How can I tell? Yes, the code looks quite a bit more polished. I'm not expert enough in JS to, e.g., know the cleanest method to inspect and modify the DOM, but I can look at code that does and tell if the approach it's using is sensible or not. Surely you've had the experience of a domain where you can evaluate the quality of the end product, even if you can't create a high quality product on your own?
Concretely in this case, I'd implemented an approach that used jQuery listeners to listen for DOM updates. Antigravity rewrote it to an approach that avoided the jQuery dependency entirely, using native MutationObservers. The code is sensible. It's noticeably more performant than the approach I crafted by hand. Antigravity allowed me to easily add a number of new features to my extension that I would have found tricky to add by hand. The UI looks quite a bit nicer than before I used AI tools to update it. Would these enhancements have been hard for an expert in Chrome extensions to implement? Probably not. But I'm not that expert, and AI coding tools allowed me to do them.
That was not actually the main thrust of my post, it's just a nice side benefit I've experienced. In the main domain where I use coding tools, at work, I work in languages where I'm quite a bit more proficient (Golang/Python). There, the quality of code that the AI tools generate is not better than I write by hand. The initial revisions are generally worse. But they're quite a bit faster than I write by hand, and if I iterate with the coding tools I can get to implementations that are as good as I would write by hand, and a lot faster.
I understand the bias towards skepticism. I have no particular dog in this fight, it doesn't bother me if you don't use these tools. But OP asked for peoples' experiences so I thought I'd share.