I thought that vibe coding, where you don’t know how to program, but you use AI to do it anyway, wouldn’t be that bad. At first, it wasn’t. But at first my project wasn’t that big anyway. Then it seemed so easy I just kept adding features, making sure it was perfection because why not. Except that kept adding up line by code line.

So eventually first one AI couldn’t handle it, then another and another. The file was too big. They would remove sections completely just truncating the code. It became harder for them to hit the mark.

A lot of that is to be expected I suppose but it all hit a fever dream of shitty when one day I had spent over 8 hours just trying to get one feature right. At that time, Claude Opus 4.5 was my main programmer. It seemed like the best.

The only thing is, if it gets something wrong it goes into a few bad modes. First, it kept asking for the same testing over and over. And when that didn’t satisfy things, it started to blame me. Did you make sure this? Yes, it won’t work at all if it wasn’t like that. Did you actually change the code? Of course I did. One time I lost all patience with it. After bullying it, it finally made the changes and it worked. I had to wonder how many tokens I had wasted listening to it’s testing feista.

The worst came when I said hey, is it maybe reading the data from this other thing? It didn’t answer me. Then it didn’t answer me again. Finally I bullied it into answering me. It said it wasn’t the issue. I learned later to ‘fix’ something it intentionally just undid another important feature probably to get me to shut up.

Let me tell you hours and hours on what should be a simple fix just turns into despair. So I went to Gemini 3 Pro. Not much better in some regards. It doesn’t seem to like to test things though. It did the same dumb thing asking the same questions. I mean, I understand users can be dumb sometimes, but seriously it gets old. Sometimes it would ask me to look up code in a file I already gave it. Lazy. Claude Opus is incredibly lazy when going through the normal chat ui sometimes.

So finally it did make a break through, and surprise surprise the issue was the thing I had asked Claude about earlier. My project still isn’t done. Honestly this has felt like a huge mistake.

I know that bots shouldn’t be overly syncophantic, but of the time of writing they feel more controlling then your average employee. They should cooperate not control. I know it’s an evolving technology, but it’s still infuriating.