The 200,000 line code repository: My AI wrote 1000 lines. I needed 100.

A codebase rarely becomes unmaintainable overnight. It grows one feature, one abstraction, one workaround, and one poorly documented decision at a time. AI coding agents risk accelerating that growth by making it easier than ever to produce code—and harder to keep the resulting systems understandable.

In this talk, we’ll explore the hidden technical and cognitive debt of AI-assisted development. How do simple requirements turn into sprawling implementations? What happens when codebases grow faster than documentation and developer understanding? And how can teams review increasingly large changes without letting unnecessary complexity slip into production?

Through a live demonstration using IBM Bob on an existing codebase, we’ll investigate an AI-generated implementation, identify redundant logic and unnecessary abstractions, improve documentation, and simplify the solution while preserving its behaviour.

The goal isn’t fewer lines at any cost. It’s a codebase where every line earns its place, every important decision can be understood, and the next developer doesn’t have to reverse-engineer what the last agent created.