A little while back I tried fully embracing the “I want it my way and I want it now!” AI mindset just to see what would happen. I wanted to see first hand if any of the hype held (spoiler: it didn’t), so I slapped together a handful of apps with LLMs to fulfill what I considered to be, at the time, niche needs that other apps didn’t do perfectly for me.
And out of the six apps I created for myself, you know how many of them I use today? Zero. Literally zero.
If you’ve ever worked on software professionally, you already know why. Stuff kinda sorta works, but not really. Things would keep breaking, like a lot. There were approximately 10,000 use cases I didn’t consider, so half the time I’d have to take a detour from what I was doing to go build it. And of course when I added those things, it would break like 20 other things. I also had to host these tools, and they’d go down regularly (weird, it’s almost like they’re not super stable!) API keys/access to services seemed to disconnect regularly and break. Oh yeah, and I have no idea how any of these were built so I’m sure they’re suuuuper secure. Basically, it sucked.
This is, of course, why all of us who’ve been doing this a while are so skeptical of “vibe coded” apps and of anything that was magically created “in a few hours”. The hard part of truly functional, well-oiled software is to serve far more than an audience of one with a barely-works-for-me version. All the actual work that takes weeks or months is where all the hard stuff gets figured out, built, and fine tuned. Conveniently overlooking all of those factors in a two hour AI slop session is your call, but let’s stop pretending any of it is shippable to real users.