Google’s Antigravity 2.0 update forces developers into the AI agent era

On a Tuesday morning in May 2026, thousands of developers opened Antigravity and discovered their code editor was basically gone. Terminals, file explorers, inline errors, and even direct code editing had been stripped back. In its place: a chat-style interface built around AI agents.

This wasn’t a bug or a misconfigured update. It was Google’s deliberate pivot to an agent-first development platform—and it landed like a grenade in the middle of active projects.

From AI code editor to full agent platform

Antigravity launched in late 2025 as a straightforward AI-powered code editor. You wrote code, the AI helped with completions, refactors, and explanations—similar to tools like Cursor or VS Code with an AI assistant.

With Antigravity 2.0, unveiled around Google I/O 2026, that model is gone. Antigravity is no longer just an editor; it’s a full-stack platform for building and running AI agents. The new release ships with five major components:

• A standalone desktop app focused on agent orchestration, not traditional editing
• A new command line interface (CLI) that replaces the old Gemini CLI
• A developer SDK for building custom agent workflows
• A managed agents layer for spinning up AI agents via API
• An enterprise deployment path through Google Cloud

In other words, Google didn’t just update a tool. It rebuilt Antigravity into an operating system for AI agents and pushed everyone onto it overnight.

The new desktop app: an “agent control tower”

The Antigravity 2.0 desktop app is designed less like an IDE and more like what Google calls an “agent control tower.” Instead of you writing code line by line, you coordinate multiple AI agents working in parallel on different parts of your project.

Here’s what that looks like in practice:

• Multiple agents can run at the same time, each handling a different task (backend, UI, tests, docs, etc.)
• You can schedule tasks to run in the background so agents keep working even when you’re not actively watching them
• Voice commands are supported, tying into Google’s broader push across Gmail, Docs, and Workspace
• The app plugs into Google AI Studio, Android tooling, Firebase, and other parts of the Google ecosystem

Google even claims it used Antigravity internally to help build Gemini 3.5 Flash, suggesting the platform is already handling real production workloads inside the company.

The forced CLI migration (and looming shutdowns)

Alongside the desktop app, Google is also forcing a fast migration from the old Gemini CLI to the new Antigravity CLI. The deadline: June 18, 2026. After that date, the Gemini CLI stops working for everyone—pro, ultra, and free users alike.

Reports also indicate that Gemini Code Assist IDE extensions will stop processing requests entirely, effectively cutting off older workflows. This isn’t a soft deprecation; it’s a hard shutdown.

The new Antigravity CLI:

• Is written in Go for better performance
• Supports asynchronous workflows and multiple background agents
• Uses the same core agent harness as the desktop app, so improvements apply across both

From a technical standpoint, the migration is simple: the old command was gemini, the new one is antigravity, with a similar structure. The problem isn’t complexity—it’s the timeline. Many teams were given less than a month to migrate critical tooling, with the old stack being shut off regardless of readiness.

Managed agents: Google’s bigger play

The most strategic piece of Antigravity 2.0 is the managed agents layer. This is where Google moves beyond “AI assistant” and into full agent infrastructure.

With managed agents, you can:

• Create an AI agent via the Gemini API with a single call
• Let that agent reason, use tools, and execute code inside an isolated Linux environment
• Maintain a persistent environment across calls, so the agent remembers what it did and keeps context
• Extend agents with custom instructions and skills

Google AI Studio now includes templates for custom agents, making it easier to spin up specialized workers for tasks like code generation, testing, data processing, or app scaffolding.

Effectively, Google is exposing the same agent infrastructure it uses internally, optimized around Gemini 3.5 Flash and designed to scale from solo developers up to large enterprises. If you’re interested in building on this stack, it pairs directly with workflows covered in guides like how to build next‑gen AI apps with Google AI Studio and Antigravity.

Gemini 3.5 Flash: the engine behind it all

None of this works without a fast, capable model underneath. For Antigravity 2.0, that engine is Gemini 3.5 Flash.

Google claims Gemini 3.5 Flash can stream at around 289 tokens per second. For comparison, benchmark numbers cited in the launch put Claude Opus 4.7 at 67 tokens per second and GPT 5.5 at 71 tokens per second. If those figures hold up, Gemini 3.5 Flash is roughly four times faster than its main competitors in this context.

Google also says 3.5 Flash outperforms the older Gemini 3.1 Pro on most benchmarks while being significantly faster. That speed matters a lot when you’re running complex agentic workflows, where multiple agents are waiting on each other in chains. One slow agent can bottleneck an entire system.

To show what this enables, Google demoed Antigravity orchestrating 93 sub-agents to build a complete operating system from scratch in 12 hours for under $1,000. The process reportedly:

• Processed 2.6 billion tokens
• Involved 15,000 model requests
• Produced a working OS capable of running Doom live on stage

Even if that demo was heavily curated, it signals the scale of work Google expects developers to offload to agents.

Pricing: Pro, Ultra, and Ultra Premium

Google is clearly positioning Antigravity as a premium, infrastructure-heavy product rather than a cheap coding assistant.

The main tiers are:

• Pro: bundled with a Google AI Pro subscription; aimed at individual developers and experimenters
• AI Ultra: $100/month, with 5× the usage limits of Pro; targeted at people running multiple agents or building serious workflows
• Ultra Premium: $200/month (down from $250), with 20× Pro limits; aimed at teams and enterprises, especially those already on Google Cloud

New subscribers get $100 in bonus credits through May 25, 2026, and the Ultra plan includes extras like 20 TB of storage and YouTube Premium. For freelancers, Pro may be enough. For fast-moving startups, Ultra is the more realistic fit. For larger teams, the pricing is designed to funnel them into Google Cloud’s broader agent platform.

Google’s ecosystem play: from idea to deployment

Antigravity 2.0 is not just a standalone tool—it’s a central piece of Google’s attempt to own the entire lifecycle of AI-powered software development.

Key ecosystem moves include:

• A Google AI Studio mobile app (pre-registration available) to capture ideas on the go and turn them into prototypes
• Native agent access to Google Workspace APIs for tasks across Gmail, Docs, Sheets, and more
• Full Android support, including the ability to build Android apps via prompts and publish directly to the Google Play Console test track from AI Studio
• One-click export of AI Studio projects into local Antigravity workspaces while preserving full project context

On top of that, Google launched the “Build with Gemini X Prize Hackathon” with a $2 million prize pool—reportedly the largest hackathon prize ever. That’s a clear attempt to bootstrap a developer ecosystem around Antigravity and Gemini as fast as possible.

How Antigravity 2.0 compares to Cursor and Copilot

When you line Antigravity 2.0 up against other AI coding tools, Google’s strategy becomes clearer.

Compared to Cursor (built on VS Code), Antigravity focuses on:

• Multi-agent orchestration and background scheduling
• Voice command support
• Deep Firebase, Android, and Google Cloud integration
• A direct enterprise path for large-scale agent deployments

Cursor, on the other hand, wins on familiarity. It lives inside VS Code, so developers don’t have to change their entire workflow or editor muscle memory.

Against GitHub Copilot Workspace, Antigravity pushes harder on automation and multi-agent workflows. Copilot Workspace still feels closer to “AI that helps you code,” while Antigravity is leaning into “AI that codes while you supervise.”

Pricing is where Google diverges most: Antigravity Ultra at $100/month is far above Cursor ($20/month) and Copilot Workspace (~$19/month). Google isn’t trying to be the cheapest—it’s betting that deep infrastructure, speed, and integration will justify the premium for serious teams.

The rollout disaster: broken setups and missing editors

While the vision is ambitious, the actual rollout of Antigravity 2.0 was rough enough to spark immediate backlash.

Because the update was automatic, many developers woke up to:

• Antigravity 2.0 and Antigravity IDE (the traditional editor) conflicting during installation and overwriting each other
• Corrupted workspace configurations and lost setups
• A gutted traditional editor experience inside Antigravity 2.0

Specific pain points included:

• Visual indicators for warnings and errors were removed or buried, making it harder to spot small issues
• Direct code editing was heavily reduced in favor of a minimalist, chat-first interface
• Git and repository management became more manual and CLI-driven, losing visual integration
• The AI sometimes ignored minor warnings as long as the app didn’t crash, leaving developers feeling “blind” to subtle bugs

There were also tracking glitches: if you closed an app that Antigravity was testing, the chat interface might still think it was running and misdiagnose the closure as a malfunction. Developers had to manually kill processes to avoid false error reports.

Within hours, Reddit and Google’s own forums filled with complaints. Many developers compared the rollout to “non-technical people shipping straight to production” and called the agent-first pivot a “massive step backward.”

The immediate workaround for some teams was drastic:

• Uninstall Antigravity 2.0
• Install Antigravity IDE separately
• Manually copy configuration files
• Roll back to version 1.23.2 and disable automatic updates

What actually got better?

Despite the chaos, Antigravity 2.0 isn’t all downside. Some improvements are significant, especially for those who buy into the agent-first vision.

Reported benefits include:

• Much better real-time monitoring of agent activity and system state
• Lower memory usage, dropping from over 1 GB to roughly 150–500 MB thanks to automatic efficiency modes
• A unified agent harness shared between the desktop app and CLI, making it easier for Google to improve both at once

Google also reportedly refunded or refilled credits for affected users as compensation. But the damage to trust from the forced update and broken environments will take longer to repair.

Three tools, three workflows

After the dust settled, developers were left with a more fragmented Antigravity ecosystem than before:

• Antigravity 2.0: the agent-first control tower for orchestrating AI workers
• Antigravity IDE: the more traditional code editor experience
• Antigravity CLI: the new command-line interface replacing Gemini CLI

In theory, this separation gives teams flexibility. In practice, the automatic rollout, lack of clear migration paths, and absence of a simple rollback option made the transition painful—especially for teams in the middle of shipping features or fixing production bugs.

If you’re trying to navigate this shift while still keeping your engineering fundamentals strong, it’s worth pairing these tools with a mindset that emphasizes code quality, testing, and review, as explored in pieces like the hidden cost of AI coding that’s quietly breaking your engineering team.

Why Google is pushing so hard

Underneath all of this is a bigger strategic shift: for Google, AI is no longer a feature—it’s the foundation.

The same pattern is showing up across the company:

• Search experiments with AI-driven, query-specific layouts
• Gemini Spark runs background tasks across Google Workspace
• Gemini Omni handles video generation and multimodal workflows

Antigravity fits into this pattern as the developer-facing piece: AI that works continuously in the background, not just when you ask for a code completion.

That also means Google is quietly redefining what it means to be a developer. Instead of primarily writing code, your role shifts toward:

• Designing workflows and constraints for agents
• Supervising and reviewing their output
• Handling edge cases, architecture, and long-term maintainability

For many engineers, that’s a big mental jump—and the overnight rollout made it feel less like an invitation and more like a mandate.

Where does this leave developers?

Antigravity 2.0 is one of the clearest signals yet that the industry is moving toward agentic workflows, whether individual developers are ready or not. Google is betting that:

• Multi-agent orchestration will become normal for serious software projects
• Speed and infrastructure will matter more than editor familiarity
• Owning the full stack—from idea capture to deployment—will lock in teams for the long term

But the backlash shows a different reality on the ground. Many developers still want:

• Direct, reliable control over their code
• Stable tools that don’t change overnight
• AI that augments their work rather than replacing their core workflows

For now, the choice is yours: embrace the agent-first future and learn to supervise fleets of AI workers, or stick with more traditional setups and lighter-weight assistants. Either way, Antigravity 2.0 is a clear sign that the AI tooling war is moving beyond code completion and into full-blown software automation.