Google Antigravity is an ‘agent-first’ coding tool built for Gemini 3

Google just launched what it's calling "mission control" for artificial intelligence workers. The company's new Antigravity tool can deploy multiple AI agents simultaneously across different digital workspaces, each operating with surprising autonomy.

Unlike traditional coding assistants that offer suggestions, Antigravity's AI agents can directly control your computer's editor, terminal, and browser. They work independently while generating what Google calls "Artifacts" - digital breadcrumbs including task lists, screenshots, and browser recordings that let humans verify what the AI accomplished.

The tool offers two distinct modes: a familiar coding environment similar to existing AI assistants, and a more ambitious "Manager view" designed for orchestrating multiple agents at once. Users can leave feedback comments without interrupting the AI's workflow, and the agents reportedly learn from previous tasks.

Available now as a free public preview across major operating systems, Antigravity supports not just Google's own Gemini model, but also competitors like Claude and OpenAI. The company promises "generous rate limits" that refresh every five hours, suggesting confidence that few users will max out the system's capabilities.