Craft Still Matters
AI can build something, ship it, and market it, but it cannot know how an animation feels or know if it will bring delight to a user.
It’s week two since I got laid off with the rest of my team, so I’ve got time again for JABA. I decided to quit working on JABA a few months ago due to time, but now that I have the time I picked it up again. I use Cursor to build almost everything which has made me think a lot about how craft still matters and that people who understand how the code works or how the system works will be needed.
Here’s the animation I made for opening an account.
When you tap an account, the card should move to the top and the other cards should move away so the details expand in place. The first versions from the LLM did not do that. It opened a full screen cover instead. That gets the details on screen, but it feels wrong. The stack disappears, the transition is cheap, and it does not feel good to use.
I went back and forth with it three or four times before the expand felt right. A lot of that was not about tweaking animation values. It was about how the views were structured. The model kept putting the card in its own separate view, not sharing views properly, not using ternaries enough, and making too many views. Once the hierarchy was wrong, the in-place expand was not even possible. I had to tell it how to architect the views so the animation could happen the way it should.
That’s what I keep noticing with this stuff. An LLM will give you something that works and is technically correct based on skills or what it was trained on, but it can’t catch the look and feel of the animation or the architecture that’s needed to make the animation work properly. It just can’t “think” that way (at least not yet…) Someone still has to care about that.
I'm looking for my next thing, whether that's full-time, consulting, or a long-term contract. If you're building something where the details matter, I'd like to hear about it. Send me an email.