What can Google Antigravity do? Websites, apps, scripts, automation

한 줄 요약: Google Antigravity는 ‘에이전트 우선’ 개발 플랫폼으로 알려지지만, 처음 쓰는 사람은 웹·앱·개발자 전용인지 헷갈리기 쉽습니다 as an agent-first dev platform, yet many first-time users still wonder whether it is for websites, mobile apps, or only professional engineers. This article walks product surfaces → what agents can actually do → sensible everyday use cases → when not to delegate—and compares Antigravity with chatbots and AI IDEs.

What it is: IDE, desktop app, and CLI

Google Antigravity is Google’s agentic software development platform. The point is not another chat box—it is an agent that plans, uses tools, and ships reviewable artifacts inside your workspace. Official docs split the product into:

• Antigravity IDE — AI IDE with Tab completion and embedded agents, one workspace per window; parallel local agents and a browser agent.
• Antigravity App (2.0+) — Standalone desktop command center for async work, project-scoped permissions, scheduled tasks, Git worktrees.
• Antigravity CLI / SDK — Programmatic access for terminals and CI/CD.

Models are Gemini-first (e.g. Gemini 3.8 Flash, Gemini 3.1 Pro; subscription tiers may include Gemini 4 Argon depending on surface—see our Gemini 4 Argon and Google AI pricing notes).

Websites, apps, scripts, and automation

Editor, terminal, browser

1. Editor — Tab completion plus conversational edits.
2. Terminal — Dependencies, tests, scripts; approvals for commands outside the sandbox by default.
3. Browser — Read/act on pages for dashboards or your own sites.

The 2.0 app adds dynamic subagents, async task queues, and voice-to-prompt transcription. If you already wire data through MCP, Antigravity is an orchestration layer with its own harness—see MCP basics for comparison.

Six everyday-friendly use cases

1. Personal or small-business sites — Describe layout and copy; review in the browser.
2. Spreadsheet drudgery — Folder-scoped rename/summarize scripts.
3. Small fixes in an existing repo — You review diffs before merge.
4. Learning by doing — Plans in artifacts beat a wall of chat text.
5. Scheduled checks — Morning health pings with a Flash-class model for cost control.
6. Split with a cloud Mac — Code locally with Antigravity; compile/sign on a remote Mac when Xcode memory is tight.

When agents are a poor fit

• No repo, no acceptance criteria—agents spin on vague advice.
• Production secrets and regulated data need isolation, not “full machine” presets.
• High-end brand/creative work still needs human iteration.
• Unattended 24/7 operation needs monitoring and rollback—not the same as a desktop session.

vs Gemini chat, Cursor, Claude Code

Chat is for answers; Antigravity is for hands-on execution with artifacts. For monthly tool comparisons see Cursor vs Claude Code. Pick Antigravity when you want Google’s agent harness across editor, terminal, and browser.

Getting started

1. Install from antigravity.google—start with IDE or App, not everything at once.
2. Create a project with only the folders you trust.
3. Keep default permissions until you understand approval flows.
4. First task: verifiable and small (e.g. fix README and run npm test).

FAQ

How is this different from Gemini chat? Chat outputs text; Antigravity agents edit files, run commands, and drive a browser—with artifacts to verify.

Can non-coders use it? Yes for small sites and scripts if you can review diffs and approvals; otherwise start with chat.

Mobile apps? Agents can work inside an existing mobile codebase; store accounts and signing stay on you.

IDE vs App? IDE for synchronous pairing; App for async, multi-project, and scheduled work.