- Developers see only their own numbers.
- Engineering managers and executives can pick any developer they may see, or All developers, from the picker at the top of the tab. It starts on You.
What’s on the tab
Developer scores
Your three scores, each out of 100, compared with the organization. A person needs at least 5 PRs that reached production in the range for a score; otherwise the tile says why it has none. Your scores are visible to you and to the people who manage you. Other developers never see them.Why this score
Pick a score to explain. The card lists the figures behind it, for example lead time, PRs per week and output for Velocity, each with your value, the organization’s value and how many points it moves your score. The points add up to the difference between your score and the organization’s. Below the figures, How to raise it gives up to three tips, starting with the figure that costs you the most points. No prompt text is read for any of this. The figures come from your code host, your AI tools’ billing and, if installed, the AI ROI Agent.Where you stand
Four tiles and a chart of your AI coding spend by tool over the range:Projects
A project is a Linear project or Jira epic when your PR names a ticket, otherwise its repository. Every merged PR lands in exactly one project.How projects compare
A scatter chart. Each dot is a project: further right means a higher median AI cost per PR, higher up means more estimated rework, and a bigger dot means more AI cost. The dashed lines mark the median project. Lower left is better.Project by project
The Vs similar verdict is one of:
Use it to find the projects worth copying (what works there?) and the ones worth a second look.
How you use AI
Three cards about one person’s AI use over the selected range. Engineering managers and executives see the developer picked at the top of the tab; everyone else sees their own. When All developers is picked, this section shows your own.Work types
Your prompts as the AI ROI Agent labelled them (Build, Fix, Understand, Docs, Ship, Other), and the Claude Code cost that followed each type. A follow-up prompt keeps the type of the one before it. Cursor prompts are counted, but Cursor bills per hour with no work type, so its cost isn’t split here. Needs the AI ROI Agent with prompt labels turned on.Cost drivers
What shaped your Claude Code cost, each compared with the previous period:
These come from Claude Code’s own telemetry, priced at list price, through the AI ROI Agent.