One Level
Less
Studio had two levels that rendered the same page, and one tile shape standing in for everything on it. Nobody could hold a picture of the product in their head without a person there to explain it. The fix wasn't a feature. I took a level out.
Three levels became one.
The last click re-opens the app you never left.
You are in one company's Customers. You need the same app in another company's project. There is no way across, only up and back down. Customers → project → company → root → the other company → its project → Customers again.
One shape stood in for three different things.
A tile was a company. A tile was a project. A tile was an app. Same size, same colour, uppercase label, no icon, no metadata. And two of the levels answered with a grid headed the same word. The breadcrumb was the only thing on screen that knew where you were, and it was the only thing.
No research budget, no feature authority, live customers.
Sole designer. Live throughout, with no migration window to hide behind. The feature set wasn't mine to change. So I didn't touch what Studio could do. I changed where its things live.
The problem was never what Studio could do. It was that its structure was invisible.
Context survives the switch.
Projects came to the front, because a project is what people actually work in; company became a label on the card and a filter in the header. Apps stopped being a destination: one overlay holds this project's apps and every project, switchable in place. Change project from inside Customers and you land in the other project's Customers. Same screen, different data.
A panel that also says what a project has.
Inactive apps stay in the grid, dimmed, so the panel answers “is this switched on here?” without a trip to settings. And a tile carries live state, so a timer running in one project stays legible from inside another.
The same model, applied across the product.
Detail views became structured records instead of rows. Team gained a real permission model where a tile used to sit with nothing behind it. And where a level genuinely matters, it is now named in words rather than inferred from which tiles are missing.
Designed in code, structure first.
Designed and built in Claude Code against a living design system, so it went to production instead of into a backlog waiting for dev capacity. Everything else inherits from the structure, which is why the structure shipped first.
Page transitions aren't implemented yet. The legacy build handled motion better than this one does, and that is still owed.
Time to a first published item. How often support is asked where to find something. And whether the switcher gets used at all, or people still route through the dashboard.