Google Antigravity

Full review

Google Antigravity is an agent-first development environment where autonomous Gemini-based agents handle the full development lifecycle: planning, coding, testing, and refactoring across editor, terminal, and browser. With version 2.0, it grew from a browser-based IDE into a native desktop app, joined by a CLI tool and an SDK for custom agent workflows.

The 2.0 desktop app

The new desktop app was rewritten from scratch, and it shows. It looks and feels remarkably like the ChatGPT and Claude desktop apps: a clean chat-first surface with the development machinery behind it rather than in front of it. Moving off the browser also brought faster performance and proper access to local files, and you can now orchestrate several agents at once or schedule tasks that run in the background. Voice commands and integrations with Google AI Studio, Android, and Firebase round it out.

The agent-first approach

Traditional AI coding assistants respond to individual prompts. Antigravity’s agents work autonomously. Give them a goal and they break it down into subtasks, execute them across multiple tools (editor, terminal, browser), and produce reviewable artifacts like implementation plans, screenshots, and walkthroughs.

The model underneath

Antigravity runs on Gemini, currently 3.7 Flash. That’s a real trade-off. Flash lags the frontier models by a fair margin on genuinely hard problems, but it is fast, cheap, and performs well on simple-to-medium tasks — which is most of what day-to-day building actually is. If you need frontier-level reasoning, this is the wrong tool; if you need volume and speed, it’s a good one.

What I’ve built and fixed

In practice, Antigravity handles a few types of work well:

• Static websites: full site generation with responsive design — my favorite playground for experiments
• Games: interactive browser-based games with logic and graphics
• Browser extensions: Chrome extensions with manifest, content scripts, and popup UI
• Computer tasks: it interacts directly with the machine it runs on. It diagnosed and fixed network and power-management issues on mine, the kind of job you’d normally ask IT support to handle. One firm rule: do this on your own machine, never the company laptop (an agent changing IT-managed settings is not your call to make), and have it explain what it wants to change before you approve

Claude Code vs. Antigravity

Both are agentic tools, but they feel different. Claude Code is more flexible and runs stronger models, which matters for complex engineering. Antigravity is more visual and approachable, and since 2.0 the experience gap has narrowed a lot. The free tier makes it easy to add to your toolkit, though the quotas have tightened since launch and paid Google AI plans lift them.

Who should use Antigravity

Developers and non-developers who want a visual, agent-powered environment for building web projects, games, and browser extensions — or for letting an agent work directly on their computer. A great complement to Claude Code for projects where speed and volume matter more than frontier reasoning.