After 30 years of coding, I finally figured out the real AI skill that matters: breaking problems into pieces AI can handle, not chasing the latest tools or models.
The story
After three decades of coding and eight years deep in AI, I just cracked something that’s been sitting right in front of me: the make-or-break skill for AI isn’t knowing the fanciest models or the latest tools – it’s knowing how to slice problems into chunks AI can actually handle.
I’ve been building with agents and prompt engineering for years, but it finally clicked when I stopped treating prompts like instructions and started treating them like code architecture.
Here’s what changed everything:
I now start every AI task by mapping the exact outcome I want, then identifying dependencies. Sounds basic, but most people skip this. They throw everything at the AI in one massive prompt and wonder why it hallucinates or misses the point.
Then I choose my decomposition approach – top-down when I’m working through complexity, bottom-up for repetitive components, divide-and-conquer when things can run parallel. Just like deciding between recursion and iteration.
The game-changer was treating prompts like object-oriented programming. I build modular prompts: one sets context, another defines format, others handle specific logic. When something needs tweaking, I update one module instead of rewriting everything.
For complex reasoning, I use chain-of-thought to walk the AI through steps. For multi-stage work, I decompose into separate sub-prompts and orchestrate them. When computation matters, I have the AI express reasoning as actual code – usually Python – then execute it.
The skills that make this work aren’t AI-specific. Analytical thinking to see systems and components. D
Recent Comments