An honest log of working with AI

This is an honest log of a day working with AI.

8AM. My new MacBook arrives. Quick setup, then get to work!

ChatGPT generated a checklist of tools for me to install.
“Since you do ecommerce, AI, and backend development, here’s what I suggest: … “. Okay …

Looked perfect on paper. Like my dating profile from 2009.

I raced through the instructions like a hacker in a Hollywood movie. Until I hit a wall with asdf (the package manager).

20 minutes of debugging later: turns out ChatGPT had outdated docs. Classic. I tag in Claude, who actually gives me the right solution. Nice save, but at this point, manually installing everything would’ve been faster.

Finally got that sorted and jumped into Varient.

First roadblock: cryptic error on startup. At least ChatGPT quickly spotted I was missing ENV variables. Small win. Look at me, living in the future!

Project running ✓
Frontend talks to backend. Backend talks to OpenAI ✓
UI crashes everywhere ✗
My confidence level ↘️

More strange IDE warnings. Both Claude and ChatGPT led me down the wrong path. It’s like having two very smart interns who are too confident.

I finally find the issue.

OpenAI’s “structured response” API returns non-deterministic JSON. Is this a joke? In human speak: you can request a specific format, but sometimes it gives you whatever it feels like. Not very different from asking Nomad (my dog) to do tricks.

I know it will get better. This is the worst AI will ever be.
But for now, don’t believe everything the hypeboiz are telling you.