground-up rebuild of his `/plan` meta-skill, specifically re-engineered for Fable 5 and Mythos-class AI models. The premise is simple but pointed: most engineers hand planning off entirely to their coding tools — Claude Code, Cursor, OpenAI Codex — and that, Dan argues, is a mistake that caps the quality of everything downstream. The video is a slow, detailed engineering vlog aimed squarely at developers who want end-to-end control over their agentic workflows.
Dan begins by writing a raw markdown brainstorm document before touching any tooling, then iteratively constructs a structured spec template. Key design decisions include a phased implementation format where every phase is self-contained and independently validatable, an embedded checklist system that agents update in real time as they execute work, and a per-phase testing strategy block that specifies exact validation commands. He also explains why he moved away from Cursor — the tool deprecated its tab-completion model in favor of a pure-agent interface, removing the last reason he was using it.
The broader argument is that the planning-to-reviewing ratio is the lever that separates strong agentic engineers from average ones, and that with each new generation of Mythos-class models, the returns on a well-structured upfront plan compound further. Developers building production AI systems who want reproducible, reviewable output at scale will find this a practical and opinionated reference.
📺 Source: IndyDevDan · Published June 22, 2026
🏷️ Format: Hands On Build
