Put myself in a whole world of hurt today.
I’m trusting AI far too much. My impatience at getting a “working thing” bit me, big time. That impatience causes me to farm out whole tasks to Claude Code, way before I have a good grasp on what I actually want to do…let alone how to do it.
I did a nice bit of re-factoring on my own this morning, which turned out perfectly fine. Tidied up error handling a bit and did some re-naming of my main error type, ApiError. I basked in the glow of that for awhile, the I thought, “let’s get password reset working”.
Famous last words.
I turned the whole thing over to the Claude web agent. Long story short, I let it build email verification with the lettre crate, forgetting the past experience that it’s not really what I want. It doesn’t actually deliver anything without an SMTP service to back it up. Rather than back away from the cliff, I decided to just press on, thinking that “oh, I’ll pick a service, won’t be a big deal”, so away I went merging everything.
Claude shat the bed. Not only did it make a poor choice, but it half-assed the changes, which I didn’t fully grasp until I pulled them into my trunk branch. Spent a hour trying to stanch the bleeding, to no avail. I don’t have the skills to fix the mess.
THIS is the core problem for me in using AI.
Anyhow, the mess deepened and I ended up nuking the whole thing from orbit and resetting against my production branch, which thankfully, was in a good state.
As a precaution, I’ve created a dev branch now, which will be my default for any further f**king around.
Now, I step back and figure how to just do things myself for the rest of the project.
Advent of Code is coming in a week…might even just take a break and actually, you know, focus on writing some content.
Comments