Agentic Coding and Expertise Framework Analysis

This technical report presents a framework for studying interactive agentic coding, based on an analysis of approximately 400,000 Claude Code sessions conducted between October 2025 and April 2026. The report evaluates the composition of tasks, human-AI collaboration, and success rates in coding sessions. It details how users primarily make planning decisions while the AI handles execution decisions, with findings indicating that the level of domain expertise significantly influences the effectiveness of using Claude Code. The report outlines that as users' domain expertise increases, the success rate of sessions rises, although the difference between intermediate and expert users is modest. Additionally, the report highlights a notable reduction in debugging time and an increase in the complexity and value of tasks over the observed period. The findings suggest that while agentic coding tools are changing the landscape of knowledge work, they do not replace the need for domain expertise, which remains crucial for effective collaboration with AI tools.