Prompt Craft
Explore how to write effective AI prompts by controlling scope, targeting specific app parts, focusing on user outcomes, and protecting existing features. This lesson helps you reduce errors and improve the accuracy of AI-generated app changes by sending clear, precise instructions one at a time.
You have a working app. Three rounds of iteration turned a rough first build into something real. Now you want to add features: a favorites system where customers can save preferred walkers, a filter bar on the homepage, and a profile page for each walker with reviews. You type it all into one message and hit enter.
The AI produces something. The favorites button exists, but it does not save anything when you refresh the page. The filter bar appeared, but the walker cards now render in a single column instead of their grid. The profile pages exist, but the navigation bar vanished from half the screens. And the booking flow you spent three rounds perfecting? It still works, except the confirmation page now shows “undefined” where the dog’s name should be.
You asked for three things and got five problems. The prompt was not wrong, exactly. It was too much at once, with too little guidance about what to leave alone. This lesson covers four skills that prevent this: controlling scope, pointing the AI at the right part of your app, describing outcomes instead of code, and protecting features you have already built.
How much should you ask for in one prompt?
One thing. Sometimes two if they are closely related, but almost never three.
This feels slow, but it is actually faster. When you bundle five requests into a single message, the AI tries to change everything at once. Each change can interfere with the others. A layout adjustment in one area shifts the spacing in another. A ...