The Archaeologist’s Copilot
Martin Fowler's article details his approach to restoring a 20-year-old Java 1.5 codebase that no longer built reliably on modern machines. He characterizes the codebase as a 'Big Ball of Mud'. His early experiments with large language models (LLMs) produced plausible answers that did not hold up when applied to the actual codebase. Progress came when he grounded the process in evidence: using AI to support analysis, validating changes in a stable Docker environment, and gradually refactoring while protected by tests. The main takeaway is that AI was most useful when constrained by evidence. The article does not name specific LLM models or Docker versions.
AI-assisted refactoring is effective only when grounded in evidence and validated in a reproducible environment.