I may be completely off base here, and this is based on my personal experience. But I find myself using AI less and less for coding.

Obviously I still use it for things I am unsure of, or for optimizations.

But when it comes to building something that I have not fully defined yet, it sucks. Yes ofcourse thats my fault, how can it know what I want?

But thats the inherent problem with AI. We are trying to solve complex problems, but are slowly loosing the ability to communicate or even think of what we want.

I recently got off the waitlist for Kiro, and wow it was refreshing. A design first approach where the AI actually helps you figure out requirements, user stories, design styles, etc. Is absolutely fantastic. And as it started building it was incredbile.

However, I very soon ran out of tokens. And what was I left with? A huge chrome extention with 20+ files which I had no idea where to get started.

Some pros: It had written the code incredbily well, modular, and had tested everything itself until issues were resolved. If I were building a corperate or zero-fault system I would love this

Cons: For a hobby project, or trying to come up with an MVP that can pivot as its built. It really can’t

There is a beauty in simplicity, and for some reason AI can’t do that.

I find that often I will request something from AI, take a look at it, and be confused why it overcomplicated the whole situation.

Yes if you are trying to win fastest solve, or have the system be so optimized thats great. But its really not great when you need to have the human in the loop.

And that brings me to my greatest concern.

How in the world is there going to be a human in the loop when the AI writes code faster than we can read it?