Google I/O 2026: Everything Developers Need to Know

Google Just Redefined What a Developer Platform Looks Like

Google I/O 2026 happened on May 19–20, and the developer keynote was one of the most dense product drops in recent memory. Not just incremental updates — actual shifts in how you build software with Google's stack.

The throughline across everything: Google is moving from "AI that helps you write code" to "agents that write, test, migrate, and deploy code for you." Here's what that looks like in practice.

Gemini 3.5 Flash: The New Default Model for Builders

The headline model is Gemini 3.5 Flash — and the numbers are worth paying attention to. It surpasses Gemini 3.1 Pro in coding, agentic, and multimodal benchmarks, while running at 4x faster output speed and at Flash-tier pricing. That's a meaningful jump: you get a better-than-Pro model for less money and lower latency.

It's already rolling out in the Gemini app, Google Search AI Mode, Antigravity 2.0, and the Gemini API. Gemini 3.5 Pro is in testing and lands next month.

There's also Gemini Omni, a new model series that accepts image, audio, video, and text as input and outputs video. It's designed for creation workflows (Google Flow, YouTube Shorts) rather than pure reasoning tasks.

Antigravity 2.0: Google's Agent Platform Gets Serious

Antigravity is Google's agent-first development platform, and 2.0 is a major upgrade. The key additions:

Specialized subagents: you can now spin up dedicated subagents to handle complex sub-tasks within a larger workflow, all from the Antigravity CLI or the web surface.
Cross-platform terminal sandboxing: agents run in isolated environments with credential masking and hardened Git policies built in — no more worrying about an agent accidentally leaking a secret or pushing to the wrong branch.
Antigravity SDK: programmatic control over the agent harness, so you can deploy your own customized version on your infrastructure.

Google AI Studio now also lets you one-click deploy to Cloud Run and integrates with Firebase services, so you can go from prompt to production-ready full-stack app without leaving the browser. Export to Antigravity when you're ready to continue locally.

WebMCP: The Browser Finally Gets an Agent Interface

This one is significant. WebMCP is a proposed open web standard that lets developers expose structured tools — JavaScript functions, HTML forms — so that browser-based AI agents can interact with your web app precisely and reliably.

Think of it as MCP (Model Context Protocol) but for the browser. Instead of an agent clicking around and hoping it finds the right button, your app explicitly declares what actions are available and how to invoke them. The experimental origin trial starts in Chrome 149.

If WebMCP becomes a real standard (which Google's backing suggests it will), it changes how you think about web app architecture: your UI components aren't just for humans anymore.

Modern Web Guidance: Expert-Vetted Skills for Your Coding Agent

Modern Web Guidance is a set of over 100 expert-vetted skills that you inject into your coding agent to get better, more standards-compliant output — accessible, performant, secure. It integrates with Baseline (web platform feature support tracking) so the agent knows which APIs are safe to use for your target browsers.

Install with one click in Antigravity or via:

npx modern-web-guidance install

Chrome DevTools for Agents

Chrome DevTools for Agents brings DevTools capabilities to AI agents: automated quality audits, real-world user experience emulation, session hand-off with auto-connect, and real-time debugging. Your agent can now run Lighthouse, inspect the DOM, and debug rendering issues without you touching DevTools manually.

Android CLI, Android Bench, and Migration Agent

Three big announcements for Android developers:

Android CLI (stable): AI agents can now tap directly into Android Studio capabilities — download the SDK, run apps on devices, handle build tasks. Works with any agent, LLM, or tool.
Android Bench: a leaderboard that ranks LLMs on Android development tasks. Gemma 4 open-weight models were added this week. Useful for choosing which model to use for Android-specific agents.
Migration Agent: converts React Native, web frameworks, or iOS apps to native Kotlin Android apps. Google claims weeks-long migrations become hours.

HTML-in-Canvas: A New API Worth Watching

HTML-in-Canvas (origin trial in Chrome) lets you embed real DOM elements inside a WebGL/WebGPU canvas. The result: immersive 3D experiences that remain fully accessible, searchable, and interactable — all declared via a new declarative API. Early days, but worth adding to your radar if you work on data visualization, games, or creative tools.

Gemini Spark and Android Halo: The User-Facing Side

On the consumer side, Gemini Spark is a personal agent that takes real actions across Gmail, Docs, Calendar, and (coming soon) third-party apps via MCP. Android Halo gives users at-a-glance visibility into what agents are doing via a subtle overlay at the top of the screen. Both are relevant for developers building on top of the Gemini ecosystem — Spark uses MCP for third-party integrations, which means your app can be Spark-compatible with the right server setup.

Resources

Here's the full developer keynote if you want to see it all in one shot:

Source: Google Developers — I/O 2026 Developer Keynote

The Bottom Line

Google I/O 2026 is not a "wait and see" event. Antigravity 2.0 and the Gemini 3.5 series are available now. WebMCP and Modern Web Guidance are in early preview but worth experimenting with immediately. If you haven't tried building with the Gemini API in the last six months, the platform looks meaningfully different today than it did.