Back to Videos

I/O '26 Recap: Everything You Need to Know

Channel Google
Date May 20, 2026
Duration 14:53
Google I/O Gemini AI Agents Google Search
TL;DR

Google's I/O '26 keynote recap is a sweeping AI overhaul across every Google product, anchored by Gemini 3.5 Flash (available today, faster than 3.1 Pro across all benchmarks), the Anti-Gravity agent platform (93 sub-agents built a working OS from scratch in 12 hours), and Gemini Spark — a 24/7 personal AI agent running on dedicated Google Cloud VMs. Google is betting its entire stack on agentic AI: 3.2 quadrillion tokens processed monthly, AI Mode in Search at 1B+ users, and a new agentic commerce stack with a cart that catches PC component incompatibilities before you click buy.

Key Takeaways

Summary

Scale & Momentum

Sundar Pichai opened with scale: token consumption has exploded from 9.7 trillion/month to 3.2 quadrillion/month — a 7x increase. Google now has 13 products with over 1 billion users each, five with more than 3 billion. The Gemini app grew from 400M to 900M monthly active users in a single year. AI Overview serves 2.5 billion monthly users. AI Mode in Search crossed 1 billion monthly users within its first year.

Models

Gemini Omni is a new any-input-to-any-output model combining Gemini's reasoning with Google's generative media models — starting with video and expanding to all modalities over time.

Gemini 3.5 Flash is available today, outperforming 3.1 Pro on almost all benchmarks while running dramatically faster. A live demo showed ~1,500 tokens per second generating a playable Chrome Dino game from a single prompt. Gemini 3.5 Pro arrives next month (currently in internal use, showing "great improvements").

On infrastructure: Jackson Pathways now seamlessly distributes training across more than 1 million TPUs globally, enabling the largest training cluster in the world.

SynthID, Google's AI watermarking system, has now marked 100+ billion images and videos. It's expanding to Search and Chrome, and OpenAI, Kakao, and Eleven Labs are adopting it — a notable cross-industry trust signal.

Coding: Anti-Gravity and the OS Demo

Anti-Gravity is Google's reimagined agent-first development platform. Paired with Gemini 3.5 Flash, it ran an audacious demo: 93 sub-agents worked in parallel for 12 hours, making 15,000+ model requests and processing 2.6 billion tokens — building a working operating system from an empty project. The OS booted and ran Doom after agents autonomously identified and wrote missing video and keyboard drivers.

Anti-Gravity makes Flash 12x faster in agentic workflows (vs. 4x in standard settings), enabled by the new sub-agent teamwork capability.

Agents: Gemini Spark

Gemini Spark is the flagship consumer agent: a 24/7 personal AI running on dedicated Google Cloud VMs, powered by Gemini 3.5 and the Anti-Gravity harness. It runs in the background even when your laptop is closed, performs long-running tasks autonomously, and integrates with Google's own tools today — with third-party tools via MCP coming in weeks.

Demo: planning a block party end-to-end — pulling RSVPs, building a tracking spreadsheet in Sheets, emailing non-responders, all in sequence without user babysitting.

AI-Native Google Search

Search now supports creating and managing multiple AI agents for ongoing tasks. Information agents run 24/7 and deliver synthesized, intelligent updates — e.g., a biotech stock monitor filtering on P/E < 15, positive cash flows, and low debt, pushing alerts the moment the market moves.

Search can now build custom stateful dashboards, trackers, and planners. Demo: a weekend planner that ingests Gmail, Photos, and Calendar, surfacing restaurant reservations beautifully on maps.

Agentic Commerce

Google's agent e-commerce stack has three building blocks:

Google Workspace

Docs Live: Verbally dictate notes and let Gemini structure the document, pulling context from Drive and Gmail simultaneously.

Ask YouTube: Synthesizes answers from YouTube's video library. Demo: how to help a toddler transition from a balance bike to a pedal bike.

Google Pix: New Workspace image creation and editing tool — hover to remove elements, resize, add/edit text, translate, all natively.

Google Flow (launched last year) gained multi-action capability: a single image can now become 16 unique videos from different angles in one agent run.

Daily Brief: A personalized Gemini digest synthesizing inbox, calendar, and tasks — designed as the first stop every morning.

Hardware: Audio Glasses

Google's first audio glasses arrive this fall, built with Samsung and designed by Gentle Monster and Warby Parker. They deliver Gemini assistance spoken privately into the ear — no display. Demo showed Gemini autonomously launching DoorDash, navigating option screens, and placing a coffee order from a voice prompt.

Gemini for Science

A new initiative bringing AI tools to accelerate research: staying on top of newly published papers, transforming research goals into runnable code, and generating new hypotheses. Framed as "a force multiplier for human ingenuity" and "a new golden age of scientific discovery."

Notable Quotes

"We were processing 9.7 trillion tokens a month across the surfaces, and fast forward to today, that number has jumped seven times to 3.2 quadrillion tokens per month."

"AI mode has been a revelation — our biggest upgrade to search ever. In just a year, it's already surpassed 1 billion monthly users."

"Over 12 hours, 93 sub-agents working in parallel made over 15,000 model requests and processed 2.6 billion tokens to take an initially empty project to the core of a functioning operating system."

"Gemini Spark — it's your personal AI agent that helps you navigate your digital life, taking action on your behalf and under your direction. It runs on dedicated virtual machines on Google Cloud, and it's 24/7. And yes, you can close your laptop."

"The whole reason I've worked on AI my entire career was because I saw it as the ultimate tool to advance science and our understanding of the world."

References

From Description

Mentioned in Video