What to know

  • Google Antigravity is an agent-first coding tool powered by Gemini 3 Pro.
  • It gives AI agents direct access to the editor, terminal, and browser to autonomously write and test code.
  • Antigravity features two interfaces: a traditional IDE editor and a manager view for controlling multiple agents.
  • The platform offers transparent, recorded artifacts like task lists and screen recordings to verify agent actions.

Among a slew of announcements, which included the Gemini 3 Pro, Google has also unveiled a new development tool - Antigravity.

Built on the Gemini 3 model, Google Antigravity is an agent-first AI coding platform that equips AI agents with direct control over code editors, terminals, and browsers. This allows Antigravity to independently plan, code, test, debug, and verify software projects.

How (Google) Antigravity Works

Google Antigravity allows AI agents to independently execute software development lifecycles. Instead of just generating code snippets on request, agents can plan complex applications, write multi-file codebases, run tests, debug issues, operate web browsers for verification, and deliver ready-to-review outputs.

The platform captures task progress with screenshots, browser recordings, and logs called "Artifacts," which developers can annotate and review for continuous feedback and control.

Antigravity runs on Windows, macOS, and Linux and builds on Microsoft's VS Code, inheriting compatibility with thousands of extensions. Users can choose between an Editor view for traditional code interaction or a Manager view to coordinate multiple AI agents working asynchronously on separate projects.

Antigravity is free to use in preview on major operating systems (Windows, macOS, Linux). There are generous rate limits refreshing every five hours, designed so only a small fraction of power users might reach these limits.