AI can write 100 lines of code in seconds and knows every design pattern by heart.
But there’s a trap: it doesn’t know when not to use them. Without context, your "Junior AI" coworker will build a rocket just to go to the supermarket, leaving you with 300 lines of over-engineered "elegance theater" for a problem that only needed 50.
In this session, we’ll move beyond simple prompts to explore a structured, senior-level refactoring workflow. We’ll start with the $440 million cautionary tale of Knight Capital to understand why code smells are business risks, then dive into a practical 3-level framework (Cosmetic, Structural, and Architectural) to categorize technical debt
You will learn a context-first methodology — "Ask first, code later" — that uses human judgment to set hard constraints for AI
We’ll demonstrate how injecting variables like team size and business stability changes the outcome of a refactor, proving that in 2026 the secret to clean code isn't a better prompt, but a better pilot

Learning Objectives:
- Categorize code smells into 3 levels to decide if a design pattern (a "spell") is actually needed or if it’s over-engineering
- Apply a "Context-First" workflow to inject human judgment into AI decisions before any code is written
- Master Constrained Execution to ensure AI refactors remain simple, maintainable, and strictly "test-green"