Stop asking AI giant messy questions and expecting clean answers. Break problems into small chunks first, then prompt each piece separately. 30+ year coder shares the one technique that made AI actually useful for real work.
The story
You know that feeling when you ask AI for help and get back a wall of text that’s 80% wrong and 20% “well, sort of”?
I’ve been coding for over 30 years and working with AI for the last 8, and I finally figured out why that happens. It’s not the AI’s fault. It’s mine.
The problem? I was asking one giant messy question and expecting a clean answer.
Here’s what changed everything: I started breaking problems into chunks before handing them to AI. Not after. Before.
I call it decomposition prompting, and it’s dead simple. Instead of “build me a REST API,” I ask the AI to list the endpoints first. Then I have it specify inputs and outputs for each one. Then I get it to write individual functions. Then tests. Then we pull it all together.
Each step is small enough that I can actually check if it’s right. And when something’s off, I know exactly where to fix it.
This isn’t new thinking for me as a developer. I’ve been modularizing code forever. But applying it to how I prompt AI? That’s been a game changer.
The accuracy goes up. The debugging gets easier. And I’m not drowning in irrelevant information because I put word limits on each chunk.
For anything complex – planning, coding, managing agent swarms – decomposition is the difference between “this almost works” and “this actually works.”
For simple stuff like “what’s 2+2,” skip it. But if you’re trying to get AI to solve real problems, breaking them down first isn’t optional anymore.
It’s the one skill that’s made everything else I do with AI better.
Recent Comments