AI Usage Dashboard
The sibling half of the June 22 split: a team-level dashboard on how well people are using AI (value by platform and work model, efficiency signals, spend cohorts), with drill-down into the developer profile.
- Jun 22 intake meeting
- 152 traceability tags
- 8 sections
Origin
In the same June 22 meeting that produced the AI Tooling Dashboard, the CEO explicitly cut the original monolithic manager dashboard in two: "a workflow-level view… versus a how-well-am-I-using-AI page as an engineering manager." Leadership pressure shaped the frame (trend lines over snapshots, progression over time) while the Head of Product supplied the sharpest product question:
If they're doing agentic, are they having to redo a lot of that work, meaning they need help with their prompts?
Data grounding
Two structural findings. Per-platform cost and sessions are solid, but value output and tool identity live in fact families that don't natively join at fine grain, so per-platform value is estimated at best. And the budget-cohort objective is fully data-blocked: no tier dimension exists, so developers cannot be sorted into plan cohorts at all. The doc names the honest alternative instead of faking one: rank by actual spend into dynamic quartiles. It is equally explicit that no true rework rate exists and prompt text is never inspected: "specificity" is a count proxy, and the design had to say so.
Design decisions
Stop over-claiming, rename the section. "Prompt quality" became "AI efficiency signals": the product measures prompt volume and output survival, not quality, and the label had to stop promising otherwise.
Spend-rank quartiles over fictional tiers.With budget tiers unbuildable, the question was reframed as "do high-spend developers produce proportionally more?"
Design the target, ship the fallback. The three-way work-model split is drawn as the target design with a two-way fallback until the underlying data populates.
Hard scope boundary. A coverage table documents what lives on this dashboard versus the workflow dashboard versus the executive report, so the split survives contact with future feature requests.
Self-review by annotation
152
Traceability tags
19
Numbered decisions
14
Data-relevance findings
7
Out-of-scope markers
As long as we define how we are saying it and people understand it, then the report can make sense.