Where AI coding spend goes: 48% code, 40% thinking (codeburn.app)

A new tool called CodeBurn has been developed to analyze AI coding API expenditures, revealing that only 47.9% of spending directly contributes to code generation. In a detailed spend breakdown of $7,890 over 105,718 API calls, nearly 40% of costs were allocated to activities such as exploration, debugging, and collaboration, reshaping the perception of AI's role from a mere code generator to a collaborative partner that invests significant resources in understanding problems before crafting solutions. The significance of CodeBurn lies in its deterministic classification of API calls into 13 task categories without using LLM calls, ensuring reliability in measuring productivity. Key features include a dashboard for activity tracking, optimization recommendations, model comparisons, and yield tracking, which assesses the effectiveness of coding sessions against git commit history. This innovative approach not only examines how money is spent on AI coding tools but also highlights potential waste and optimization areas, making it a valuable asset for developers striving for efficiency in AI-assisted coding.