There is a strong tendency to integrate LLMs with existing apps. Barring few errors that could eventually get corrected, LLMs are fundamentally programmable computational environments. With that notion, every app could be an environment that is dynamically instantiated. Do we need to fundamentally change software delivery? Treat LLMs as dynamic computational environments, generate purpose-specific interfaces on-demand, create adaptive, context-aware interaction models. Eliminate fixed application paradigms?