By Google dropped AgentKit 2.0 for Antigravity this week — 16 specialized agents, 40-plus domain-specific skills, and tightened A2A protocol support — and the timing is deliberate. Google I/O 2026 runs May 19–20, and this is Google’s clearest statement yet that Antigravity is a serious competitor to Cursor and Claude Code. The feature set is real. So is the quota cliff that’s been locking developers out for days at a time.
What AgentKit 2.0 Actually Ships
AgentKit 2.0 brings 16 specialized agents covering the full development stack: frontend design, backend logic, security review, testing, infrastructure, SEO, mobile layout, UX, and database configuration. Each agent arrives pre-loaded with domain-specific skills — 40-plus in total — that only activate when your request matches the skill’s description. That keeps context lean rather than stuffing everything into every prompt.
The framework uses an Agent MD file to define operational rules, giving you precise control over how an agent behaves on a given task. Skills can be scoped globally (available across all projects) or to a specific workspace. Google has published authoring documentation and Codelabs, so writing custom skills for your own stack is a supported workflow, not a hack.
The update also ships 11 pre-configured commands and deeper integration with InForge for backend automation — real-time deployment and performance monitoring included.
Why A2A Matters More Than the Skill Count
The most strategically important part of AgentKit 2.0 isn’t the agent roster — it’s the strengthened Agent-to-Agent (A2A) protocol support. A2A is Google’s open standard for agent-to-agent communication, announced April 2025 and now backed by over 150 organizations: Microsoft, AWS, Salesforce, SAP, LangGraph, CrewAI, AutoGen, and Semantic Kernel among them.
A2A uses HTTP, Server-Sent Events, and JSON-RPC 2.0 as transport with Agent Cards for capability advertisement. It’s designed to complement MCP: MCP handles agent-to-tool calls, A2A handles agent-to-agent delegation. With AgentKit 2.0, your Antigravity agents can delegate subtasks to — and receive from — agents built in completely different frameworks.
If A2A becomes the dominant standard for multi-agent interoperability — and adoption numbers suggest it’s moving that direction — Antigravity becomes the natural orchestration hub. That’s the bet Google is making.
How It Stacks Up Against Cursor and Claude Code
Antigravity is still in free public preview as of May 2026 — no credit card, no waitlist, all features unlocked. That’s a meaningful advantage over Cursor Pro ($40/month) and Claude Code Pro ($20/month). The platform’s 1-million-token context window via Gemini 3.1 Pro is also a real differentiator for large codebases, where Cursor and Claude Code cap out around 200K tokens.
The honest take: Antigravity sits in the middle on performance. Cursor wins on raw speed. Claude Code produces the most thoroughly-architected output. Antigravity auto-verifies its own work — running tests, capturing screenshots, comparing against spec — which reduces debugging time but adds latency. For developers who already use Cursor daily, Antigravity is a compelling addition for complex multi-agent tasks. It’s not a full replacement yet.
Context rot is also real. Within a single session, Antigravity can lose track of file structures and architectural decisions made just a few prompts ago. That’s a ceiling on how much you can trust it for long, sustained workflows.
The Quota Problem Isn’t Fixed
AgentKit 2.0 ships strong features. But Google’s March 2026 quota cuts haven’t been reversed. Google silently reduced free-tier quotas by roughly 92 percent and switched the refresh window from every five hours to weekly. Developers reported being locked out for entire weeks after 20 to 30 minutes of active use.
The billing unit is “compute effort” — not tokens, not requests — which makes it nearly impossible to predict how much quota a task will consume. Competitors use transparent token-based metering. Heavy Antigravity users are being pushed toward a $250/month plan, well above every comparable tool on the market. Google has not publicly addressed the community complaints.
What’s Coming at I/O 2026
The official Google I/O workshop on May 19 — “Vibe once, run anywhere with Google Antigravity and Flutter” — demonstrates the Flutter GenUI integration, which generates adaptive UI components from natural language at runtime. Gemini 4 is expected at I/O and will likely ship with Antigravity integration.
If the quota issue gets addressed alongside those announcements, AgentKit 2.0 starts looking like a genuine inflection point. If it doesn’t, the developer community’s patience is already thin. Try Antigravity while the free preview holds — just be aware of the weekly quota ceiling before you commit it to anything mission-critical.
