On November 18, Google announced the release of the Gemini 3 family of neural networks, claiming it’s the most advanced and powerful family of neural networks with advanced agent-based and reasoning capabilities. Big upgrade, obviously. But while the Gemini 3 Pro launch grabs headlines, the more important part isn’t the model. It’s Antigravity - Google’s push into agentic development.
Available in public preview from the day of its launch, Antigravity is Google’s own agentic development platform, which, according to Google, is built around an agent-based approach: instead of a single assistant sitting in a sidebar, Antigravity gives developers a whole fleet of agents that can read/write code, operate terminals, and literally click through browsers.
It’s clear that agentic coding isn’t new, but right now we’re constantly witnessing rapid iteration on the concept. Actually, Google already had multiple coding tools - Jules in IDEs, Gemini CLI, Code Assist. Antigravity isn’t replacing them; it's an attempt to unify the workflow around agents that can operate autonomously. It puts it directly up against Cursor, Claude Code, and Codex-style coding agents. Some observers note that Antigravity looks similar to Windsurf, which makes sense, as Google hired the Windsurf team, including CEO Varun Mohan, in July for $2.4B.
At first glance, Antigravity is yet another VS Code fork, Cursor clone. It's a desktop application you install that then signs in to your Google account and provides an IDE for agentic coding against their Gemini models. While on the surface, it might look like "just another AI editor," underneath the hood, it's actually a fair bit more interesting than that.
Antigravity combines what has become a pretty standard AI-centric IDE with a few innovations that, according to Google, evolve the IDE “towards an agent-first future.” Google is embracing a new development model where engineers don't write code but rather oversee agents who plan, sketch out features, launch the app, and click in the browser. In other words, Google wants developers to manage processes, not write every line of code. In a demo shown during the presentation, Antigravity built a flight tracking app, tested it, and provided a video report of its successful operation.
Here are those interesting new ideas in Antigravity:
1. Different modes for different tasks
Antigravity offers users two main workflows: a traditional VS Code-style editor and an agent manager dashboard. The standard Editor mode resembles familiar IDEs such as Cursor or GitHub Copilot, where the agent operates in a sidebar. The new Manager mode is designed for orchestrating processes. In this mode, the developer has a unique "mission control for creating, coordinating, and monitoring multiple agents," allowing them to manage multiple autonomous processes in different workspaces in parallel.
2. The concept of artifacts
Google clearly understands the trust problem associated with the agentic tools available today - either they show too much low-level noise or they hide everything, giving you mysterious output. Neither works and helps build trust. A key feature of the platform to address this issue is its reporting system. As Antigravity executes tasks, it generates so-called artifacts - detailed plans, task lists, screenshots, and even video recordings of browser activity. It allows developers to quickly verify the correctness of AI actions without having to analyze complex logs. As company representatives note, such visual evidence is easier for users to verify than digging through logs which competitors typically offer.
3. Learning and Feedback
An important feature is the system's ability to learn from past work. Antigravity stores prior snippets, patterns, and reasoning steps from past missions so future tasks can compound. They decided to pursue not “model fine-tuning”, but a persistent memory approach. Over time, the agent becomes more opinionated about how you build things. It is the part that, if it works, could be a real multiplier as it is exactly where agentic IDEs become exponentially more useful than assistants bolted onto editors. Google has also implemented the ability to leave comments on specific artifacts without interrupting the agent's workflow. The agent incorporates that feedback on the fly, so the workflow keeps moving instead of restarting every time something’s slightly off.
Availability and Model Choice
Antigravity is currently available for free on macOS, Windows, and Linux. You can try it for free, but first users are already complaining that Antigravity simply won't launch, it seems the workload is currently too high. And yes, it's free for a reason and not because Google is generous. Every click, every snippet, every project you try provides data they can learn from, improving Antigravity’s future performance, suggestions, or decision-making. Data is the currency here, not money. So it's best to consider which projects are worth testing the new IDE on, and which ones are best avoided.
When it comes to supported models, the native Gemini 3 isn’t the only model that developers can choose. In addition to the native Gemini 3 Pro, the platform supports Claude Sonnet 4.5 and OpenAI's GPT-OSS. Google and Anthropic recently announced a major cloud cooperation worth tens of billions of dollars, so maybe this was part of the deal. Google knows that winning this market isn’t about one model, it’s about owning the workflow.
The company also promises "generous limits" that are updated every five hours, assuring that only a "very small fraction of experienced users" will ever reach these limits.
Early Feedback
Early reactions to Google’s Antigravity IDE are mixed, but some patterns stand out.
The multi-agent workflow is genuinely interesting. You’re not just typing alongside a single AI - you can spin up coding, review, and refactor agents across a project. For larger repos, that changes how work gets done. Some developers are already seeing wins: long-standing bugs fixed, dormant projects revived, terminal commands automated, even basic UI testing handled automatically. Features like Chrome integration, inline comments, and video artifacts give visibility into what the AI is actually doing. Enterprise teams may find this tool especially valuable: Google’s security, compliance, and infrastructure are unmatched.
At the same time, stability is a real pain point. Crashes, incomplete edits, and reports of original files being deleted when reverting changes mean Git isn’t optional. Solo developers or small teams report that Cursor is faster, more reliable, and less restrictive. Right now, many see Antigravity as essentially a VS Code wrapper with AI sprinkled on top. The promise is there, i.e., the agents, multitasking, and artifact-based workflows, but the execution is rough. The takeaway: Antigravity shows what agentic IDEs can do. Today, it’s mostly worth experimenting with rather than fully relying on.
Final Thoughts
We’re moving toward a world where developers spend less time typing and more time exercising judgment - deciding what to build, reviewing artifacts, enforcing quality, and guiding autonomous systems. Cursor did this first, making agentic coding practical in an IDE. Claude Code is close. Now Google is entering the arena with more compute, more money, and more patience than anyone else. And they’re not just competing, they’re trying to reset expectations for what an IDE can be. Winning in this market requires staying in the game long enough to outlast competitors. Google has more than enough resources to do exactly that.
FAQ
What is Antigravity?
Antigravity is Google’s agentic development platform where developers manage autonomous AI agents that can write code, run terminals, click through browsers, and report back as verifiable artifacts.
When was Antigravity launched?
Antigravity was launched on November 18, 2025, alongside the Gemini 3 family announcement.
How is Antigravity different from other AI coding tools like Cursor or Claude Code?
Antigravity offers a multi-agent workflow, artifact tracking, and real-time feedback, letting developers oversee AI processes instead of typing the code.
Which models does it support?
Antigravity supports Gemini 3 Pro (native), Claude Sonnet 4.5, and OpenAI’s GPT-OSS.
Which platforms is Antigravity available on?
It is available on macOS, Windows, and Linux in free public preview.
Is Antigravity reliable for production projects?
Early users report crashes, slow performance, and occasional file issues. Git is recommended to prevent data loss.
What are its main competitors?
Antigravity’s main competitors are Cursor, Claude Code, Codex, and Windsurf.
