New Tool CodeBurn Details AI Coding Spend
TL;DR. A new tool, CodeBurn, tracks AI coding API calls to reveal less than half the spend goes to code generation, with significant costs in exploration and debugging. - The tool categorizes AI coding activities into 13 tasks, providing a deterministic breakdown of token usage and costs. - The analysis shows AI models spend about 40% of their budget on understanding problems and reasoning, acting as a collaborator. - CodeBurn reads session data locally, ensuring no API keys or sensitive data leave the user's machine for privacy.
- CodeBurn tracks AI coding API calls, identifying where money is spent beyond code generation.
- Only 47.9% of AI coding spend directly produces code, with 40% dedicated to thinking and problem-solving.
- The tool provides deterministic classification of 13 task categories without using LLM calls or external APIs.
Sources
- Where AI coding spend goes: 48% code, 40% thinking — codeburn.app