One thing I use often use ChatGPT for is drafting UI mockups for C#/WPF apps and web apps. Currently, I have to copy over the code locally to preview what the code should be doing. I think it would be a great feature to have a canvas option that shows what the code provided should render, whether that be an artificial generation or something that actually runs code on the backend and streams back what it actually produces. The GPT would determine what data it needs to simulate or request that data from the user. An additional utility with this could be a component hierarchy navigator that works similar to current browser and IDE tools so you can see how the pieces fit too.
This could be further expanded into a UI/UX design tool. Imagine being able to tell the GPT to ensure alignment of text inputs and labels or perhaps use it to preview what your HTML/CSS code looks like on a series of platforms (different browsers, mobile devices, etc) to see if there are any design or compatibility issues. You could give it a palette or mood of colors to use to optimize the look and feel of your app. You could ask it to make sure it has accessibility options for those who need it.