Level 2: Directing AI with Prompts

The most common mistake: dumping your entire vision into a single prompt and hoping for the best.

"Build me a complete booking system with a calendar view, customer accounts, email notifications, payment processing, and an admin dashboard."

This fails for several reasons:

AI loses focus. The more you ask for at once, the more likely AI will cut corners on each part. It's trying to satisfy everything, so it satisfies nothing well.

You can't course-correct. When everything comes out at once, you can't tell which part went wrong. Was it the calendar? The accounts? The payments? You end up starting over instead of fixing.

Context gets diluted. AI has a working memory. Fill it with ten features, and the details of each one get fuzzy. Ask for one thing at a time, and you get AI's full attention on that thing.

You skip the thinking. Asking for everything at once usually means you haven't thought through the sequence. What needs to happen first? What depends on what? AI can't figure that out for you.

The solution isn't to ask for less. It's to ask in sequence—one clear request at a time, reviewing the output before moving on.

AI doesn't know what kind of help you need unless you tell it. The same request gets very different responses depending on the role you assign.

Think of it like hiring. You wouldn't ask "someone" to help with your project. You'd hire a frontend developer, or a designer, or a copywriter—each brings a different lens.

Assigning a role does two things: 1. It focuses AI's knowledge on what's relevant 2. It sets the tone and format of the response

The key to effective prompting is sequencing. Instead of one massive request, you break work into steps—each one small enough to review before moving on.

This isn't slower. It's faster. You catch problems early, when they're easy to fix. You build on working foundations instead of shaky ones.

AI will give you something that looks right. Your job is to verify it actually is right before building on top of it.

You don't need to understand every line of code. You need to check that the output matches what you asked for and works as expected.

Getting things right on the first try is rare. Iteration isn't failure—it's the process.

Professional developers don't write perfect code immediately. They write something, test it, find problems, fix them, and repeat. Vibe coding works the same way, except AI does the writing and you do the directing.