Start with a minimal prompt and add features gradually instead of building something complex from the start. This lean approach makes it easier to ship and maintain, and lets you fix problems as they come up rather than getting stuck with an over-engineered mess.
I have a quick tip on prompting for you. Always start out small. So in my experience, what I've seen is as AI models get better and better, they have a tendency to overcomplicate builds and solutions. So the best way to counteract that, at least for me, is to start out small. If I give it a simple set of features that I want to build, I can always kind of build on top of that and fix errors as they come, as opposed to having like this massive Frankenstein of an application that I can't even ship or deploy.
