The Google I/O 2026 Developer keynote, published by the Google I/O team on the Google Developers Blog, announced Google's transition from AI that simply assists to autonomous agents capable of handling intricate workflows. The keynote centered on the Gemini 3.5 series of models and the enhanced Antigravity agent-first development framework.
Antigravity 2.0 introduced a new CLI enabling developers to 'spin up specialized subagents to tackle complex workflows' with built-in sandboxing and credential protection. The Managed Agents offering in the Gemini API eliminates infrastructure setup friction, allowing technical teams to define agentic behavior while Google Cloud handles the underlying infrastructure. Additionally, the new Antigravity SDK provides programmatic control for custom deployments, giving developers flexibility in how they integrate agentic capabilities.
Google AI Studio received significant upgrades with native Kotlin support for Android development, Google Workspace integration for accessing organizational data, and one-click Cloud Run deployment that simplifies the path from prototype to production. The first two app deployments are complimentary, lowering the barrier for developers to experiment with agent-powered applications.
For Android development, the stable Android CLI now grants AI agents access to Android Studio capabilities, including SDK downloads and device testing. Google open-sourced Android skills to help LLMs execute best practices for complex migrations. Android Bench, an LLM leaderboard specifically for Android development tasks, was expanded to include Gemma 4 and other open-weight models. Perhaps most impressively, a new Android Studio migration agent can convert React Native, web framework, and iOS code into native Kotlin applications, turning 'migrations that would have taken weeks into just hours.'
Web development received equally transformative announcements. WebMCP was introduced as a proposed open web standard that enables browser-based AI agents to execute complex tasks with enhanced precision. Chrome 149 launches an experimental WebMCP origin trial, allowing developers to expose structured tools so that browser-based AI agents can interact with web applications programmatically rather than through brittle screen-scraping approaches. Modern Web Guidance provides agents with vetted skills covering 100+ use cases, while Chrome DevTools for agents facilitates automated quality audits and debugging.
The HTML-in-Canvas API, available in origin trial, represents a significant technical innovation that combines real DOM elements with WebGL and WebGPU rendering. This enables developers to build immersive 3D experiences that remain fully searchable, accessible, and interactable -- addressing a longstanding tension in web development between rich visual experiences and web fundamentals like SEO and accessibility.
The developer keynote also highlighted the Agent Development Kit (ADK) 2.0, which takes a code-first, engineering-focused approach to building custom agent meshes. ADK 2.0 features a unified graph-based engine with a slider from dynamic, model-led reasoning to strict, deterministic workflows. New collaborative workflow modes include 'chat' for full user interaction, 'task' for assignment-based collaboration, and 'single-turn' for parallel execution without user interaction. ADK now supports Kotlin alongside Python, Go, and Java, enabling on-device mobile agents that can seamlessly coordinate with backend Python agents.
Notably, Google emphasized openness and interoperability throughout the keynote. The Agents CLI is designed to work with any AI coding agent, including Antigravity, Gemini CLI, Claude Code, and Cursor. The A2A (Agent-to-Agent) protocol ensures interoperability across the entire agent development stack, so an agent built with low-code tools can be called as a sub-agent by a fully custom-coded agent mesh.