HyperAIHyperAI

Command Palette

Search for a command to run...

Google Unveils Antigravity: Agent-First Coding Tool Powered by Gemini 3 Pro with Multi-Agent Management and Verifiable Artifacts

Alongside the launch of Gemini 3 Pro, Google has unveiled Antigravity, a new coding tool designed for an “agent-first” future. Built to leverage Gemini 3 Pro alongside third-party models like Claude Sonnet 4.5 and OpenAI’s GPT-OSS, Antigravity enables AI agents to interact directly with the code editor, terminal, and browser—giving them deeper control over development workflows. A standout feature of Antigravity is its ability to generate Artifacts—structured outputs such as task lists, execution plans, screenshots, and browser recordings—that provide transparent, verifiable proof of what an agent has done and intends to do. Unlike raw logs of model actions and tool calls, Artifacts are designed to be intuitive and easy for developers to review, enhancing trust and accountability in AI-driven development. Antigravity introduces two distinct user views to support different workflows. The default Editor view delivers a familiar Integrated Development Environment (IDE) experience, similar to tools like Cursor and GitHub Copilot, with an AI agent operating in a side panel. The new Manager view, however, is built for advanced users who want to orchestrate multiple agents simultaneously. It functions like “mission control,” allowing developers to spawn, manage, and monitor several agents across different workspaces in parallel. Feedback is also more seamless in Antigravity. Users can leave comments directly on specific Artifacts without interrupting the agent’s workflow, enabling iterative refinement without context loss. Additionally, agents can learn from prior tasks by retaining useful code snippets or procedural steps, improving efficiency over time. Currently available in public preview, Antigravity runs on Windows, macOS, and Linux. It’s free to use with generous rate limits for Gemini 3 Pro, which refresh every five hours. Google notes that only a very small fraction of power users are likely to exceed these limits. The tool supports multiple models, giving developers flexibility in choosing the AI best suited for their tasks.

Related Links