Google recently took the developer world by surprise with the release of Gemini 3, their newest flagship AI model, and Antigravity, their agent-first AI-powered IDE. The announcement sparked immediate curiosity and discussion across the internet, as many developers wanted to understand how Google’s new AI editor is different from the ones they’re already familiar with and how it might influence the way software is built and maintained.

One of the most interesting parts of the announcement is how tightly Antigravity is connected to Gemini 3. Google is treating these releases as parts of a unified platform, with the IDE serving as a showcase for what the new model can do. It also signals that Google wants to compete directly with the top AI coding tools on the market today, and if they execute well on this integration, there’s a real possibility they could overtake them.

In this article, we’ll explore both releases from a developer’s perspective. We’ll briefly review what’s new in Gemini 3 and, more importantly, explore how Antigravity reshapes the development workflow. You’ll learn how it works, how to use it, its major features and current drawbacks, and how it’s different from other AI text editors like Cursor and GitHub Copilot.

🚀 Sign up for The Replay newsletter

The Replay is a weekly newsletter for dev and engineering leaders.

Delivered once a week, it’s your curated guide to the most important conversations around frontend dev, emerging AI tools, and the state of modern software.

What’s new in Gemini 3

Gemini 3 is a major advancement for Google’s AI ecosystem and for the LLM landscape as a whole. It builds on Gemini 2.5, retains the one-million-token context window, and introduces several important improvements. Below are some of the most notable ones:

• Core model upgrades: Gemini 3 delivers stronger logical reasoning, more reliable problem-solving, and better long-context understanding. It also handles multimodal inputs more smoothly, so it can work with text, code, images, audio, and video in a more consistent way than earlier versions.
• Generative interface: One of the most talked-about additions in Gemini 3 is its generative interface capability. This feature allows the model to create interactive visual interfaces directly from plain language descriptions. A user might request an explanation of how DNA and RNA interact or a simple simulation of a supply-chain flow, and the model produces a rendered, interactive interface instead of a static response.
• Nano-Banana Pro improvements: Google’s image-generation system also received a major upgrade. It now produces more realistic images and handles embedded text with greater clarity, which has been one of the hardest challenges for image models to solve. It is also more consistent when generating a series of related images and is less prone to hallucinations.

These updates give a clear picture of what’s new in Gemini 3. Now, let’s look at how Google built on these capabilities in its new AI-powered editor, Antigravity.

What is Antigravity?

Antigravity is Google’s new agent-first AI IDE, and it was created to demonstrate how Gemini 3 can use tools and reason through practical development tasks. Similar to all modern AI IDEs, Antigravity was also built on a fork of the open-source Visual Studio Code project, with the environment redesigned in a way that places AI agents at the center of the workflow rather than the text editor itself.

Some months back, Google acquired the Windsurf team, and this acquisition played an important role in Antigravity’s creation. Windsurf was known for its AI-augmented editor and for its early experimentation with agent-based development tools. After joining Google, the team brought much of its experience into the Antigravity project, and parts of its earlier work appear to have influenced the current design. The former Windsurf team now leads the engineering and product direction for Antigravity, guiding its shift toward a more agent-centered development experience.

Antigravity introduces a development environment that gives AI agents a central role in creating and managing projects. In addition to helping you generate code, it also helps you plan, execute, and monitor AI-driven actions across an entire project. Let’s look at some of its features below.

Customizable agentic workflow

Antigravity provides multiple agentic workflows that let you choose how much autonomy to give its AI coding agent. For example, you can grant complete autonomy, which gives the agent full control over actions like creating or deleting files or running terminal commands without asking for confirmation:

You can also opt for partial autonomy, where the agent handles code generation and routine commands but asks for your input before performing critical actions such as file deletion.

Dedicated agent manager

Antigravity includes an agent manager that allows you to run multiple agents simultaneously. For example, you can run Agent A to add dark mode to your website while Agent B adds a feature to an existing codebase or builds an entirely separate app in parallel:

The manager provides a clear overview of what each agent is working on and how far along they are in their assigned tasks. Additionally, agents can interact with your web browser to automatically preview and run your application, verify whether the output meets your requirements, and iterate accordingly.

Nano Banana integration

Another interesting feature in Antigravity is its integration with Google’s Nano Banana generative image model. It can generate images on demand and use them directly in your project. This capability isn’t available directly in other AI IDEs and typically requires workarounds, such as connecting the IDE to an MCP.

Familiar AI-IDE features

Antigravity also includes the core features you’d expect from modern AI text editors, such as:

• Tab auto-complete and chat in code: Accept AI code suggestions with a single press of the Tab key, and chat with the AI directly from your code.
• **MCP integration: **The IDE supports MCP integration, which allows your editor to securely connect to your local tools and external services such as GitHub, Heroku, and Netlify.
• Multi-model selection: In addition to Gemini 3, Antigravity supports other popular models, such as Claude Sonnet and OpenAI’s open-source GPT-OSS model, with more models to be added soon.

Now that we’ve explored its features, let’s look at how you can get started using Antigravity.

How to use Antigravity

To use Antigravity, you’ll first need to download it from its official website. Once it’s downloaded and installed, you’ll also need to sign in with your Google account.

After signing in, you should immediately see the familiar VS Code interface, as shown in the image below. And if you’ve used other AI IDEs like Cursor, Windsurf, or GitHub Copilot, the rest of the interface should be easy to navigate as well:

To start a new project or update an existing one, use the conversation box in the right sidebar and describe what you want to build. You can also choose between two modes: Planning and Fast:

In Fast mode, the agent will carry out the requested task immediately. In planning mode, however, the agent will first create an implementation plan document, then complete the task based on that plan, and when it’s done, generate an overview walkthrough document describing the changes it made.

Hands-on experience

To put Antigravity to the test, I gave it a vague prompt and asked it to “create a picture puzzle app” using Planning mode. It began by generating an implementation plan document for my review and then proceeded with the task. While the agent is working, you can add comments directly to the plan to guide its decisions, and it will incorporate those adjustments as it continues generating:

The Nano Banana integration also came into play here. Since a puzzle app needs an image to start with, Antigravity inferred this automatically and generated a default picture using Google’s Nano Banana model. This was not something I explicitly asked for; it recognized the requirement from the prompt and handled it on its own:

Once it finished building the app, Antigravity generated a walkthrough document summarizing what it had done, the features it implemented, and recommended next steps to improve or extend the project.

Overall, the workflow feels similar to other AI IDEs, such as Cursor and GitHub Copilot, but the built-in image generation really stands out as an integrated experience.

Notable drawback

Antigravity is still in its early stages, and you might notice certain latency issues or the software becoming unresponsive after heavy usage. However, these shortcomings are expected at this early stage, and the team will likely address them soon.

How Antigravity compares with Cursor and GitHub Copilot

The AI-IDE landscape is getting increasingly competitive, and most developers naturally compare new tools to Cursor and GitHub Copilot, the two most widely adopted options today. Here’s how Antigravity stacks up against them:

Category	Antigravity	Cursor	GitHub Copilot
Core Philosophy	Agent-first IDE designed around autonomous workflows	AI-enhanced editor; strong planning tools	Autocomplete-first assistant integrated into VS Code
Agent Support	Full and partial autonomy, multi-agent manager, browser interaction	Limited agents; no dedicated manager	Minimal agentic behavior
Model Support	Gemini 3 (default), Claude Sonnet, and GPT-OSS	Multi-model: OpenAI, Anthropic, Google	Multi-model: OpenAI, Anthropic, Google, Grok
Unique Features	Built-in Nano Banana image generation, artifact tracking, browser-driven validation	Mature UX, fast execution, workspace-level edits	Best-in-class autocomplete, deep GitHub integration
MCP Integration	Yes	Yes	Partial

Antigravity is still in its early stages, but with its agent-centric workflow and deep Gemini 3 integration, it enters the AI-IDE space as a serious competitor.

Wrapping up

Right now, Cursor is still the one everyone looks at when they think of the top AI IDE. It’s fast and reliable; however, its pricing model is a frequent point of frustration among users, and that continues to be a major drawback.

Antigravity enters this space as a credible alternative. It has most of the same core features as Cursor and is powered by Gemini 3, a state-of-the-art model. On top of that, Google has something Cursor doesn’t. They have the compute and the resources to make things better and way cheaper if they decide to. If they iron out the early-stage issues, Antigravity could easily give Cursor a real run for its money.

That said, the market is big enough for everyone. More options just mean better tools, faster improvements, and more ways to build software the way you want. And that’s a win for all of us.

Thanks for reading!