Antigravity 2.0 ships as standalone agent-first developer platform

Antigravity 2.0 graduates from a feature inside the Gemini ecosystem into a first-class desktop application engineered around multi-agent orchestration. The release ships four production surfaces: a desktop client, a CLI for terminal-driven workflows, an SDK for embedding agents in third-party apps, and Managed Agents inside the Gemini API for hosted execution. Gemini Enterprise tier-ins bring SSO, audit trails, and compliance controls for regulated deployments.

Architecturally, Antigravity treats every dev task as a graph of agent invocations with shared state and tool access — closer to LangGraph or Anthropic's Claude Code than to a chatbot wrapper. The CLI in particular targets the developer cohort Claude Code has captured over the last six months: terminal-native engineers who want to script agent workflows rather than chat with an IDE assistant.

The move sets up a direct three-way collision between Antigravity, Claude Code, and OpenAI Codex/ChatGPT-for-developers. Ethan Mollick observed this week that Anthropic and OpenAI are converging on a unified experience, while Google's surfaces are diverging — Antigravity, Studio, Gemini app, and the consumer Gemini are all distinct products. That's a strategic bet: rather than forcing one tool for all developer audiences, Google is segmenting by workflow.

What to watch: pricing — Antigravity Managed Agents will charge by compute consumption, fitting Google's new consumption-based pricing pivot — and ecosystem traction. Claude Code has a year's head start with a fanatical user base; Antigravity needs both a flagship customer story and a credible answer to the developer question "why migrate?" The CLI and SDK suggest Google understands the answer must be "because it's portable" and "because the agent runtime is hosted."