Google Antigravity, a New Era in AI-Assisted Software Development (antigravity.google)

Google today unveiled Antigravity, an “agentic” development platform that rethinks the IDE for a world of autonomous coding agents. Built around Google’s Gemini 3 and integrating Anthropic’s Claude Sonnet 4.5 and an OpenAI GPT-OSS option, Antigravity combines a familiar AI-powered editor (tab completions, inline commands, side-panel agents) with a new agent-first “Manager” surface that orchestrates and observes multiple agents across workspaces. The product emphasizes task-level transparency—agents produce verifiable Artifacts (task lists, implementation plans, walkthroughs, screenshots, browser recordings) instead of raw tool-call logs—so users can track progress, validate outcomes, and give feedback without interrupting an agent’s run. Technically, Antigravity enables agents to operate across editor, terminal, and browser surfaces autonomously (e.g., write code, run localhost, and actuate the browser to test features), supports asynchronous interaction patterns and live feedback that’s incorporated into ongoing execution, and treats learning as a core primitive by adding actions to a shared knowledge base. Public preview is free for individuals (Mac/Windows/Linux) with model-usage rate limits refreshed every five hours and correlated to the amount of agent work. The launch signals a shift toward higher-level agent orchestration in development workflows—promising major productivity gains but also raising new priorities around verification, rate-limit management, and governance as agents take on more end-to-end software tasks.