Google Antigravity

Filed by the Vibe Code Weekly desk · Updated Nov 12, 2026

The story

Antigravity is Google's attempt to make the agent, not the editor, the center of software work. Instead of one assistant beside your code, it gives you a family of tools (a desktop app, an IDE, a command-line client and a Python SDK) for dispatching and supervising several agents at once.

Antigravity starts from the agent rather than the editor. Google calls it an agentic development platform for the agent-first era, and its design assumes you will hand off whole tasks. Subagents can run long jobs in the background while others take on research, implementation or testing. The human role shifts toward reviewing results. Google describes reviewing work at the task level, with artifacts and verification outcomes showing what an agent did, and feedback you can attach to any surface or artifact to push the output from nearly done to finished.

Each surface fits a different working style. The desktop app supervises parallel local agents, sorts conversations into Projects, spans multiple workspaces and sends scheduled messages for recurring chores. The CLI is the keyboard-driven option: it runs autonomous coding agents, executes shell commands directly and keeps background subagents in view. The IDE is the full editor, with an agent manager, artifacts and deep knowledge of the codebase. The SDK exposes the same agent harness to Python, so you can prototype custom agents, script engineering tasks or run evaluations with little code.

Google's example projects lean toward hardware and multi-agent work. Robotics researchers were asked to build a self-playing pinball machine and a controller for an inverted pendulum. Another demo refined a flight tracker's design with Nano Banana, and a collaborative whiteboard gained features through several agents working in parallel. Recent posts on the Antigravity blog cover a plugin marketplace, voice mode, use for knowledge work and Gemini 4 models such as Argon, Carbon and Hydrogen.

Google positions Antigravity as an agentic development platform where the agent, not the editor, is the center of work. Its design assumes users hand off whole tasks and review results at the task level, with artifacts and verification outcomes showing what an agent did.

The good

One agent harness shared across app, IDE, CLI and Python SDK. Background subagents can split research, implementation and testing. Artifacts and verification outcomes give a reviewable record of agent actions. Scheduled messages handle routine work without manual prompting. Free to download for macOS, Windows, Linux and Googlebook.

The catch

Depends on a desktop installation. Local agents tie the workflow to your own machine. The SDK is Python-based. Four separate surfaces mean deciding which one fits each job.

Who should read on

Builders who want to delegate complete tasks to agents and review the outcome. Developers coordinating several agents on one codebase at the same time. People who work mostly from the shell and want agents there too. Engineers building their own agents or evaluation scripts in Python. Frontend teams offloading repetitive UX chores to agents that work in the browser.

Who can skip it

Anyone who wants a purely web-based tool with no download. Users looking only for inline code suggestions rather than delegated agent work.

The price

Available at no charge.

Reader mail