Prototype to decide

In the old world, someone wrote a design doc. Teammates left comments. You argued in a meeting. Eventually someone just went and built it and there was often a significant delay before the team discovered whether the approach missed important requirements.

Coding agents make prototyping nearly free. When prototyping is free, the fastest way to settle a design debate is to build it. An engineer can spin up a rough prototype in minutes, share the result with their team, and make a decision grounded in something real rather than something imagined.

Ideas typically start as agent prototypes. You run Claude Code, get something rough, and bring that context into a plan to iterate with your team. But prototyping also happens mid-plan. Hit an unknown? Fork the context, spin up a quick prototype to validate it, bring the result back. The boundary between planning and building is intentionally blurred.

Plans should contain code — not as attachments, but as first-class messages. When a teammate wants to challenge an approach, they don't write a comment arguing about it. They fork the context, prototype their alternative, and bring the result back. Proving a theory about the intent, not just debating it.