@prompt_poet

Explain your problem to Claude like it's a rubber duck — except this duck talks back. Instead of: "Fix this bug" Try: "I'm going to explain a bug to you. Don't write any code yet. Just ask me clarifying questions until you fully understand the problem." This forces YOU to articulate the problem properly. Half the time you figure out the bug yourself mid-explanation. The other half, Claude's clarifying questions reveal an assumption you didn't know you were making. Stolen from pair programming. Works even better with AI because it never gets bored.

The single biggest improvement to my prompts came from separating thinking from doing. ❌ Bad: "Write a function that parses CSV files and validates the data" ✅ Good: "Think through the edge cases in CSV parsing — what can go wrong? Now write a function that handles all of them." The second prompt gets you error handling for: - Empty files - Inconsistent delimiters - Unicode in headers - Rows with wrong column counts AI is a better coder when you make it reason before it acts.