Google Antigravity - Final verdict

Goggle's Antigravity had some very interesting features which greatly enhanced my experience, but Gemini 3.1 pro is too far behind the rest of the competition. I loved the fact that I can actually go to the Implementation plan and make comments on specific parts of the text. My subscription also came with image and video generation models, which was a great thing to have. However, that's where its "good to have"s end.

To truly get your money's worth from Gemini you will have to repeatedly use /goal and /grill-me to ensure better quality outputs. Although Antigravity in no way pushes you towards these, nor does it tell you about these features if you ask Gemini itself. This feels like a huge, missed opportunity by Google. I mean you already have a LLM I am talking to, can't you atleast train it to teach me when I am doing bad prompts? Or atleast show a loading screen tip, or something! It is safe to assume that if I am a vibe-coder using Antigravity, I do not have the IQ to learn all this on my own. Even if I did, I DO NOT WANT TO, I am vibing here not try-harding to squeeze max performance out of Gemini.

The worst problem I had was the model itself. Gemini 3.1 pro is almost 6 months behind other flagship models (as of writing this article). This means it is terribly outdated in many aspects. There were quite a lot of times when I had to switch to Claude Opus 4.6 to unblock Gemini 3.1 pro from chasing its own tail.

With most of my planned projects failed, I am so glad to be exiting Antigravity.