Google Antigravity 2.0 at I/O 2026: Agent-First Dev Platform With CLI, SDK, and Managed Execution
Google renamed Gemini CLI to Antigravity and shipped a full agent-first developer platform at I/O 2026. The keynote demo built an OS framework in 12 hours for under $1,000 using 93 parallel sub-agents.
Google shipped Antigravity 2.0 at I/O 2026 on May 19. This is not a rebrand of Gemini CLI with a fresh coat of paint — it’s the most complete agent-developer platform Google has shipped, and it puts Cursor, Windsurf, and GitHub Copilot Workspace in a materially different competitive position.
What Shipped
Antigravity 2.0 is five interconnected pieces:
- Desktop app — a standalone agent-first IDE with integrated task management and sub-agent orchestration
- Antigravity CLI (
agy) — the command-line agent, evolved from Gemini CLI, with multi-agent coordination built in natively - Antigravity SDK — for embedding custom agent workflows into existing toolchains and CI pipelines
- Managed Agents in Gemini API — deploy your own sub-agent pipelines on Google’s infrastructure without managing execution servers
- Gemini Enterprise Agent Platform — the enterprise tier with compliance controls, audit logging, and dedicated capacity
The Demo That Set the Bar
Google’s live I/O demo used Antigravity 2.0 to build the core framework of a working operating system in approximately 12 hours. The session spawned 93 sub-agents, processed billions of tokens, and came in under $1,000 total compute cost.
The benchmark is deliberately provocative. The point is not that you should build an OS this way — it’s to show the platform can coordinate long-horizon, multi-agent work at a price point that makes industrial use economically rational. Twelve hours of distributed agent work for less than four figures is a different cost structure than anything available six months ago.
Pricing
- AI Pro (existing plan) — unchanged
- AI Ultra — $100/month, 5× higher usage limits vs. Pro
- AI Ultra+ — $200/month, 20× higher usage limits vs. Pro (reduced from $250 at announcement)
The underlying model is Gemini 3.5 Flash, now generally available, priced at $1.50/$9 per 1M tokens (input/output). Google claims 4× faster throughput than comparable frontier models on coding tasks, which is load-bearing for multi-agent workflows where sequential latency compounds.
Why This Is Different From Interactive Coding Tools
The CLI + SDK combination is the key differentiator. Teams can wire agy into CI pipelines, trigger Managed Agent workflows from existing tooling, and deploy custom sub-agents without standing up server infrastructure. This is positioning Antigravity as an orchestration layer for software factories — not a smarter text editor.
The integration story is tighter than most competitors. Native hooks for Firebase, Google AI Studio, and Android Studio mean teams already in the Google ecosystem get the highest leverage on day one, without third-party bridge code.
What Is Missing at Launch
The gaps are real. GitHub and GitLab integration is third-party only — there is no native PR/issue workflow comparable to Copilot Workspace. Local model support is absent, which matters for security-sensitive environments. The Managed Agents tier ships without per-agent cost controls, which is a meaningful risk for teams automating production workflows where runaway agent loops can rack up unexpected bills.
These are tractable problems, and Google’s track record with Gemini API iteration suggests they will close quickly. But teams evaluating Antigravity for production use today should factor them in before committing to the platform.