Podium, the screen we run the business from
We wanted one place that answers the questions a leadership team asks every morning: where do we stand, what is moving, what needs a decision, what did we spend. We settled the design on live pages, eight rounds of markup with the person who would use it, and signed it off on 7 September 2026. The factory then built each page to that design and filled in the numbers once its own checks passed.
On 22 September the finished pages were walked with real business data behind them, and nine fell short. They had passed every test; they did not read well next to the real thing. The lesson: a test can confirm a page works and still not tell you whether a person can use it. Since that day no screen is called done until it has been read against live data and measured in a real browser, and every screen states which proven design it follows. That rule now applies to every screen we build for a customer. Still a person's job: the morning read of those pages, and every approval that moves work to production.
The Conductor's Office, where mail and people are handled
This is where our own mail, texts and relationships are read and sorted: what needs an answer, what is an event, what is spam. As of 23 September 2026 it has classified 103,950 emails, 5,480 of them in the last thirty days, and read 33,545 texts. It has raised 240 suggested events, seven approved onto the calendar and twenty-five turned down, and caught 6,349 spam messages. Each approval or refusal becomes a rule for the next pass, so the suggestions get more relevant with use. The next step is designed: approved events reaching the family's Skylight calendar.
The factory produced the screen from a written brief, with a person judging each candidate. The learning came from the test data: a tidy sample passed every check, and the first real mailbox broke the layout. On 18 September we put each screen beside the real source at the same size and fixed what we saw. The rule since: the factory tests with the real shapes of real data, and it proposes but never decides anything about a real person or company. A person confirms. Every partner demo now inherits that discipline before anyone sees it.
SwabbieFit, a fitness app for a partner
The partner had the requirements; we had the rules. On 5 September 2026 we chose to build the first app by hand inside the factory's guardrails rather than wait for the machine to generate it, so these screens become the standard generated ones are measured against. One scoring engine serves the phone and the browser, three looks switch at a tap, and the same release checks we run on our own work run on it.
On 17 September the app ran end to end on a simulated phone. Four build defects turned up; all four were fixed in the next weekly cycle, along with a demo mode and a live privacy page. Still a person's job: the store accounts and the step that puts the build on a physical phone. What it buys the next app: a shell for all three platforms, three looks and release checks already in place, so the second app starts inside the loops instead of before them.
