expansion and contraction technique

When prototyping with AI agents, I noticed I started using a pattern that I called expansion and contraction

Expansion: I create multiple features and give full autonomy to the agent about UI, architecture etc.

In this phase, I want to validate the user experience and flow, not technical details

Contraction: In this phase, I start deleting what is not useful, enforce rules about HOW I want things to be done technically, and reduce the scope of surface area the agent picks decision autonomously

This pattern emerged naturally over many projects and experiments, I find it creates a great balance in not overthinking too early and take back control when the ideas are a bit more clear