Hi Product Hunt community!

I have a genuine question for builders and product managers here about the necessity of native iOS/Android apps for simple utility tools.

Years ago, if you wanted to build a smooth, high-performance tool like a video compressor, photo editor, or drawing board you had no choice but to build a native mobile app. Web apps were simply too slow and lacked access to heavy device processing.

But with WebAssembly (WASM), WebGPU, and modern browser engines, we can compile heavy native libraries and run them directly in the browser at near-native speeds.

I tried a lot of ways to work with AI tools and most of them made a mess. Here is the one that finally stuck, and I am curious if anyone else landed in the same place.

I build small. One piece at a time, and I run it before I move on. I never let the model build two related things at once, because when something breaks I want to know exactly what broke. I keep the context tight so the model is not guessing about code that does not exist. And I read everything before I trust it.

Before I let a model build something now, I ask myself one question. If this tool disappeared tomorrow, could I rebuild this myself?

The answer does not have to be yes. Some things I build are genuinely throwaway, and that is fine. But I need to know which kind I am building before I start, because the two projects need completely different decisions.

I tested my vibecoded project for two days before showing it to anyone. Clicked everything, tried every button, felt confident.

The first person who used it broke it in four minutes.

Set a bunch of strict rules for myself when starting out about keeping things simple and never taking quick shortcuts.

A couple years in half of those rules are gone because real life hit and decisions got way more complicated.

How do you tell if you're just adjusting to real business needs or actually losing your original vision?

I did not realize how much of my job was taste until I started generating code fast.

When you can produce ten versions of a feature in an afternoon, the bottleneck moves. It is no longer how fast you can build. It is how well you can tell which version is good. And that skill lives completely outside the tool.

I have noticed a split in myself that I am not proud of. I read the code I wrote by hand. I do not read the code the model writes for me. I just look at the result and decide if it seems right.

That is fine until it is not. And when it is not, I am in a stranger's house with no idea where the light switches are.

There is a specific moment in vibecoding that nobody warns you about. The demo works. It does the thing. You show someone and they say wow.

Then they actually use it, and it falls apart in ways the demo never could.

I kept convincing myself I was fine because tasks were still getting shipped but taking calls on weekends and working past dinner slowly drained all my energy.

I didn't need a long vacation I just needed to stop letting work leak into my evenings.

Taking a full step back and enforcing a hard no work rule after a specific hour made a massive difference in my sleep and overall focus within a month.

How do you keep small work exceptions from turning into permanent bad habits?