Stop Coding and Start Planning

Klaassen argues that AI made developers sloppy by eliminating the perceived need to plan. The thesis: vibe coding (prompt and hope) produces features that work but teach the system nothing, while planning with AI produces features that compound — each plan teaches the system how the developer thinks.

The framework introduces “three fidelities”: quick fixes (just do it), mid-complexity features where planning yields massive ROI, and big uncertain projects that need “vibe planning.” For Cora’s email bankruptcy feature, one hour of planning with a research agent saved days of building the wrong thing by surfacing Gmail rate limits and timeout issues before any code was written.

The planning agent analyzed Figma designs and produced implementation plans grounded in existing codebase patterns. A second agent then compared screenshots of the built result against the original designs. Five screens came out pixel-perfect, including mobile layouts that were never designed.

RDCO mapping: The planning-before-execution pattern applies directly to content strategy and operational work. The insight that plans teach systems how to think while code only teaches how to solve individual problems reinforces the vault-first approach — documenting reasoning creates reusable institutional knowledge.