I built an app using AI. It actually worked.
I built an app called HeyMom using mostly AI and it actually worked out much better than I expected
What made me start this project?
I started working on a project for a friends mums’ birthday. He wanted me to create an app. The only issue was that I had never created an app before but he still seemed to have some weird level of faith in me that I would get it done. He then told me that he wants me to use AI for as much of it as possible which I thought was weird, I’ve never had someone tell me to use AI as much as I can, normally it’s the opposite and how I won’t learn if I use AI and everything else they say. I was intrigued and I had some spare time to work on side projects which I wanted to do for my CV for the longest time so I told him sure. The idea behind the app was simple - A child could upload an image to their parent once a day allowing the parent to know what they’re up to.
What was my goal for this project?
Initially, I thought the app was purely for his mums’ birthday so I put countless hours into feeding prompts into Cursor to get this app done. That was the goal. Get the app done by any means by this date. To be honest, I thought a week in, I was going to have to tell him no because I couldn’t imagine AI being able to basically fully make me an app and boy was I wrong. First day, I put a very simple prompt, something along the lines of make me this app with these features. Cursor took a minute to start spewing code at me and then it happened. It spewed code everywhere. Frontend? Did it. Backend? Did it. I was honestly in disbelief at how quickly it had completed some of the features that would have otherwise took me hours.
What challenges did I run into?
Don’t get me wrong, AI was also the most annoying thing in the world to work with. Some prompts it felt like it had 3 brains and with some other prompts, it felt like speaking to a toddler who didn’t even know what an app was. I’d ask for a change, it would claim that it was done, I’d check and what do you know? It hadn’t changed a single thing in relation to what I had asked of it. I’d then provide a more detailed prompt and same thing again. Eventually after wasting some credits making changes to nothing I had asked it to make changes to, it would get it done but sometimes it felt like I was there forever for a simple change.
What did I learn from this experiment?
Over the experiment, I learnt that AI was actually a solid tool, it had times where it made me feel like an incompetent software engineer. I would sit there considering how long it would take me to implement the feature that AI had completed in literally two minutes. It was honestly mind boggling how smart it could be. In about two weeks, I had created an app that used Flutter, Python, AWS S3 and Clerk without any knowledge on any of them except for Python. It blew my MIND. While it was great, it also showed some clear and obvious flaws. You want security vulnerabilities in your code? AI will be your best friend. It would send data from the backend to the frontend that would allow any user to access any image stored on the AWS servers which wasn’t great. Overall, I learnt that AI can be incredibly helpful but to be careful about the tasks that I assign to it in the future.
My plans from here
I’ll be continuing my work on FiveM (Grand Theft Auto V modding platform) but I’m also genuinely interested in the way AI works now. I want to see if it could make an application that is even harder than this without all the issues I ran into provided I gave it better prompts. I truly believe that AI could help software engineers work way faster than they are currently working and rather than seeing it as a threat, software engineers should be more open to using the tools they have.