AI increases throughput. It also increases the speed at which teams can manufacture entropy. Better agentic programming is therefore mostly a problem of decreasing entropy as much as possible.
The idea of entropy is from The Pragmatic Programmer. Roughly, the idea is that entropy in code (like in physics) increases over time due to small structural compromises accumulate until the next change becomes harder, riskier, and more confusing than the last.
High software entropy is characterized by hugely complex code bases.
Complexity is anything related to the structure of a software system that makes it hard to understand and modify the system.
Fix 1: Track "entropy hotspots”
- files with high churn
- files touched by many unrelated features
- modules with wide import graphs
- modules nobody can test without a full environment