Something quietly significant happened at Google I/O 2026. Not the flashy Gemini demos or the Search overhaul everyone was talking about. The real shift came from a product most non-developers barely know: Google Antigravity.

With version 2.0, Google didn't just update an app. It retired a philosophy.

The old Antigravity was a code editor smart, AI-assisted, but still fundamentally an IDE where a human sat in the driver's seat. Antigravity 2.0 removes the steering wheel. You describe the destination; AI agents figure out the route, split the work, and keep going while you do something else. That's not an update. That's a new category.

Here's everything you need to know.

The Biggest Change: It's No Longer an IDE

Let's get this out of the way first, because every other headline is burying it.

Antigravity 2.0 is not an updated code editor. Google has built a separate standalone desktop application available on macOS, Linux, and Windows that has no IDE at all. No file tree. No syntax panels. No repository-first thinking. The primary object in Antigravity 2.0 is a task, organized into projects that can span multiple folders and contexts.

Google describes it as an "agent-first operating surface." That's corporate language for something real: you now talk to agents, they do the work, and you review the output. The IDE still exists separately for developers who want it. But Google's energy and its roadmap has moved on.

Why does this matter? Because it signals that Google thinks the dominant workflow for software development is about to change. Not in five years. Now.

What's Actually New in Antigravity 2.0

Multi-Agent Orchestration

This is the headline feature. Instead of one AI working on your problem, Antigravity 2.0 can spin up dynamic subagents focused mini-agents that each handle a piece of the work in parallel. One agent researches. Another writes code. A third tests. A fourth deploys.

Google's reasoning is practical: a single agent with a huge context window trying to do everything tends to lose focus and make mistakes. Subagents keep things tight. They mirror how real teams work someone coordinates, specialists execute, results get merged.

The architecture also allows for asynchronous task management. Long-running work continues in the background without blocking your current conversation. If an agent is installing dependencies or running tests, you don't stare at a loading screen you move on.

Scheduled Tasks (This One's Quietly Huge)

Antigravity 2.0 lets you set cron-like schedules that trigger agents automatically. Examples Google gives: run a code review every morning, generate reports weekly, monitor deployments hourly, refactor stale code overnight.

This is the moment AI moves from "tool you prompt" to "infrastructure that operates." Most developers haven't wrapped their heads around this yet. A scheduled Antigravity agent isn't just automating a task it's replacing a workflow that previously required a human to remember to do it.

New Slash Commands

Antigravity 2.0 ships with four slash commands worth knowing:

/goal The agent works until the task is fully complete, without asking for constant check-ins. Basically: stop interrupting, finish the job.

/grill-me Before starting, the agent interrogates you about requirements. This is one of the most practically useful features in the release. Vague briefs are the #1 cause of bad AI output. /grill-me forces clarity upfront.

/schedule Converts a prompt into a recurring autonomous workflow.

/browser Explicitly grants the agent permission to use a browser. Google admitted that agents weren't reliably deciding on their own when to browse, so they made it a deliberate choice. That's honest, and it's the right call.

Live Voice Transcription

Voice input now uses live transcription via Gemini Audio models rather than uploading an audio file. The difference in feel is significant it's the difference between dictating to a recorder and actually having a conversation. Google is clearly pushing toward voice as a first-class input method.

Hooks and Project-Level Permissions

JSON hooks let you intercept and customize agent behavior at specific points in a workflow. Projects carry scoped permissions you define what the agent can access and what it can't touch. For teams thinking about governance before autonomy, this is the feature to study.

Gemini 3.5 Flash Is the Engine Under the Hood

Antigravity 2.0 runs on Gemini 3.5 Flash Google's current top agentic and coding model which Google says was co-developed using Antigravity itself. The model handles the heavy lifting across parallel agents, and Google claims it can perform tasks comparable to larger frontier models at lower cost and faster output speeds.

More importantly: multiply those speed gains across multiple simultaneous subagents, and the compounding effect is real. Projects that took hours in Antigravity 1.0 are reportedly finishing in minutes.

The Gemini CLI Is Going Away What You Need to Do

This is the part most articles are treating as a footnote. It isn't.

Google has announced that Gemini CLI which has over 100,000 GitHub stars, 6,000 merged pull requests, and millions of users is being retired in favor of the new Antigravity CLI. The old tool stops serving requests on June 18, 2026 for AI Pro, AI Ultra, and free individual users.

The migration isn't automatic. Google says it will release video walkthroughs to ease the transition, and has confirmed that key Gemini CLI features have been ported over: Agent Skills, Hooks, Subagents, and Extensions (rebranded as Antigravity plugins). The new CLI is described as snappier and more capable of orchestrating multiple background agents.

Business users already on Gemini Code Assist don't need to take action the migration is handled on their end. Individual developers do.

Action item: If you're using Gemini CLI today, bookmark the Antigravity migration docs and plan your move before mid-June.

The New Pricing Tiers

Google introduced a new pricing layer alongside Antigravity 2.0:

This mirrors what Anthropic and OpenAI have done with tiered plans a $100/month mid-tier for developers who've outgrown the basic plan but don't need enterprise-level access. As a launch offer until May 25, 2026, new and existing Ultra subscribers can claim $100 in bonus credits.

The rationale is practical: agentic workflows burn more tokens than chat. If you're running parallel subagents on real projects, a $20 plan will cap out quickly.

The Broader Ecosystem: It's Not Just an App

What Google actually announced at I/O 2026 is bigger than the desktop application. Antigravity is now a platform:

Antigravity CLI For terminal-first developers who don't want to leave their workflow. Replaces Gemini CLI.

Antigravity SDK Lets developers build custom agents optimized for Gemini, then deploy on whatever infrastructure fits their organization.

Managed Agents via Gemini API Agents get an isolated Linux environment to operate in. This is Google offering hosted agent infrastructure, not just model access.

AI Studio Integration Projects built in AI Studio can now be exported directly to local Antigravity, carrying over all context. A dedicated AI Studio Android app is also launching, letting users capture ideas on mobile and return home to find a working prototype.

Google Cloud + Firebase integration Antigravity connects natively to Google's existing developer infrastructure.

The strategy here is clear: Google wants Antigravity to be the layer that ties together model access, developer tools, cloud infrastructure, and enterprise workflows. They're not just competing with Cursor or Claude Code on features. They're trying to own the agent ecosystem end to end.

What This Means If You're Not a Developer

Google has been explicit about something most of the coverage hasn't mentioned: Antigravity is expanding beyond coding.

The removal of IDE-centric design is partly philosophical (agents don't need file trees), but it's also a market signal. Google's next target isn't just developers. It's marketers who need landing pages built. Analysts who need reports generated on a schedule. Operations teams who need recurring workflows without hiring engineers.

The /grill-me command, scheduled tasks, and multi-folder project support all make more sense when you imagine a user who doesn't know what Git is. That's intentional. Google is building for a future where coding agents serve knowledge workers broadly not just the people who currently write code for a living.

The Question Worth Sitting With

A year ago, AI coding tools were autocomplete on steroids. Antigravity 2.0 is a multi-agent desktop operating system with scheduled autonomous workers and background task orchestration.

The honest question isn't whether this is impressive. It clearly is. The question is what it changes about how we work. And more specifically: what does a developer's day look like when agents handle the implementation and humans handle the direction?

The answer isn't settled. But Google has clearly decided which side of that question to build toward.

FAQ

What is Google Antigravity 2.0? A standalone desktop agent platform (macOS, Linux, Windows) built around autonomous AI agents rather than a code editor. Released at Google I/O 2026.

Is it replacing the original Antigravity IDE? No the IDE still exists. Antigravity 2.0 is a new, separate app focused on agent orchestration. Google's focus and development energy has shifted to this new surface.

What happened to Gemini CLI? It's being retired. Users must migrate to Antigravity CLI before June 18, 2026. Key features from Gemini CLI have been ported over.

How much does it cost? Free access exists with limits. AI Ultra at $100/month offers 5x limits vs. the $20 Pro plan. The top Ultra tier is $200/month (reduced from $250).

What model powers Antigravity 2.0? Gemini 3.5 Flash, which Google says was partly developed using Antigravity itself.

Can non-developers use it? That's clearly the direction Google is heading. Multi-folder projects, scheduled tasks, and voice input all lower the barrier for non-technical users significantly.