Two Principles for Effective AI-Assisted Development
Practical observations from months of building production software with AI coding assistants.
Principle 1: Describe outcomes, not procedures. The AI consistently delivers better results when given a clear description of the desired end state rather than step-by-step implementation instructions. Detailed procedural guidance often leads to circular debugging loops where the AI tries to satisfy conflicting constraints. Counterintuitively, high-level directives like 'make it performant and beautiful' produce more coherent solutions.
Principle 2: Install Karpathy's coding rules. The four laws (40k+ GitHub stars) fundamentally improve AI behavior: think before coding, prioritize simplicity, make surgical changes, maintain goal-driven execution. When loaded as a system prompt or skill, the AI asks clarifying questions instead of guessing, avoids unnecessary complexity, and stays focused on the task at hand.