Developers watching Google’s latest Gemini 3 announcements got an extra surprise: Antigravity, a coding environment built around autonomous agents rather than traditional prompts. The tool gives AI direct access to the editor, terminal, and even a browser, turning the system into something more like a junior teammate than a passive assistant.

What separates Antigravity from most coding copilots is its ability to document its own workflow. As each agent works, it generates Artifacts — structured notes, screenshots, test outputs, and browser recordings. These snapshots act as a kind of transparent audit trail, making it easier to trust the steps behind the code instead of blindly accepting a result.

Another standout feature is the dual-view layout. The familiar Editor view looks like any modern IDE, with a side-panel agent offering suggestions. But the new Manager view feels almost futuristic, letting users oversee multiple agents working simultaneously across different tasks. It’s a shift toward orchestration rather than simple input-output interaction.

Google says early testers appreciate the ability to comment directly on Artifacts without interrupting the work cycle. Combined with the promise that agents can “learn from past tasks,” Antigravity may mark the first big step toward AI tools that feel truly persistent. The preview is available now on all major desktops, with free access and generous rate limits.

Read more