I’m a QA Tester. AI wrote the code, but my QA skills actually shipped my first app.

For over 7 years, I have worked as a QA tester. Every day I find bugs in big apps. I always wanted to build my own app, but I couldn't code from scratch.

Then AI appeared. People online say you can build a startup in a weekend with it. That is a lie. AI is like a very fast junior developer. It writes code quickly, but it makes stupid mistakes. Here is the reality of building my first app, Lupi (a local-first subscription tracker).

1. The Spaghetti Code and Bad Math
   AI generated my screens fast, but it did not understand components. Every button and header was a new piece of code. To change one margin, I had to edit 15 different files. Also, AI is terrible at calendar math. It broke my weekly payment cycles and forgot that February sometimes has 29 days. I had to fix the logic myself because financial apps that calculate things incorrectly would be dead in a day after release.

2. The Bureaucracy Hell
   AI is blind outside the code editor. It cannot help you with the Google Play Console, Google Cloud, and RevenueCat. Connecting product IDs across these three platforms for my PRO plans (Monthly, Yearly, Lifetime) was a nightmare. This is where most "AI weekend projects" die.

3. State Management Fail
   Because I am a QA, I wanted Lupi to be perfect. I used AI to write 778 tests in Jest. At first, I was happy because all tests were green (PASS). But I quickly found out that many tests were "blind." The AI wrote code that returned a PASS result, but it did not actually check the logic. It was just checking if the code exists, not if it works. I found a major bug that the AI tests missed: If a user canceled their PRO subscription, the app still gave them PRO access. The user only lost access if they force-killed the app. AI is good at writing syntax, but it is still poor at covering complex app states and business logic with tests.

4. The Useless Closed Beta (and my Copycat)
   In Poland, Google Play forces you to get at least 12 testers. I found 18 random testers on Facebook groups. It was a waste of time. I got zero feedback from them. But it gets worse: one of those testers actually stole my UI and logo and published a copycat app on the App Store while I was still in beta. I was so angry. But then I realized: if someone steals your UI before you even launch, your idea must be good, and still, my UI looked much better :)

5. Why I built Lupi
   I built Lupi because app stores do not remind you when your free trials end. Why? Because they make money when you forget. Lupi tracks your subscriptions, expenses and reminds you before you get charged. It works 100% offline with a local database. AI helped me start, but the experience of QA helped me finish.

If you are interested, check out Lupi on iOS and Android. Try to break it and let me know what I missed!