Follow a real feature from loose idea to verified deployment. Each step builds directly upon the boundaries you established in previous steps.
Define the core problem, the user, and what must be strictly excluded from the first version.
Naming what NOT to build gives the AI assistant narrow guardrails.