On the morning of May 19, 2026, thousands of developers opened their computers and found their code editors had effectively disappeared. Terminals gone. File explorers gone. The ability to directly edit code: stripped out. What replaced it was a chat interface. Google had pushed an automatic update to Antigravity overnight, and within hours Reddit and Google's own developer forums filled with people reporting their active projects were now unusable.
This wasn't a bug. It was intentional. And it tells you exactly where Google thinks software development is going.
Table of Contents
- What Actually Changed in Antigravity 2.0
- The Five Components of the New Platform
- Managed Agents: The Part Most Coverage Missed
- Speed Is the Whole Argument
- The OS Demo: What It Actually Showed
- Pricing Breakdown
- What Broke and Why Developers Are Angry
- Antigravity 2.0 vs. Cursor vs. GitHub Copilot Workspace
- My Take
- FAQ
What Actually Changed in Antigravity 2.0
Antigravity launched in November 2025 as an AI-powered code editor. The pitch was simple: open it, write code, the AI assists. It competed directly with Cursor. That product is now gone, at least as the default experience.
Version 2.0 is not an upgrade. It is a rebuild. Google repositioned Antigravity from a tool that helps developers write code into what it calls an "agent control tower" — a platform for supervising multiple AI agents working on a project simultaneously. The developer's role shifts from writing code to directing agents that write code.
That is a fundamentally different product. And Google shipped it as an automatic overnight update without an opt-in.
The Five Components of the New Platform
Antigravity 2.0 ships as five separate pieces:
1. Standalone desktop app. Designed as the agent control tower. Multiple AI agents run in parallel on different parts of a project simultaneously — one on the back end, another on the interface — without waiting on each other. Scheduled tasks run in the background even when the developer is not actively working. Voice command support is included.
2. Antigravity CLI. A new command-line interface that fully replaces Gemini CLI, built in Go for speed and responsiveness. The old command was gemini; the new one is antigravity. It supports asynchronous workflows and shares the same agent harness as the desktop app. Improvements to the core agent system apply to both products simultaneously.
3. Developer SDK. For building custom agent workflows on top of the platform.
4. Managed Agents. An API layer for spinning up AI agents on demand (covered below — this is the most significant piece).
5. Enterprise path through Google Cloud. Targeted at teams already running on Google Cloud infrastructure.
The platform also connects directly into Google AI Studio, Android development, and Firebase — the full Google ecosystem in one surface. Google stated it used Antigravity to help build Gemini 3.5 Flash internally, which suggests the tool is already handling production-level workloads at Google itself.
The Gemini CLI shutdown deadline is June 18, 2026, and it applies to everyone — pro, ultra, and free tier. Gemini Code Assist IDE extensions stop processing requests on the same date. Migration is not optional.
Managed Agents: The Part Most Coverage Missed
The managed agents layer is where Google is making a move that goes beyond giving developers a better coding assistant.
A single API call through the Gemini API spins up an agent that can reason, use tools, and execute code inside an isolated Linux environment. Each conversation starts a new environment that persists across subsequent calls — it remembers what it did, maintains context, and picks up where it left off. Developers can extend these agents with custom instructions and skills. Google AI Studio now provides custom agent templates to simplify this.
This is Google handing developers the same agent infrastructure it uses internally, co-optimized with Gemini 3.5 Flash for exactly this kind of workload.
Speed Is the Whole Argument
None of the parallel agent architecture works if the underlying model is slow. In agentic workflows, agents wait on each other in chains. One slow agent bottlenecks the entire system.
Google claims Gemini 3.5 Flash hits 289 tokens per second. The video cites Claude Opus 4.7 at 67 tokens per second and GPT 5.5 at 71 tokens per second for comparison. If those figures hold, that is roughly four times faster than the nearest competitor.
Google also reports that Gemini 3.5 Flash outperforms the older Gemini 3.1 Pro model across most benchmarks while being substantially faster. The speed gap is not a minor detail — it is the structural reason parallel agent orchestration at this scale is possible at all. For a closer look at what Google shipped alongside this at I/O, see Gemini Omni Flash explained.
The OS Demo: What It Actually Showed
During the I/O keynote, Varun Mohan, who runs Google's Antigravity platform, demonstrated the system building a complete operating system from scratch. The numbers: 12 hours, under $1,000 in compute costs, 93 sub-agents running simultaneously, 2.6 billion tokens processed across 15,000 model requests. At the end, Doom ran on the new OS live on stage.
A working kernel. A game running on it. Built in half a day for less than a thousand dollars.
Stage demos are optimized for stage demos. But even accounting for that, the scale of what was shown is a clear signal of where Google is pointing this platform.
Pricing Breakdown
Three tiers:
Pro — bundled with a Google AI Pro subscription. Entry point for individuals and experimentation.
AI Ultra — $100 per month. Five times the usage limits of Pro. Includes 20TB of storage and YouTube Premium. For developers running multiple agents regularly or building production workflows.
Ultra Premium — $200 per month (reduced from $250). Twenty times the Pro usage limits. Aimed at teams and enterprises already on Google Cloud.
For context: Cursor is $20 per month. GitHub Copilot Workspace is $19. Google is not competing on price. It is competing on infrastructure depth.
New subscribers had access to $100 in bonus credits through May 25, 2026. Google also announced a Build with Gemini X Prize Hackathon with a $2 million prize pool — framed as the largest hackathon prize pool ever — to accelerate developer adoption on the platform.
What Broke and Why Developers Are Angry
The product vision is coherent. The launch execution was a mess.
The automatic update installed Antigravity 2.0 and gutted the traditional IDE experience without notice. Terminals gone. Visual error and warning indicators gone or degraded. Direct code editing stripped back in favor of a minimal chat interface. Git and repository management shifted from visual integration to manual command-line work.
Antigravity 2.0 and Antigravity IDE — the separate traditional editor — conflicted during installation, fighting over the same directory and overriding each other. In some cases, one installation deleted the other. Existing workspace configurations were corrupted. Developers lost their setups.
There were also tracking failures: close an app Antigravity was testing and the chat interface would still register it as running, then diagnose the closure as a malfunction. Developers had to manually kill processes to stop false error reports.
One developer described it as non-technical people shipping code straight to production. Another called the agent-first pivot a massive step backward. The common fix: uninstall 2.0, download Antigravity IDE separately, manually copy configuration files, or roll back to version 1.23.2 and disable automatic updates.
There were genuine improvements in the update. Real-time monitoring improved significantly. Memory consumption dropped from over a gigabyte to 150-500MB with automatic efficiency mode. Google refilled usage credits as compensation. But developers now face three separate downloads for what used to be one tool: Antigravity 2.0 for agent orchestration, Antigravity IDE for traditional coding, Antigravity CLI for terminal work.
Antigravity 2.0 vs. Cursor vs. GitHub Copilot Workspace
Against Cursor, Antigravity leads on multi-agent orchestration, background task scheduling, voice commands, Firebase and Android integration, and the enterprise path through Google Cloud. Cursor's advantage is familiarity — it is built on VS Code, so developers do not have to restructure their entire workflow to use it.
Against GitHub Copilot Workspace, Antigravity has substantially stronger multi-agent capabilities and more aggressive automation. Copilot Workspace at $19 per month serves a completely different use case than Antigravity's $100 Ultra tier.
The positioning is clear. If you are building on Android or Google Cloud, Antigravity is designed to be the obvious choice. If you want an AI assistant inside your existing editor without changing your setup, Cursor still makes more sense.
My Take
The technology is real. 93 parallel agents building a working OS in 12 hours is not a gimmick, and 289 tokens per second — if it holds under production load — changes the economics of agentic workflows in a way that matters.
But Google shipped a platform transition as a surprise. No opt-in. No rollback path. Developers running production work woke up to broken environments because Google decided the timeline was now. That is the part that will cost them trust in the developer community, and trust in developer tools is slow to rebuild once lost. For context on how Anthropic is moving in the same space, the Anthropic Mythos cybersecurity findings show a different approach to deploying agent-level AI.
The product is where development is going. The launch is what you do when you don't respect where developers currently are.
- Antigravity 2.0 is a complete platform rebuild, not an update — five components replacing what was a single code editor.
- The traditional IDE is not gone but must be downloaded separately. The automatic update did not include it.
- Gemini CLI shuts down June 18, 2026. Migration is mandatory for all tiers.
- Google claims Gemini 3.5 Flash runs at 289 tokens/second — roughly four times faster than the competition.
- The OS demo (93 agents, 2.6B tokens, 12 hours, under $1,000) is the clearest indicator of what this platform targets.
- Rollback path: version 1.23.2, disable automatic updates.
Frequently Asked Questions
What is Google Antigravity 2.0?
Antigravity 2.0 is Google's AI agent development platform, announced at Google I/O on May 19, 2026. It replaces the previous Antigravity code editor with a five-component platform — desktop app, CLI, SDK, managed agents API, and enterprise Cloud path — built around parallel AI agent orchestration powered by Gemini 3.5 Flash.
Did Google remove the code editor from Antigravity?
The traditional code editor was not included in the automatic 2.0 update, which is why developers woke up to a chat-only interface. Antigravity IDE — the editor — still exists but must be downloaded separately from the main Antigravity 2.0 app. Google has since pushed UI updates to make this clearer.
What is the Gemini CLI shutdown deadline?
June 18, 2026. After that date, Gemini CLI stops working for all users — pro, ultra, and free tier. Gemini Code Assist IDE extensions also stop processing requests. The replacement is Antigravity CLI, which uses the same command structure with antigravity replacing gemini.
How do I roll back Antigravity to the previous version?
Uninstall Antigravity 2.0, then download version 1.23.2 from Google's previous releases page. After installing, disable automatic updates in settings to prevent Google from pushing the new version again. Manually copy your configuration files to restore your previous setup.
How does Antigravity 2.0 pricing compare to Cursor?
Cursor costs $20 per month. GitHub Copilot Workspace is $19. Antigravity's AI Ultra tier — the one designed for regular agent use — is $100 per month. Google is not competing on price. The value proposition is multi-agent orchestration, deeper Google ecosystem integration, and an enterprise path through Google Cloud that neither Cursor nor Copilot Workspace offers at the same scale.
Where This Is Going
Google is not treating AI as a feature inside a development tool anymore. It is the foundation. The same pattern shows up across their products — AI building custom search layouts per query, Gemini running background tasks across Workspace, Gemini handling video generation. The direction is AI that works continuously, not just when you ask it to.
Antigravity 2.0 is the developer-facing version of that bet. Whether the developer community was ready for the transition is a separate question from whether the transition is coming. Google forced the answer. The backlash shows many weren't ready. The real question over the next 90 days is whether the platform stabilizes fast enough that the developers who rolled back give it another look — or whether the trust damage sticks. If you want to understand why the industry is moving this direction at all, Karpathy's self-training experiment is the clearest explanation of where the underlying logic comes from.
0 Comments