Google dropped something at I/O 2026 that caused a lot of noise — and for once, the noise seemed justified

Antigravity 2.0 is a free AI coding platform that can build entire websites from scratch, run several AI agents at the same time, and schedule tasks to keep running overnight. You don’t need to write code to use it. You don’t need a designer. You don’t even need a subscription to get started.

If that sounds like a lot of marketing speak, fair enough. But after watching it build a full D2C (direct-to-consumer) brand website live — without a single line of code being written by a human — it’s hard to dismiss.

Here’s what Antigravity 2.0 actually is, what it can do, and whether it’s worth your time.

What is Google Antigravity 2.0?

Antigravity 2.0 is Google’s agentic AI development platform, announced at Google I/O 2026. Think of it as an AI assistant that doesn’t just answer questions — it actually goes off and does things for you.

The original Antigravity launched in 2025 as an AI-integrated code editor. Version 2.0 is a much bigger deal. It now includes three components:

  • A desktop app (a fork of VS Code, rebuilt around AI agents)
  • A command-line tool (the Antigravity CLI, written in Go)
  • An SDK for developers who want to build their own custom agent workflows

The big shift is the move from “assistant” to “agent.” An assistant waits for you to tell it what to do. An agent figures out what needs to happen, breaks it into subtasks, and runs them — often in parallel — without you babysitting every step.

How does it actually work?

When you give Antigravity a task, a “manager agent” takes over. It reads your request, breaks it into smaller pieces, and assigns each piece to a specialized sub-agent. One agent might write the HTML, another handles the CSS, a third tests the layout in a browser window.

They run at the same time. They check each other’s work. They iterate until the output passes all the checks.

This is meaningfully different from asking ChatGPT or Claude to write some code and then copy-pasting it into your editor. Antigravity is in the loop from start to finish — reading your files, running commands, fixing errors, verifying results.

Under the hood, it uses Gemini 3.5 Flash, a version of Google’s model specifically optimized for this kind of agent work. It outputs at around 289 tokens per second, which is about 3-4x faster than competing models. Speed matters here because agents are doing a lot of back-and-forth reasoning.

What can you build with it?

The demo that got the most attention was building a full premium D2C brand website — the kind of thing that would normally take a web designer, a developer, and a few days of back-and-forth.

Antigravity did it in a single session. Multiple agents ran in parallel: one working on the product page layout, another writing copy, another setting up the checkout flow. The result wasn’t a template — it was a customized site built to the brief.

Beyond websites, Antigravity 2.0 handles:

  • Scheduled background tasks — you can set agents to run overnight and come back to finished work in the morning
  • Code refactoring and debugging — give it a broken function and it’ll trace through, find the issue, and fix it
  • Documentation generation — useful if you’ve inherited a codebase with no comments and no README
  • API integrations — agents can make web requests, interact with external services, and handle structured data

One thing worth noting: it runs code in a secure Linux sandbox. The agents can’t do anything to your machine beyond what you’ve authorized. This matters because you’re essentially letting an AI system run commands on your behalf.

Who is it actually for?

This is where it gets interesting.

Antigravity 2.0 has a very clean, minimal interface — a file browser on the left, a conversation panel on the right. You describe what you want, and the AI handles the implementation. There’s no coding required to use the basic features.

That makes it genuinely accessible to people who’ve never written a line of code. A small business owner who wants a professional website, a content creator who wants to build a newsletter landing page, a founder who wants to prototype an app idea — these are people who previously needed to hire someone or spend months learning.

At the same time, there’s a CLI and an SDK for developers who want to go deeper. The SDK exposes the same agent primitives (code execution, file management, web browsing) through Python, TypeScript, or Go, so you can plug Antigravity into your own tools or CI/CD pipeline.

It’s one of those rare tools that’s trying to serve two audiences without being watered down for either.

Is Antigravity 2.0 actually free?

Sort of. Here’s the honest breakdown:

The free tier gives you access to everything but hits rate limits during busy hours. For light, occasional use — testing a concept, building a small personal site — that might be enough.

For sustained use, you’re looking at:

  • Google AI Pro — $20/month (same tier as many other AI subscriptions)
  • Google AI Ultra — $100/month (5x the usage limits)
  • Google AI Ultra+ — $200/month (20x limits, for heavy professional use)

Usage is compute-based and refreshes every 5 hours, not daily. So you’re not locked out for the rest of the day if you hit your limit at noon — you can try again a few hours later. You can also top up with pay-as-you-go credits.

The $20/month tier is competitive with Claude Code, Cursor, and other AI coding tools. Whether it’s worth it depends on how much you’re using it and what you’re building.

How does it compare to Claude Code and Codex?

Quick version:

Claude Code (from Anthropic) is a terminal-only tool that runs in your command line. It produces higher-quality code on complex tasks according to SWE-Bench benchmarks, but it’s more of a power-user tool. No desktop app, no parallel agents by default.

Codex (from OpenAI) sits in the middle — good speed, strong ecosystem integration with other OpenAI tools, but still sequential rather than parallel agent execution.

Antigravity 2.0 wins on breadth (desktop + CLI + SDK), raw speed (289 tokens/sec vs roughly 80-120 for competitors), and multi-agent parallel execution. It loses slightly on raw code quality for very complex tasks, and the SDK is still new with a smaller community around it.

Which one is “best” depends on what you’re doing. Antigravity is the most accessible starting point. Claude Code is the better choice if you care about code quality on hard problems and live in a terminal. Codex makes sense if you’re already deep in the OpenAI ecosystem.

The thing nobody mentions

Most coverage focuses on what Antigravity can build. The less-discussed part is the scheduled tasks feature.

You can set up an agent to run at a specific time — overnight, weekly, whatever — and it’ll execute the task in the background without you being present. This is genuinely different from anything that’s been available before at this price point.

The obvious use case is automating repetitive development work. But think about it more broadly: automatically updating a website with new content, running weekly code health checks, generating reports from live data. These are things that used to require a developer to write and maintain automation scripts. Antigravity does it through natural language.

Should you try it?

If you’ve ever wanted to build something on the web and been stopped by the technical barrier, Antigravity 2.0 is probably the most accessible entry point available right now.

If you’re already a developer, it’s worth running a few tasks through it and seeing where it saves you time. The parallel agent architecture is genuinely fast, and the scheduled tasks feature opens up some interesting automation possibilities.

The free tier is enough to test it properly. Start there, see what it can do for your specific use case, and decide from there whether the $20/month makes sense.

Sources: