Does Code Cleanliness Affect Coding Agents?

As autonomous coding agents gain popularity, their evaluation has largely centered on task completion rates. However, the impact of the underlying code's structural and stylistic quality, or 'cleanliness', on an agent's ability to navigate and modify it remains underexplored. Our study introduces a new evaluation protocol using minimal pairs to isolate the effects of code cleanliness. Findings indicate that while code cleanliness does not affect the agent's pass rate, it significantly reduces operational footprint, suggesting that traditional maintainability principles are still crucial in AI-driven development.