Google Rolls Out Gemini 3.5 and Agent-First Development Tools at I/O 2026

Google has shifted its focus from AI assistants to autonomous agents at this year's I/O developer conference, announcing the Gemini 3.5 series of models alongside major upgrades to Antigravity, its agent-first development platform.

Key Announcements#

Agent Orchestration & Development The upgraded Antigravity 2.0 and new Antigravity CLI let developers spin up specialized subagents for complex workflows with built-in sandboxing, credential masking, and hardened Git policies. For those wanting self-hosted solutions, the new Antigravity SDK provides programmatic control to deploy agents on custom infrastructure.

AI Studio Enhancements Google AI Studio now supports native Kotlin for vibe-coding Android apps, integrates with Google Workspace, and enables one-click deployment to Cloud Run with Firebase support. Developers can export complete project state directly to Antigravity for extended development.

Android Development Tools

• The stable Android CLI lets agents interface directly with Android Studio to manage SDK downloads, device testing, and app builds
• Android Bench, a new LLM leaderboard, measures how different models perform on Android development tasks, now including open-weight models like Gemma 4
• An Android migration agent feature can convert React Native, web, or iOS code to native Kotlin Android apps in hours instead of weeks
• Open-sourced Android skills help LLMs execute best practices for complex workflows and APIs

Web Development & Browser Standards

• WebMCP, a proposed open web standard, lets browser-based AI agents execute complex tasks by exposing structured tools and JavaScript functions. An experimental origin trial launches in Chrome 149
• Modern Web Guidance provides 100+ vetted skills to help agents build performant, accessible, and secure web experiences, with integration to Baseline standards
• Chrome DevTools for agents enables AI to automate quality audits, emulate user experiences, and debug code in real time
• The new HTML-in-Canvas API allows developers to build immersive 3D experiences using real DOM elements with WebGL and WebGPU, maintaining searchability and accessibility

Google plans to livestream additional sessions May 19–20, with over 85 sessions, codelabs, and resources available on demand starting May 21 on .