LLMs Are Super Helpful (Except When They Aren’t) I think that currently, LLMs help a lot with writing and refining code, but it really depends on the use case. I tried to have Replit’s AI create a habit-tracking app with a specific stack setup for me in one go, and it really didn’t work all that great. The code it put together didn’t account for some corner cases and the whole app was build very fast but essential parts were wrong which would require me to do some serious research into why the controlling logic didn’t work. In other words, these AI coding helpers are awesome, but they’re not magic.

Coding on an iPad? Totally Doable One really positive thing, though, is that it massively improves the developer experience. With GitHub Codespaces, you can essentially do all your coding on a mobile device. The Replit app on iOS (especially on an iPad) is super intuitive and really well designed. I can totally imagine starting a new app and doing all the coding through prompts with this kind of setup.

App Maintenance: The Thief of Joy My biggest problem today is that I want my apps to be as maintainable as possible. Maintenance is the thief of all joy when developing apps. In the JavaScript ecosystem, your codebase can basically enslave you. Making sure that old packages still run is extremely painful. It’s really just about making the developer’s life easier, with no real benefit to the user. But with AI and Vanilla Code comes to the Rescue. I think this is where tools like Replit are extremely powerful because they allow you to write as much vanilla code as possible. These AI assistants are great at replacing those frameworks that are basically just template solutions to problems that take some extra legwork to solve. So basically, you avoid dependency hell and keep things simple.