For the "ask" phase, I feel like that the chat interface makes sense as the first step for the adjust phase because ChatGPT is a LLM, which works seamlessly with any natural language prompts.<p>But it's a bad interface for most applications. It took me out of context on the work I'm doing and forces me to describe what I am working towards in natural language. I will still use it now because it is a big step forward from Googling. But I think the bigger challenge (which brings the much better experience) lies in how we could deeply integrate the iterative creation process with ChatGPT into the app workflow itself.