92% reviewed.
78% AI-assisted.
Two quarters. Nine engineers. One internal full-stack dashboard that turned a vibe into a baseline — and a baseline into a habit. Now scaling toward the 60-engineer Sportsbook channel.
Jira and GitLab couldn't show AI usage. The team had no shared baseline.
Anecdotes everywhere. Some engineers had Copilot pinned, others wouldn't touch it. Leadership wanted a number. The team wanted to be left alone. Neither group had what they needed: a shared, honest picture of where AI was actually moving the work.
Build the dashboard nobody asked for.
Internal dashboard, two weeks.
Stitched GitLab events, Jira metadata, and self-reported PR labels (ai-reviewed, ai-assisted) into a single team view. Owned the schema; owned the read path.
DORA + AI-usage metrics, side-by-side.
Lead time, deploy frequency, and AI-assist rate on the same chart — so quality and speed couldn't be argued in isolation.
Surfaced gaps weekly, not quarterly.
A 10-minute Friday review. Names off, patterns on. Trends visible before they became culture.
Shared prompts, not policies.
A small library of prompts and review checklists, versioned in the repo. Suggestions, not mandates.
The dashboard didn't change the team — the conversation it forced did. Once the number was visible, debate moved from “should we” to “where next.” Adoption stopped being a top-down ask and became a peer signal.
- —Ship the metric before the dashboard. A spreadsheet for two weeks would've validated the signal faster than a polished UI.
- —Define "AI-assisted" earlier. The first month was spent re-litigating what counted. A loose, public definition would've moved faster than a perfect one.