Slop.

I've been seeing myself get "worse" at using AI over the year. It started so subtle, just having AI help me write out the "boring parts," to then suddenly spiral into "I want to make an account management system where I can invite people to my squad, but have a lot of different squads, and these squads aren't aware of each other."

When prototyping, or at least out of sheer interest in what the models can do, that's totally fine. Using primarily Claude Code, I'd go into Plan Mode and see what it would come up with. It would explain the case to me, I'd read it thoroughly, be happy enough with it (read: I see nothing breaking), and then: go ahead, let's see!

But later in the year, something shifted. Being so amazed by Claude and Gemini and, in general, believing that even if they mess up I can just fix it later, I grew more and more complacent. I started thinking "yeah, not reading those 12 lines, I'm sure it's fine" and just made it do its thing.

Long story short: many of my creations still work and support plenty of users, so I'm sure they work. But I also see things being slower and more buggy. It works, but not in a way I'd be proud of.

This year the slop ends. I'm fully aware there's larger slop in AI that people are making (and noticeably so). But even if I end up not-writing-so-much-of-my-code, I still want to feel like I've thought through all of it. And I believe I let that slip a bit. Well, a lot.

This year will be better.