Google’s Antigravity: An ‘Agent-First’ Tool for Gemini 3
▼ Summary
– Google has launched Antigravity, a development tool that uses Gemini 3 Pro and other third-party models to support multiple agents with direct access to the editor, terminal, and browser.
– Antigravity produces Artifacts, task lists, plans, screenshots, and browser recordings, to help users verify the work completed and planned by the agents.
– The tool offers two main views: the Editor view for a familiar IDE experience and the Manager view for controlling multiple agents autonomously across workspaces.
– Users can provide feedback by commenting on specific Artifacts, and agents can learn from past work by retaining code snippets or task steps.
– Antigravity is available in a free public preview for Windows, macOS, and Linux with generous rate limits for Gemini 3 Pro and support for other models like Claude Sonnet 4.5 and GPT-OSS.
Google has unveiled Antigravity, a powerful new development tool built around the freshly announced Gemini 3 Pro model. This platform is engineered for what the company describes as an “agent-first future,” supporting multiple AI agents with direct access to essential developer tools like the editor, terminal, and browser. Antigravity also integrates third-party models, positioning it as a versatile environment for complex automated workflows.
A standout feature of the tool is its unique reporting system. As agents perform tasks, they generate detailed records known as Artifacts. These include comprehensive task lists, strategic plans, screenshots, and even browser recordings. The purpose of these Artifacts is to provide users with a transparent and verifiable account of both completed work and planned future actions. While the system does log all actions and external tool usage, Google emphasizes that Artifacts offer a more intuitive and user-friendly method for validation compared to sifting through raw logs of model activities.
The platform introduces two distinct operational views to accommodate different working styles. The default Editor view presents a familiar Integrated Development Environment, or IDE, experience. This layout, with an agent situated in a side panel, will feel comfortable to users of tools like Cursor or GitHub Copilot. For more complex projects, the Manager view acts as a central command hub. This perspective is designed for overseeing multiple agents simultaneously, enabling them to operate with greater autonomy. Google likens this function to a “mission control” center, capable of spawning, orchestrating, and monitoring numerous agents across various workspaces in parallel.
Enhancing the collaborative dynamic between users and AI, Antigravity incorporates more refined feedback mechanisms. Users can now leave comments directly on specific Artifacts. This allows developers to provide guidance and corrections without interrupting an agent’s ongoing workflow. Furthermore, the agents possess a learning capability, enabling them to retain and reuse valuable snippets of code or remember the precise steps needed to accomplish recurring tasks, thereby improving efficiency over time.
Antigravity is currently available as a free public preview for Windows, macOS, and Linux operating systems. It operates with what Google characterizes as “generous rate limits” for its flagship Gemini 3 Pro model, with these limits refreshing every five hours. The company notes that only a tiny percentage of power users are expected to encounter these restrictions. For added flexibility, the platform also supports other leading models, including Claude Sonnet 4.5 and OpenAI’s GPT-OSS.
(Source: The Verge)
