The first globe was one of the first things I made with AI, back in the summer when I was putting this website together. It went through at least a dozen versions, and I was fairly happy with where it landed. It showed eight cities that are meaningful to me, for personal or professional reasons. Lines connected them, and a red dot hopped from city to city, typing out each name.
That was four or five months ago, and I've built a lot of other things since. This week, while refreshing this site, I gave the globe a second try. Both versions are running below, exactly as they first went live.
The difference is obvious, and it comes down to two things.
Know your models. I built the first globe with an earlier version of Claude Opus and the second with Opus 5.5. Not all models are built the same, even within one company. Find out which ones are best at the kind of work you want to do (I keep a tracker of the major ones). Just as important, know which ones you actually have access to. Someone using an enterprise tool at work, someone on a free account and someone paying for a higher-tier plan are often working with different models, even inside the same app. Before you judge what AI can do, check which model you're using.
Guiding code is different from prompting. I don't write code. What I've gotten better at is guiding it. The first globe went through at least a dozen rounds too. The difference now is that I'm better at the rounds. This time every one was an editorial call. São Paulo goes first, because that's where I spent a decade. Orosei replaces Cagliari, because this site is named after it. Every city sits where it really is on the map. The line to San Francisco waits until the globe has crossed the Pacific. Every city's name fades out the same way before it turns out of view. None of that is coding. It's editing, and it's the same instinct I use on a script.
The contours of planet Earth haven't changed in thousands of years. What has changed are the models, and how I use them.