Antigravity: Google's Agentic Code Editor
The Team Behind Antigravity
Antigravity marks the first release from a team that originated at Windsurf. After selling non-exclusive IP rights, the founding members joined Google and built this product on top of that foundation. The result is an editor that feels familiar if you have used VS Code, Cursor, or similar forks, but introduces several new abstractions for agent interaction and testing.
Getting Started
Antigravity is currently in public preview and available to try for free. Given the attention it has received, expect rate limits during this phase. The interface opens to an agent manager that serves as your central coordination hub.
The agent manager contains an inbox where you can spawn different agents and see which ones require attention. If you work across multiple workspaces or projects, you can coordinate everything from this view. When you are ready to dive into code, press Command+E or click the editor button in the top-right corner.
Once you add a directory, it appears on the left side. A dropdown lets you toggle between workspaces and start new conversation threads for each agent. You can add images to your prompts and use @mentions as you would expect.
Models and Modes
Antigravity offers two distinct modes: planning mode and fast mode. The model selection includes Gemini 3 Pro (high and low configurations), Claude Sonnet 4.5, and surprisingly, GPT-OSS 120. The first two represent some of the best models available. The inclusion of OpenAI's open-source model, while notable, is an odd choice given its limited adoption in coding contexts. Notably absent is GPT-5, which is understandable for competitive reasons.
Fast Mode in Action
When you submit a request in fast mode, the agent immediately begins working through the task. You can watch it spawn work, create files, and build out the application in real-time.
One standout feature is automatic testing. Without any prompting, Antigravity opens your application in a preview and begins testing it. The agent navigates through the interface, clicks buttons, scrolls, and validates functionality on your behalf.
This browser automation shows you exactly what is happening: mouse movements, hover states, button clicks, and scroll actions. The agent reasons between each step, explaining what it is doing and why. This level of integrated testing is rare in local development tools. While Devon and Emergent Labs offer similar capabilities, this is the first time such thorough automated testing has been built directly into a mainstream IDE interface.
Planning Mode for Complex Projects
For more involved work, planning mode changes the workflow. Instead of immediately executing, the agent develops a structured plan first. You can review this plan and leave comments on individual tasks or planning stages before execution begins.
This creates additional surface area for interaction. You can skip steps, modify requirements, or provide feedback on specific parts of the plan. The comment system also works with images you pass in, letting you give precise visual feedback.
Integrated Image Generation
Antigravity incorporates Nano Banana directly into the product. You can generate images within the same interface where you build applications. For example, you might generate a reference image of a plant store landing page with specific styling requirements, then ask the agent to build a Next.js application based on that visual reference.
The image generation is currently rate-limited, but the integration points toward a future where visual design and code generation happen in the same workflow.
The IDE Experience
Opening the full editor reveals an environment that will feel familiar to VS Code or Cursor users. Your agent conversation sits on the right side, and you can continue sending edits and refinements just as you did with the initial prompt.
The ability to hop between the agent manager, preview mode, and full IDE creates a flexible workflow. You can start with a high-level request, watch the agent build and test the application, then drop into the editor for fine-tuning.
Video Context and Future Possibilities
Gemini 3 Pro supports video input, which opens interesting possibilities for future workflows. Feeding video context directly into the agent could enable new forms of interaction, such as recording a bug and having the agent diagnose it from the footage, or capturing a design walkthrough and translating it into implementation tasks.
Bottom Line
Antigravity brings together several capabilities that were previously scattered across different tools: multi-agent management, automatic browser testing, integrated image generation, and structured planning workflows. The VS Code foundation means developers can adopt it without learning an entirely new environment, while the agent-centric features push beyond what existing AI-powered editors offer.
For developers already using AI coding assistants, the automated testing and planning mode alone justify exploring the preview. The question remains how Google will price these capabilities once the preview period ends.
