Cost-attribution dashboard for AI coding agent sessions on real codebases

The author instrumented 87 Claude Code sessions because nobody could say where the money actually goes on a real codebase. The reusable product is a profiling layer that breaks agent spend down by task, file, and context re-send, so teams can find and fix the expensive patterns.

Signal strength

Why this is worth a look — three verifiable signals, not a black-box score.

• Cross-source

  1 independent platforms · 1 related discussions

  Weak

• Discussion heat

  not yet measured

  Weak

• Freshness

  5 days ago

  Strong

Related topics