Your AI cost dashboard tells you how much. Not why.

Description and use

A dashboard that attributes AI costs to workflow shapes enables pattern spotting over time, helping teams adjust practices, but requires linking agent logs to commits.

Problem

Cost dashboards report total spend but lack insight into the nature of the work that incurred the cost, making it hard to distinguish productive exploration from wasteful activity.

Outcome

By tagging AI costs with workflow shapes and linking them to commits, teams can spot patterns such as increasing exploration costs relative to shipped features, enabling targeted process improvements.

Workflow

• Collect agent logs from Claude Code and Codex.
• Parse logs to identify session boundaries and actions (e.g., file reads, edits).
• Correlate session data with git commits to determine which sessions led to code changes.
• Classify each session into workflow shapes (exploration, implementation, review, failed attempts, rework).
• Aggregate costs and hours by workflow shape per project.
• Display aggregated data in a dashboard that shows cost trends over time by shape.
• Use trend analysis to identify projects where exploration cost is rising relative to shipped cost.
• Adjust team practices or tooling based on insights (e.g., improve task descriptions, refactor codebase).

Evidence

A cost dashboard tells you how much you spent. It almost never tells you why.

Source

The console can tell you how much a workspace spent, but "what it can't tell you is which human beings caused it" — or, I'd add, what kind of work caused it.

Source

To classify a session by workflow shape you'd need to correlate agent logs (which you have) with what actually reached a commit (which lives somewhere else entirely — git, not the session file).

Source

Where this is actually useful, I think, is pattern-spotting over weeks, not judging a single session.

Source