So, I installed AntiGravity because obviously I would like a proper AI integrated IDE with agentic capabilities...
To test this supposed super IDE I downloaded a project I had AI Studio iterate for me and had it refactor everything.
Since AI Studio is obsessed with making things in react and I detest react on all levels. Not to mention the absolute irritation that is typescript. I figured it would be a pretty decent real world test.
As you can see, the project is nasty with all the react code puking out html. It has no business doing that yet it seems to be the standard for developers these days. Surely a sad state of affairs.
Anyway, I didn't really touch on Antigravities agentic abilities too much but still got a pretty decent taste of how it can be to work with it.
I have not used claude code or cursor before but I think on a baseline level they would be similar. Basically I am able to tell the model, in this case GEMINI 3 , what I need it to do. It comes up with a plan and I change it or give the go ahead.
Then it iterates until everything is done. I do think it is much better at testing and iterating than other options available. Which is likely a baseline agent trait.
In most AI tools you have to give feedback, but in the more agent like or let's call it modular ones, the model basically follows a set of instructions and will continue to do so until it has checked off every item on the list.
Basically AI on a loop.
In that regard Antigravity is quite nice, and the fact that I can tell it to write in any language or framework I want.
I do like that it has an actual gui, albeit a butchered version of vscode. That is why I never tried claude code because I just don't like dumping files and such to the terminal. Maybe it is just something I need to get use to. For now I will stick to my noob gui interface though.
Since I have not setup any advanced workflows or agents I think I am not really seeing the worth of Antigravity vs just having an agent like copilot in vscode.
However I do prefer it being a separate program because I specifically remove AI from my vscode. Although I can't seem to add my own LLM's to Antigravity and have to rely on the pittiful free tier provided by the couople of LLM's it supports out the box. It does seem to be a great option for quick prototyping locally.
In fact I like Antigravity quite a bit as a local AI sandbox for making tedious changes or mvp projects that I might give other AI editors a look simply because I think it will be a while before Antigravity becomes truly powerful and user friendly.
Specifically user friendly. There just isn't much it provides in the way of making life simpler besides the AI running on a loop to finish a project plan. Sure that is nice but everything feels disconnected and even if you tell the AI to wait for approval it all feels out of your control and disconnected.