Google Antigravity 2.0

Antigravity is Google's attempt to skip the plugin era of AI coding entirely. Rather than bolting an assistant onto an editor, it forks VS Code and rebuilds the workflow around agents: a familiar Editor View sits alongside a controlled Chrome instance the agent can drive itself, and an Agent Manager that behaves less like a chat window and more like an inbox for delegated work. It launched as a free preview in November 2025 and graduated at Google I/O in May 2026 into something closer to a platform, with a rebuilt desktop app, a Go CLI, an SDK and a pay-per-run Managed Agents tier in the Gemini API.

The pitch lands hardest for front-end and full-stack developers who are already comfortable handing off whole tasks. The browser surface is the genuinely novel part here — an agent that can launch your app, click through it and screenshot the result is closing the loop that most AI coding tools leave dangling, where code compiles fine and looks broken. The Manager View's five parallel agents push the same idea further: this is a tool designed for someone supervising work, not typing it.

It is emphatically not for everyone. If you live in IntelliJ, PyCharm or WebStorm, there is nothing here for you, and adopting Antigravity means adopting a whole new editor rather than adding a feature to the one you have. At $20/month for AI Pro it lines up exactly against Cursor Pro and Windsurf Pro, so the question is not whether it is cheap — it is whether it is dependable.

On ambition, Antigravity is the most interesting thing in this category. Self-verifying front-end changes via a real browser, and an inbox metaphor for orchestrating five agents at once, are the kind of design decisions that suggest someone actually thought about what agentic development feels like day to day rather than shipping a chat sidebar with a bigger context window. The 2.0 additions — CLI, SDK, managed agent runs — read as a credible bid to be infrastructure rather than an app.

The problem is the meter. Four quota cuts in four months between December 2025 and March 2026, reportedly locking paying Pro subscribers out for days at a stretch, is not a rough patch; it is a pattern. Add an opaque credit system, a five-rung price ladder that runs from free to $200/month plus $25-per-2,500-credit managed runs, and a $249.99 tier that was quietly repriced, and the impression is of a product whose economics are still being worked out in public at customers' expense. An agent platform you cannot rely on being available is a worse proposition than a weaker one you can.

Two further reservations. The provenance is murky — reviewers point to internal 'Cascade' references and stripped VS Code branding as signs of Windsurf lineage, which is a footnote rather than a flaw, but it does undercut the idea of this as clean-sheet engineering. And the model story is genuinely confused: depending on which surface and which release you look at, you are being told the defaults are Gemini 3.1 Pro/Flash with Claude and GPT alternatives, or an Antigravity-tuned Gemini 3.5 Flash. That inconsistency matters when speed and quality are exactly what you are paying for.

Our rating reflects that split. Judged on design, Antigravity earns a strong recommendation; judged as something to build a team's workflow on at $20/month, it has not yet earned trust. Try it on the free tier, keep your existing setup installed, and revisit once Google has gone a full quarter without changing the rules.

• Verdict: buy into the ideas, not yet the subscription.
• Best fit: front-end and full-stack developers already on VS Code who want agent delegation and are willing to tolerate quota risk.
• Avoid if: you are a JetBrains user, or you need predictable availability for billable work.