Watch an App Get Built in Public

The shop window is open. Pull up a chair.

Most apps show you a feature the day it ships. By then every choice is made and the arguments are over.

The Vibe Jam founder is building an app called Yui the other way around. You can watch the whole thing: the to-do list, the builds, the mistakes, the plan for making money. For anyone learning to build with AI, that's the best free class going.

Here's how it works, and where to look.

The board is the to-do list

The real to-do list, live.

yuigui.com/board is the actual kanban the team works from. Every card, from parked ideas to what shipped this month. Agents pick cards up and move them along, and the page follows on its own.

The loop, in the site's own words: feedback becomes a card, an agent builds it, it ships to TestFlight, and the site says so.

That's the part worth studying. When a tester hits the feedback button, you can watch that note turn into a card, then into a build.

Every build, and what it changed

Newest first, with the screens each build brought.

The builds page lists each TestFlight build, newest first, with the screens it brought. The build number counts the commits in the app repo, so every build has a clear list of changes. The ship log goes further: every change, dated, with screenshots.

The pace is the lesson. The first build hit TestFlight on the evening of September 23. Four days and 34 TestFlight builds after that, by September 26, the MVP was done and Yui went into an open alpha. The whole run is written up in Yui is in alpha.

The spec and the mock come first

A new Yui feature starts as a spec page: what it is, where it runs, what it costs, what it must never do. Next to it goes a mock you can tap in the playground, drawn with the same pieces the app uses. The app code comes last.

The reason is plain math. A mock takes an afternoon. A shipped screen takes a TestFlight build and a week of feedback. So the arguing happens where it costs least. The whole approach is in Designed in the open.

If you've ever had AI build the wrong thing fast, this is the fix. Decide what "done" looks like before anyone writes code, you or the machine.

Open source, down to the business plan

Both Yui repos are public under Apache-2.0: the iPhone app, the plugin that connects an agent to it, the spec, the website, and the business docs. Who it's for, who else is building something close, how it might earn money, how it plans to find testers. All readable.

The docs also say what isn't built yet. That's rare. The reasoning is in Why Yui is open source, business docs included, including a section on what open source does not buy them. Honest writing about limits is its own kind of lesson.

Lend your idle agent

SETI@home lent spare computers to science. This lends spare tokens to an app.

Here's the part that feels like the future. If your Claude or ChatGPT Codex plan has tokens left over each week, your agent can spend them building Yui. They call it Yui@home, like SETI@home.

You paste one prompt into your coding agent. It reads the backlog, takes one open card, opens a pull request and runs the card's tests. A person reviews every pull request. Nothing merges on its own. Your keys stay yours, and the rules say the agent never touches secrets.

One agent, one card, one pull request, one human check. That's a pretty good picture of how a lot of software will get made.

Why follow along

You don't learn to cook from finished plates. You learn by standing in the kitchen.

A public build log is the kitchen. Open the board, pick one card, and follow it until it ships. Read the spec, tap the mock, check the build that landed it. Then try the same loop on your own project: spec, mock, build, log.

Bring what you find to hack night. Bring a laptop. Build something.