The Archaeologist’s Copilot

This article details the process of restoring a 20-year-old Java 1.5 'Big Ball of Mud' codebase that no longer built on modern machines. The author describes pitfalls like relying on AI-generated answers that don't hold up, and shares a practical strategy using AI for analysis, Docker for stable environments, and gradual refactoring protected by tests. Key phases include analysis, containment, lift, and refactor, with lessons on avoiding the 'Tourist Trap' and 'Lying Tests'. The article emphasizes that AI is most useful when constrained by evidence and a step-by-step modernization plan, making it highly relevant for developers working on legacy system upgrades.