A new AI code editor has just appeared on the market, and this time from Google. It is none other than Antigravity, which is presented alongside the new Gemini 3 Pro model. In this post we will see what it is about and if it is really worth it.
Google presents Antigravity
Google presents Antigravity as, of course, a fork of Visual Studio Code, but to which they have made some improvements that honestly do not look bad. Is it really a Cursor killer?
Well, to put you in context, I will tell you that Antigravity for the moment is a totally free IDE, it does not cost money, not even if you wanted to pay.
As we can see on the web, it only lets us use the free version and in the future there will be a paid one. But we already know Google, obviously it will be paid, just like what happened with Jules which I brought to the blog at the time. Surely they will keep the free version but with limits just like now, and the pro version that is probably included in their Google AI Pro plan.
But leaving theories aside, let’s go with what we know, and that is that right now it allows us to use its new model Gemini 3 Pro totally free, and apart from that Claude Sonnet 4.5.
Hard blow for the rest of code editors…
Basic functionalities: What we already know
At the level of programming experience, what we have is the typical combination of smart autocomplete, side chat and natural language commands on the code. Nothing revolutionary so far, right? We have already seen this in Cursor, JetBrains, etc. But where Antigravity tries to differentiate itself is in the concept of “agents”.
The Agent Manager: Coordinating your AI team
Here comes one of the most curious parts: the Agent Manager.
Google sells it as an “agent first” experience, meaning that your way of working is not so much “me and my editor”, but “me coordinating a small team of AI agents”.
You have a kind of unified “mission control”, where you see which agents are active, what tasks they are working on and what they are modifying in your project. From there you can launch several agents at once on any workspace, and they all talk to Gemini 3 Pro behind the scenes. The idea is that you pose a high-level task (“migrate this module to architecture X”, “add authentication with OAuth”, etc.) and let the agents execute the steps.
Execution modes: Planning vs Fast
For that they have set up two task execution modes:
1. Planning Mode
It is the “serious” mode: ideal for complex tasks. Here Antigravity investigates, proposes a detailed plan, shows you that plan and does nothing until you approve it. It takes longer, but it takes it calmly.
2. Fast Mode
This is the “don’t warm my head, do it” mode.
Launch an agent faster, with less ceremony, for specific changes or things you don’t want to overthink.
In both cases there is one thing that I liked quite a bit: it always builds a plan and shows it to you. You can read it, mark things that don’t fit and literally tell it “iterate it again before touching anything”.
Cross-Surface Agents: Total environment control
Another very powerful part is how they handle what they call cross-surface agents, which basically means that agents not only see the editor, but also the terminal and the browser.
That is, you can install the browser control agent of Antigravity, and from there the system can open your app in the browser, launch QA flows, monitor dashboards and use that info as real-time feedback. Instead of “send me a screenshot” or “pass me the copied error”, the agent itself sees what is happening in the browser, executes actions, reloads, and uses that as context. It’s like having a junior who also controls Chrome for you.
The interface: Artifacts and total control
In the interface you have an area where you see artifacts (basically “work pieces” that the agent is generating like a plan, a new file, a script, etc.), verification results and the conversations between agents.
The interesting thing is that you can inject feedback at any point: you can comment on a specific line of code, highlight a section and say “improve only this part”, or ask them to prototype a new feature based on a specific artifact.
Then you have a change review section, where you see all the diffs that are going to be applied, even if they affect multiple files. And again, you decide whether it is applied or not. This, for me, is the key line: Antigravity is not about letting the AI go and do whatever it wants, it goes about you maintaining control, but delegating the heavy lifting.
Installation and initial configuration
The installation is quite simple. It welcomes you, lets you import the configuration from Cursor or Visual Studio Code (rascals), choose appearance and here very important it asks us what type of agent we want.
Agent types
I agree with them, the best is the recommended one: Agent Assisted, since it has automatic executions and the agent makes the decisions. But if that scares you, you can choose Review Driven and that needs your confirmation for each action, here it is up to your taste.
It lets you have the same shortcuts as in VIM, I leave it normal. And a command line that instead of code . is agy .
Important to log in with Google, and ready we would have Antigravity installed.
Browser extension
What I see when I open it is that it asks you to install its Chrome extension, because that is how it is going to work with the frontends it generates. Really here applause for Google. And then we see customizations of rules and workflows.
MCP Marketplace
Apart from that I see like a kind of marketplace for official MCPs that seems brutal to me, in addition to the typical custom configuration to add MCPs of course. If you don’t know what MCPs are, I have a post where I explain it in detail.
Testing Antigravity: My real experience
Well, enough chatter, I started testing it with what I usually always do and that is asking for a Python API.
The planning process
I have to tell you one thing: I love that it does this. It has generated a file with the steps it is going to follow to do it, and of course we can review it and if we don’t like something put a comment on it. It is brutal honestly I love it.
And nothing, I gave it to generate and… I like it, I like it a lot. I expected it to be slow like Visual Studio Code with Copilot, but no way, it went fine, it did what I asked, which wasn’t complicated either and honestly, I see it very much at the level of Cursor.
Testing parallel agents
On the other hand not everything was going to be good, I tried the agents to do parallel tasks, among them add auth, database connection and add a new model and… to my surprise it could only do one, in fact the easiest one, adding a new model.
The other two failed due to capacity, since obviously being free it is limited, but I don’t see any retry button or anything like that. It is the only thing that would put me back, since if you want to work on something serious this would pose many problems… but well let’s think it’s a thing of the first days and surely with time it will be solved.
Is it a Cursor killer or not?
I would tell you that, as of today, it is not as simple as “yes” or “no”:
The good
On the one hand, having Gemini 3 Pro and Claude Sonnet 4.5 free inside the IDE is a punch on the table. That agents can coordinate editor, terminal and browser is very powerful at the level of real workflow. And the approach of “I show you the plan, I show you the changes I’m going to make and you decide” seems much healthier to me than the typical crazy autocomplete.
The not so good
On the other hand, being free it has many limitations and saturation failures. Honestly I would prefer to pay a small fee so that the service is much better, because I think currently the free version can hardly compete with those other IDEs.
The future
But watch out, in the near future it is very likely that it will. The competition (Cursor, JetBrains with Junie, etc.) is not going to stand still and that is the good thing about this sector. Also, Google has the resources to do this really well.
And as I always say, however good the agents are, if you don’t have technical criteria, the one who breaks the project is still you.
Conclusion
In any case, what is clear is that Antigravity raises the bar of what we expect from an AI editor: it is no longer just that “it helps you write code”, it is that it poses itself as an orchestrator of agents that understand your project, your environment and even your browser.
My recommendation
Sign up and try it because it is free and has potential, but don’t use it for production yet since limits and stability are an issue. Stay tuned because Google has the resources to do this very well. And above all, maintain your criteria, tools are just that, tools.
That has been all. If you liked it and want me to bring more analysis of this style, let me know in the comments. See you in the next post.
And remember, AI is not going to replace programmers, but programmers who use AI are going to replace those who don’t use it! 🚀
What do you think?
Leave your opinion, question or suggestion. Comments are synced with GitHub Discussions .