[whazor]

You can have small design up front as-well. You write down one or two pages, let the LLM generate code and tests. Keep iterating. If you believe in 100% LLM coded applications, then it makes sense to manage the English input as specifications. Instead of throwing the prompts away, you neatly organize them. Plus you can add additional constraints when the AI does something you don't like.

But I don't trust LLMs to program anything critical, and only do sandbox/tests/demo's. Things where code quality is less important.