Google Antigravity

Google Antigravity is an agentic software-development platform from Google that reorganises programming tools around autonomous AI agents rather than around the code editor. First released in November 2025 alongside the Gemini 3 model family, it is free for individual developers and became one of the most widely adopted AI coding environments of 2026, with Google reporting millions of users of the original IDE by May 2026.^[1][2]

History

Antigravity launched in November 2025 as an "agent-first" development platform. It shipped with two surfaces: a familiar AI-powered editor and a Manager surface where agents could be spawned, orchestrated and monitored as they planned, executed and verified tasks across the editor, terminal and a controlled browser. Unlike earlier coding assistants that lived in a sidebar, Antigravity gave agents their own workspace, including the ability to produce and annotate artifacts such as implementation plans, screenshots and walkthroughs. At launch the platform offered a choice of models: Google's Gemini 3, Anthropic's Claude Sonnet 4.5 and OpenAI's GPT-OSS.^[1]

In May 2026 Google released Antigravity 2.0, a separate standalone desktop application with no IDE at all, built for synchronous and asynchronous interaction with fleets of agents. The release also expanded the platform into a family of tools: an Antigravity command-line interface, a software development kit, programmatic access to the Antigravity agent through the Gemini API, and an integration allowing projects to be exported from Google's AI Studio with their full agent conversation history.^[2][3][4]

Key concepts and technology

Antigravity's architecture centres on an agent harness that Google has co-developed with its Gemini model and training stack. Core capabilities include dynamic subagents that decompose tasks into parallel workstreams, asynchronous task management so long-running commands do not block the main agent, JSON hooks for governing agent behaviour, and projects that group folders with their own permissions rather than binding agents to single repositories. Scheduled Tasks allow agents to run on cron-style recurring timetables, and the platform introduced slash commands such as /goal, used to run an agent until a task is fully complete without intermediate prompting, and /browser, which steers agents toward browser automation.^[2]

Google positions the platform as evolving from assisting with code toward general knowledge work. In an enterprise configuration delivered through Google Cloud's Agent Platform, agent activity runs inside the customer's cloud boundary with Google Cloud privacy terms, and the company has published vertical feature bundles such as a Science Skills collection; in September 2026 the Antigravity SDK added support for local AI models.^[2][5][6]

Applications and impact

Antigravity competes directly with a cluster of agentic coding tools - Cursor, Windsurf, Claude Code, OpenAI's Codex and others - that are reshaping software teams' workflows. Google's May 2026 materials featured enterprise deployments at firms including Accenture and Deloitte, citing governed autonomous engineering workflows and agent-driven quality assurance.^[5] Because agents can execute commands, edit files and browse the web on a user's behalf, the platform's permission model and review flows are a central part of its design, and security researchers have highlighted the wider class of risks that autonomous coding agents carry in enterprise settings.^[2][5]

>See Also

• Cursor
• Windsurf
• Vibe Coding
• Claude Code
• Model Context Protocol
• Agentic AI