Don't waste time on specs: /prototype instead

Matt Pocock argues for throwaway prototypes when prose cannot resolve how a feature should look, behave, or model complex logic. His Prototype skill creates a higher-fidelity artifact for feedback and later implementation; it does not make exploratory code production-ready by itself.

Key Points Covered

• Prototype to answer a named question: Cheaper code generation makes spikes more accessible, but discussion remains the lower-cost tool when the unresolved decision does not need working behavior [00:00:00]-[00:02:53].
• Use alternatives to expose design choices: The diagram-search demonstration generates several materially different approaches so feedback can address grouping, filtering, placement, and layout rather than one assumed implementation [00:03:57]-[00:06:03].
• Increase fidelity only when it improves feedback: Pocock iterates toward a preferred alternative and favors testing inside the live route for a more representative experience [00:06:03]-[00:07:55].
• Separate exploration from production cleanup: After accepting the prototype, he proposes a fresh implementation pass that removes prototype code and restores conformance with the original requirements [00:07:55]-[00:08:58].
• Prototype logic as well as interfaces: Small terminal harnesses can exercise state transitions and edge cases when the uncertainty lies in back-end behavior rather than presentation [00:08:58]-[00:10:01].

The video does not compare prototype cost or implementation quality with a specification-only workflow, and it leaves isolation and regression safeguards for live-route experiments unresolved.

Full video: https://www.youtube.com/watch?v=n0VhIVtviC0(opens in a new tab)