Google just released Antigravity 2.0, and it basically tears apart the traditional IDE. The agent manager that used to live inside an editor is now a standalone desktop app that runs on macOS, Windows, and Linux.
The agent's boundaries are no longer limited to a single repository. The new version replaces workspaces with projects, allowing a project to span multiple local folders, each with its own read/write permissions. For complex tasks, the main agent can now dynamically spawn sub-agents to work in parallel, preventing the main context from being clogged with noise. Compilation and testing — the slow stuff — all run asynchronously in the background.
Scheduled wake-ups are new to the desktop. Users can type /schedule to set one-shot timers or cron jobs, waking the agent to run scripts or inspect code at specific times.
New slash commands draw clear lines around how much autonomy Google trusts its agents with. /goal gives full control to pursue a task relentlessly. /grill-me forces the AI to ask clarifying questions before acting. /browser requires explicit permission before going online. Google admits that AI still can't reliably decide when to search the web.
Alongside the standalone desktop app, Google also released a CLI, SDK, and API. The company says the old Gemini CLI will eventually be fully migrated to Antigravity CLI.
