Catching an AI Usage Problem Before It Became a Habit

I built this for a friend who oversees three teams at his company — sales, marketing, and operations, 15 people total under his purview. He knew what I could build and asked for help after rolling out both GPT and Claude across those teams. The tools weren't the problem. Seeing how they were actually being used was.

Tracking meant logging into two separate platforms and trying to piece together who was doing what, with no way to correlate the two or see any one person's full picture in one place. As he put it, he has too many other things to do all day than check usage caps between users and platforms one by one.

Specifically, he needed a single view that could answer: who's actually adopted the tools and who hasn't; whether people lean toward GPT or Claude for different kinds of work; when someone's about to hit a cap; whether one team is carrying more weight than the others; whether usage clusters around certain days; whether he's paying for seats nobody's using; and whether people are doing the same task twice across both platforms.

We built a single dashboard that pulls live usage data directly from OpenAI's and Anthropic's official usage APIs — every person across the three teams, side by side, filterable by role, access level, team, or name. No manual exports, no jumping between tabs.

The interface runs on a visual system I designed myself first: a wave that fills a progress bar as usage climbs from 0 to 100%, styled with a claymation look rather than a flat gray bar. It's fully collapsible — two compact bars on screen, expandable to show both the 5-hour and 7-day usage windows, or tucked away entirely when it's not needed.

In the first week, the dashboard flagged a new team member blowing through their usage cap. That visibility led to an actual conversation about what they were doing that was eating through usage so fast — the answer turned out to be a repeated manual process, which got turned into a reusable skill instead. Their usage dropped, and the fix now applies every time that same task comes up again.

He's said the dashboard finally gives him a quick read on where everyone stands, without having to carve out time to dig for it himself across two platforms and fifteen people's worth of caps.

The live view solves today's problem, but both platforms reset every 5 hours and weekly — so nothing sticks around long enough to show a trend. The next phase, already underway, is a database that snapshots daily and weekly usage by person and platform so that history doesn't disappear. Once it's in place, he'll be able to chart usage by team or individual over time — and start making real decisions about a larger, company-wide rollout instead of guessing at it.