LLMs are becoming sufficiently capable of generating correct code for complex scenarios, but the intricacies of incremental front-end changes, from simple one-off stylistic updates to swapping out entire UI pages while keeping existing application logic are full of surprises and pitfalls. In this talk, I’ll discuss a simple web UI DSL that can describe abstract UI concepts and act as an intermediate language between concrete design, and UI and application logic, and train ChatGPT to read and write this DSL to WebSharper+F# code. This approach yields excellent results, and with minimal tooling, it can be augmented to yield easily applicable, incremental UI changes for prompts of any size.