How to use Google Antigravity
Corbin Brown
2 min read
Google Antigravity is Google's entry in the AI coding editor race — a free development environment built around Gemini where the agent, not the editor, is the main character. You describe what you want; agents plan the work, write the code, and verify their own results (including opening your app in a browser and clicking through it). If Cursor feels like pair programming, Antigravity leans further toward management: you spend more time reviewing what agents did than watching them do it.
What's actually different about it
- The manager view. Alongside the normal editor there's a surface for running several agents at once and checking their status — closer to a team dashboard than a chat window. The parallel agents workflow is a first-class citizen here.
- Artifacts instead of trust-me. Agents produce plans before building and evidence after — task lists, screenshots, browser recordings of the feature working. You verify from the artifacts, which is exactly the review habit beginners should build anyway.
- The price. It's free during preview with generous Gemini limits, which makes it the cheapest way to try agent-first development on a real project.
Getting started, concretely
Download it, sign in with a Google account, open a folder (empty is fine), and give the agent a buildable first task — the same discipline as any vibe coding tool: one clear feature, not “build my startup.” Let it produce its plan, read the plan before approving, and check the evidence it hands back. The skills transfer completely: if you've prompted Cursor or Claude Code, you already know how to talk to this one, and the review muscles are identical.
FAQ
Antigravity vs Cursor — which should a beginner pick?
Either genuinely works. Cursor has the larger community and more tutorials (useful when stuck); Antigravity is free and pushes you toward the delegate-and-review workflow the industry is heading into. Trying both on the same project for a weekend settles it faster than any comparison article.
Is my code sent to Google?
The agent sends relevant project context to Gemini's servers to generate code, like every cloud-model tool. If that's unacceptable for your project, the alternative is local models — with a real capability tax. Same trade-off as Cursor with OpenAI or Anthropic models.