AIDA: Asymmetric Impact of Developer AI-Assistance

Abstract

In late 2025, a mid-sized startup software team implemented an AI coding assistant (Anthropic's Claude Code) during a six-week pilot study. The core insight from this implementation is that the AI did not simply boost the volume of output; it fundamentally transformed the nature of the team's work. Routine coding tasks were completed approximately 20% faster, enabling additional time to be allocated to more complex tasks (with a 27% increase in time spent) and to enhancing code quality. Key quality metrics improved: average pull request (PR) size decreased by about 47%, indicating more focused and incremental changes, and the proportion of code commits containing tests rose from 26% to 47%, a 21 percentage-point increase. The team did not deliver more features in less time; rather, they produced higher-quality output within the same timeframe by completing simpler tasks more quickly, addressing complex problems more thoroughly, and substantially increasing testing. These findings highlight the asymmetric impact of AI assistance, with tangible benefits distributed unevenly across work types and team members. We introduce the AIDA (Asymmetric Impact of Developer AI-Assistance) framework as a way to make these distributional changes visible and discuss implications for CTOs and researchers.