Angel Campa

Sixteen products in 430 days,
alongside a full-time job.

In May 2025 it felt like a gold rush. Everyone was saying that anyone could build an app now. That was not true then, and I am not sure it is true now. I started building anyway.

None of this code was typed by hand. Models wrote it, and I directed them: at least 5,039,037 lines, 116,830 test cases and 21,898 commits, each figure counted by the repository that produced it. Checking them is the skill, and it took a year to learn.

The core sample

Each product sits at the month it began. The width is how long it ran. Two seams tell the story: the thin slow one at the top, and the crowded one below it. Pick any band.

GatherGroveGeoLeapSkillLedgerCapVeriCAMAudit v2LextractGrantPipeKaiplanPebbleDeskFlorivaGavelhousePHIGuardFloriva WebVentora PlatformVentora SequencerVentora CRMDecision-making skills

Seventeen bands here. The headline counts sixteen, because the bottom one is still going.

The same thing as a list, by subject and by era.

01 · The first six months

It was slow, and it was buggy.

The bugs were stupid ones. Text running past the edge of its box. Things badly aligned. Happy paths that simply failed. It was not reliable. I found that out by opening the app and looking.

GatherGrove is the one I am still glad about from that stretch, and the reason is that I tested it by hand, over and over. That works. It does not scale past one person paying attention to one product.

02 · What changed

I stopped trying to catch things myself.

Writing the tests first was supposed to be the answer, and it is not. A model will code a bug straight through a test it wrote itself. I watched that happen often enough to stop reading a green suite as good news.

So the checking had to be built. Coverage gates that fail the commit rather than the conscience. A second engine kept alive to disagree with the one that shipped. Adversarial passes. And the one that still catches the most: an agent driving a real browser, using the product, taking screenshots and looking at them. That is the one I would keep if I could only keep one.

03 · How it ended

Then I shut it down.

On purpose, not by accident. Deploy scripts rewritten to refuse to bring the services back. A dated export of everything. Hostnames left resolving to nothing. The last one came off both app stores in August 2026, after a final release that made everything in it free.

I like to do things right. It felt awful.

17 products, one page eachThe workWhat each one did. What it ran on. Every number with the command that produced it.What it all adds up toThe ledgerLines, tests and commits across 17 products. And what each count covered.How you can check any of itThe methodEvery gate a change had to pass. And the one that caught more than all the others.

I can build anything. To build a business you have to love the industry you are in, or just stay an engineer.