TIL : Only 14% of my AI sessions produce a commit

Across 267 Claude Code sessions tracked since December 2025, 38 produced at least one Git commit. 14.2%.

The reflex is to think the agent “writes code for me”. In reality, most sessions are exploration (reading code, understanding a bug), debugging (reproducing, diagnosing), or review (analyzing a diff).

The most-used tools confirm: Bash (2700 calls), Read (1083), then only Edit (545). We read three times more than we write.

This isn’t a problem — it’s the nature of the work. An agent that helps understand and diagnose is as valuable as one that generates code. The commit is not the measure of agentic productivity.