Today, I declare that Le Steu AI is officially closed. It was fun while it lasted. But I want to focus on new ventures. So while Le Steu AI is closed, the landing page will still be there, the domain will be in use for future projects, but it won't be under Le Steu AI umbrella anymore.
Anyways, before closing down Le Steu AI, I finally finished with Robin Talks, an AI voice app for non-English speakers who want to speak English more confidently. Shout-out to my buddy, Sigsten Gustavii, who created Borel.one, an AI mobile app builder. I used it to build Robin Talks. Check it out.
But I think I'm through with mobile app development. Unlike web apps, which I have several now (still kicking), there's too many hurdles to it:
- Building with React Native is fine. It's everything surrounding it is a pain. Upgrading Expo SDK version, compiling without Expo Application Service (EAS). Can you imagine looking at Gradle the wrong way and it compiles the wrongly-sized APK, which cannot be installed on an Android phone? I know building the app on EAS exists, but it is limited by the number of free builds you can order, and I generally don't want to rely on third-parties whenever possible, especially in this age of AI, when you can swap out providers easily provided the new ones have docs.
- Publishing to Google Play. They request facial, address verification, then make you set up many configurations. In the end, I still have to wait for its review, and have to find 12 people to test the app in closed beta. By this time, my startup DearDish has already been planning to implement smart search for recipes. Stay tuned, btw.
- Don't forget, you upload AAB file to Google Play Console to publish, not APK.
I just don't know how everyone else builds their mobile app. Maybe they have more people to test than I do. At this point, learning Rust is probably easier than dealing with app publishing.
Another point I have is AI app builders. While I owe my start to Borel.one and Lovable, I can't say I like using them. They are cool to build something fast, an MVP, a prototype. And nowadays, I encourage people with creating things with Lovable because its UI library is really nice. But once you scale, you will run into limitations. Maybe their fees jack up. Maybe the spaghetti codebase prevents even your AI agent from untangling it. It's a price we pay for low-code, no-code solutions, I guess.