Mobile July, 2025
N2Self — Hourly Task Tracker
When my child was born, my attitude toward my profession shifted — I began to approach it in a more practical way. We had a very specific problem: we needed an interval tracker to measure the time between feedings.
Sure, I could have set alarms or reminders for the whole day — but what if we planned to feed at 13:00 and ended up feeding at 13:10? That would mean editing all the upcoming alarms and reminders. Not exactly convenient.
I thought, “The app shouldn’t be that complicated — with GPT I can definitely build it.” But I quickly realized how much underlying logic such an app actually needs:
In the end, I built this “super simple” three-screen app in two weeks — working from early morning to late at night (and sometimes until morning), often arguing with my computer.
I used only GPT (4o, 4o-mini, and 4.1).
My choice was purely practical: I showed my design to GPT and asked it to create a comparison table of where and how much it would cost to build something similar. The table included FlutterFlow, Cursor, Play, Lovable, and a few others. We concluded that building it in Xcode would be the cheapest and most suitable option for me:
I didn’t like that Lovable builds web apps — meaning they aren’t fully native to the OS.
Through trial and error, I learned that GPT-4o is an excellent consultant: it tells you where to look, how to register, and where to put things. But in terms of code, it tends to be “creative” — constantly redesigning the UI — and needs extremely precise instructions. As a consultant, though, it was great: fast, clear, and it helped me get into Xcode quickly, organize the project, and connect fonts.
GPT-4o-mini was my main coding partner: we built layouts together, fixed bugs quickly, and it was very attentive to my design and requirements.
GPT-4.1 helped with code review, separating business logic from UI, and building the architecture. But it over-explained things — which slowed the process — so I only used it for analysis.
When my child was born, my attitude toward my profession shifted — I began to approach it in a more practical way. We had a very specific problem: we needed an interval tracker to measure the time between feedings.
Sure, I could have set alarms or reminders for the whole day — but what if we planned to feed at 13:00 and ended up feeding at 13:10? That would mean editing all the upcoming alarms and reminders. Not exactly convenient.
I thought, “The app shouldn’t be that complicated — with GPT I can definitely build it.” But I quickly realized how much underlying logic such an app actually needs:
- What if the cycle starts from 00:00 to 08:00?
- What if the user sets a card with reminders in the morning for the evening? How should the timer behave?
- And so on.
In the end, I built this “super simple” three-screen app in two weeks — working from early morning to late at night (and sometimes until morning), often arguing with my computer.
I used only GPT (4o, 4o-mini, and 4.1).
My choice was purely practical: I showed my design to GPT and asked it to create a comparison table of where and how much it would cost to build something similar. The table included FlutterFlow, Cursor, Play, Lovable, and a few others. We concluded that building it in Xcode would be the cheapest and most suitable option for me:
- The app was for personal use only.
- It was iOS-only.
I didn’t like that Lovable builds web apps — meaning they aren’t fully native to the OS.
Through trial and error, I learned that GPT-4o is an excellent consultant: it tells you where to look, how to register, and where to put things. But in terms of code, it tends to be “creative” — constantly redesigning the UI — and needs extremely precise instructions. As a consultant, though, it was great: fast, clear, and it helped me get into Xcode quickly, organize the project, and connect fonts.
GPT-4o-mini was my main coding partner: we built layouts together, fixed bugs quickly, and it was very attentive to my design and requirements.
GPT-4.1 helped with code review, separating business logic from UI, and building the architecture. But it over-explained things — which slowed the process — so I only used it for analysis.
So yes, I didn’t experiment with too many tools. I had patience and treated GPT like a senior developer I had to clearly brief — and it showed me how to implement my ideas.
After finishing the first working version, I started using it myself, even making a home screen widget. I shared a screencast of my experience on social media, and many people recommended trying Cursor as a “magic bullet.”
The app had one bug and a few UI inaccuracies, so I decided to give Cursor a try.
Cursor is fast. With GPT, you drive and it gives you directions. With Cursor, it’s autopilot — smooth and quick. It immediately connected Git, committed everything, rewrote all files, and made them beautiful. I was so impressed that I built a release and published it to TestFlight — and even the App Store — without hesitation.
But when I started using the app, I found a couple of bugs. I asked Cursor to fix them, and it broke the layouts. I thought, “No problem, Cursor has everything saved — it can just roll back to yesterday’s version.” But instead, every rollback made the design and functionality drift further from the original. Eventually, it reverted to my very first buggy build.
Since Cursor has full desktop and file access, it moved a folder with duplicate layouts into the project folder — and then deleted them. In short: I lost control. We crashed.
Now I no longer have the source files of my latest build — but I still have the working version in the AppStore, which might be useful to others.