What Is Antigravity

Google Antigravity is an agentic development platform that reimagines the relationship between developer and tool. Instead of an editor that waits for you to type and offers completions, Antigravity lets you describe an outcome and hands it to one or more AI agents that plan the work, edit files across your project, run terminal commands, and drive a browser to verify the result. You become a reviewer and director rather than the person writing every line.

This lesson covers what Antigravity actually is, the three surfaces it ships with, its place in Google's developer ecosystem, and what you get at each tier.

What you'll learn

• Antigravity is a full agentic development platform -- IDE, CLI, and SDK -- not a chatbot sidebar
• It launched November 2025; version 2.0 shipped at Google I/O 2026 as a standalone desktop application
• Free tier available for individual developers; enterprise tier for teams on Google Cloud
• Gemini 3.5 Flash is the default agent model, with third-party model support available
• The platform's defining feature is agents that autonomously work across editor, terminal, and browser

The problem

A traditional AI coding assistant lives in a sidebar. You tell it what to change, it writes a suggestion. You accept or reject. This works for single-line edits but falls apart for the kind of work most developers actually do: multi-file features, cross-project refactors, end-to-end testing, dependency upgrades that touch 40 files. With a sidebar assistant, you become the bottleneck -- you have to prompt it for every file, every step, every verification. The tool is an assistant, not a worker.

Antigravity's thesis is that the next step is not a better assistant. It's an agent that owns the task.

Options & when to use each

Surface	What it is	When to use it	What it costs you
Antigravity IDE	The full desktop application with editor, agent manager, artifacts, and browser integration	Daily development work, multi-file features, delegated tasks where you want visual feedback	Must leave your existing editor; agent-first workflow has a learning curve
Antigravity CLI	Terminal agent, lightweight, headless-capable	CI/CD pipelines, servers, quick terminal tasks, migrating from Gemini CLI	No visual editor or artifact browser; terminal-only output
Antigravity SDK	Python SDK for embedding agents into your own tools	Platform teams building internal developer tooling, custom agent pipelines	Requires Python infrastructure; you manage deployment

Build it: install and authenticate

Install the desktop application directly from the official source, which supports macOS, Windows, and Linux. Once installed and authenticated, verify the environment by running a quick diagnostic query in a new standalone conversation.

# Download from antigravity.google/download
# The install is platform-specific:
#   macOS: Drag to Applications
#   Windows: Run the installer
#   Linux: Extract and run the AppImage or .deb

# After install, open Antigravity and sign in with your Google account.
# You'll land in the agent manager view -- this is the default starting surface.

# Verify the agent is working by opening a new conversation and asking:
# "List the files in the current directory and tell me what language they're in."

Antigravity 2.0 is a standalone application. It is not a VS Code extension and does not require VS Code installed. When you open it, you see the agent manager on the left side panel (conversations, projects, settings) and the editor surface on the right. The agent operates across both.

If you used Gemini CLI on Days 1 and 2, you will notice the agent's behavior is similar -- it runs shell commands, reads and edits files, and reports back. The difference is that Antigravity gives you a richer interface for managing multiple agents simultaneously and reviewing their output through Artifacts.

What goes wrong

Mistake	How you notice it	The fix
Treating the agent as autocomplete	You type one-line prompts and accept/reject single edits	Write task-level objectives: "Add OAuth login with tests," not "write an import statement"
Not reviewing the agent's plan	Agent makes architectural decisions you would not have made	Antigravity agents produce a plan before touching code. Review it. Reject it if the approach is wrong
Running too many agents on one codebase without coordination	Merge conflicts between agent branches	Use Projects to scope each agent to its own workspace or worktree. Let agents work in isolation and merge after review
Confusing Antigravity CLI with Gemini CLI	Commands or flags that worked in Gemini CLI do not work	Antigravity CLI is the successor. Migration docs are at the Antigravity blog. Some features have different syntax

Confirm it worked

Open Antigravity, create a new standalone conversation (no project), and ask:

"I have a project in this directory. Read the package manager file and tell me what dependencies it has, then suggest one dependency upgrade that is safe to apply."

The agent should:

1. Find and read your project's dependency file (package.json, requirements.txt, Cargo.toml, etc.)
2. List the dependencies it found
3. Suggest one specific upgrade with a reason

If it does all three, the agent is working across editor and terminal correctly.

Next: The IDE & Agent Manager