22 days ago
- The study investigates whether the cleanliness (structural and stylistic quality) of code affects the performance of autonomous coding agents.
- It introduces a minimal-pair evaluation protocol, comparing repositories identical in architecture, dependencies, and behavior but differing in static-analysis violations and cognitive complexity.
- Results from 660 trials with Claude Code show that code cleanliness does not significantly change task pass rates.
- However, agents working on cleaner code use 7–8% fewer tokens and reduce file revisitations by 34%, indicating improved efficiency and reduced computational cost.
- The findings highlight that traditional maintainability principles remain relevant in AI-driven development, influencing agent behavior alongside model choice, harness, and prompting.