I tried Antigravity and I don’t think I’ll go back. Maybe it’s because LLM models have improved so much in such a short time, or maybe it’s because Google has closed the gap with (and perhaps even surpassed) OpenAI. That’s why I replaced VS Code for developing my personal projects.
Programming With AI: A New Normal
If you remember, my first post here was about GitHub Copilot and its inline suggestions in VS Code. Shortly after mid-August, I disabled it because I was dissatisfied with how it worked. So why should I now prefer an IDE that offers the same service as its main feature?
Time flies, especially when it comes to information technology: it only took four months for me to change my mind. Actually, I have to say that Google simply implemented the changes I suggested to GitHub. In fact, you can ignore the suggestions by pressing the [esc] key. It wasn’t that hard after all!
Not that anyone at Big G read them, of course, but Gemini’s inline suggestions on Antigravity work pretty much as I had hypothesized. I must say that, compared to GPT, the results are better. I tried it with both JavaScript and Python, and I must admit that I received some useful suggestions.
The Issue of Extensions and Language
As you should know, Antigravity (by the way, I hate the name) is a fork of VS Code and is compatible with a number of features from Microsoft’s editor. However, it does not use the official registry of extensions: it uses one managed by the Eclipse Foundation called Open VSX Registry.
This does not necessarily mean that all extensions available for VS Code are also available for Antigravity. Although it is possible to import settings from the first editor to the second, for example, Italian localization is not included by default, but must be installed separately as an unofficial extension.
Many extensions are derived from the official ones, including language support, and you will generally find them with the prefix ms- in the publisher name. I prefer to keep the number of extensions installed to a minimum, but if you use them extensively, you may have to wait for them to become available.
What I Don’t Like and Would Change
Among the many useful options, unfortunately, there is one that stands out as particularly annoying. Antigravity has the bad habit of automatically changing the indentation of the code, even when the indentation written in the source code is the desired one. That’s not nice at all.
Of course, you can disable this feature, but it’s not as easy as it should be. The automatic change is present in several menus, and to be sure you don’t get any surprises, you have to search through them all, depending on how many extensions you have installed.
This feature, which many may find useful, becomes a problem especially when you contribute to open source projects. It has already been pointed out to me that I modified the structure of some lines of code written by someone else, without realizing it, just by opening the same file.
What I Like and Would Keep
I don’t know if it’s thanks to the MCP server, the permissions I’ve granted locally, or both, but when I copy a block of text from GitHub, I find the same structure as an inline suggestion on Antigravity. This is particularly useful when writing documentation — and I write it very often.
The most innovative part concerns the Agent, i.e., a tab on the right dedicated to interaction with generative AI. This allows you to chat in two modes: Planning and Fast. The first mode allows you to review changes before they are applied to the sources, while the second mode proceeds directly to saving.
Antigravity connects directly to your Google account, so you can share settings between different devices and take advantage of your paid plans. In addition to Gemini, Claude Sonnet, Opus, and GPT-OSS are also supported. An integrated MCP Store allows you to use your credentials to access various services.
Thoughts on Google and Generative AI
Google has gone all-in on generative artificial intelligence. Not that it was ever behind, mind you: before the advent of transformers, DeepMind was the benchmark for state-of-the-art technology. Gemini and Gemma raise the bar for LLM models. I still need to get Nano Banana into focus.
Antigravity is just the latest tool in a long line. Jules, Opal, etc. are still experiments, but they represent Google’s path to just as many uses for generative AI. Creating an IDE from scratch would have been foolish when you could create a fork of VS Code, and Google made the best choice.
I don’t like that they’ve further lowered the free tier limits on APIs, because trying out generative AI services involves higher costs than traditional software. It must be said that the competition does not offer any free tier at all, so Google still has a better policy. It’s even better if you use Gemma locally.
If you like, follow me on Bluesky and/or GitHub for more contents. I enjoy networking.
Top comments (0)