Beyond UML: Making Sense of AI-Generated Code through Visual Architecture

Generative AI allows developers to create large amounts of code faster than ever before. But while code generation accelerates, understanding and maintaining that code often becomes harder.

In practice, AI-generated code frequently leads to oversized classes, unclear responsibilities, and complex dependency structures. But even if we prevent “bad code” — we are missing orientation.

In this talk, Sven introduces a different way of thinking about software architecture: visual feedback instead of static documentation. Using interactive dependency visualizations, developers can see structural problems early and instinctively improve design decisions — often without consciously “doing architecture”.

The session explores why traditional UML diagrams are comparable to paper maps, while interactive visual models behave more like Google Maps: context-aware, exploratory, and immediately actionable.

Rather than presenting another framework or set of rules, this talk focuses on reducing cognitive load, improving developer experience, and staying in control when more and more code is written by AI.

The session is based on real-world experience and is aimed at developers who want to understand complex systems better — not just produce more code.