Today I saw Thomas Ricouard share this image.
He didn't explain it, but it's clearly an image generated by ChatGPT Images 2.0, and I'm pretty sure he was imagining a dungeon crawling game as a curious hybrid of fantasy art and a native SwiftUI app.
It made me wonder about using the latest models to not just vibe code UI, but to provide conceptual examples of UI that are more rapid than even vibe coding.
And it's not that these UIs are good! It's rather that, as with lots of vibe coding examples, they overcome that initial hump that can be so daunting, because they provide something to work against. As if they trigger an allergic, "well, not that!" type of response. And then make one say, "Let me show you how it should be…" (Lalit Manganti has a good post on this, saying, "The key is the emotional reaction I have immediately to the LLM’s response, either excitement or frustration." [Lalit's emphasis.])
After I saw Thomas's image, I began trying my own, wondering, "What might a real-time space-based 4X game look like as a native SwiftUI app."
Again, it's not great. But it's something. And it's not bad; not because it's good, but rather because it's believable enough to see something there and say, "well, no, but maybe like this."
Or here's a similar game concept as a terminal UI. There's so much that's wrong here, but that's exactly the point. It's bad enough to be unusable, but believable enough to trigger something polemical.
Unusable but useful. Maybe I'm just lazy, but that feels like a helpful place to start.