Google I/O 2026: The Agentic Pivot
How Google is betting its distribution advantage on agents that act — and a multimodal layer that lets them see, hear, and create.
Upcoming Events:
AI-Native Founders: Workflows, Wins & What’s Breaking | Time: Friday, May 29, 6:00 PM - 9:00 PM at Stanford | Register: https://luma.com/founde-wkzr
Google’s annual I/O developer conference opened with a keynote that was less a parade of product updates than a declaration of intent. The message was unmistakable: Google believes we have entered the “agentic Gemini era” — a shift in which AI stops being a passive assistant waiting for prompts and becomes a proactive agent that works on the user’s behalf across a redesigned, deeply integrated ecosystem. The agentic push is the headline bet, but it is not the only one. Alongside it, Google is moving aggressively to embed multimodality — voice, image, and video — into nearly every product it ships, from Search and Docs to YouTube and its eyewear line. Together, the two bets define Google’s strategy for the year: agents that act, and a multimodal layer that lets them see, hear, and create.
Part 1: Google’s Real Position — Distribution vs. Mindshare
Google’s own numbers from this year’s I/O tell a striking story about scale. AI Overviews reaches 2.5 billion monthly users, AI Mode 1 billion, and the Gemini app 900 million — figures that dwarf any competitor. But the chart flatters Google more than it should. AI Overviews and AI Mode both live inside Search, a surface 2.6 billion people already use by reflex; those users didn’t choose an AI product, AI was simply placed in front of them. Only the Gemini app reflects a deliberate opt-in choice.
That distinction matters, because it is exactly the gap ChatGPT and Claude are exploiting. ChatGPT has reached roughly 1 billion monthly users with none of Google’s distribution — no browser, no operating system, no Search box — meaning every one of those users arrived on purpose. Claude, at around 30 million consumer users, looks tiny by comparison; yet Anthropic draws roughly 80% of its revenue from enterprise and API customers, a segment where Claude leads. The three companies are, in effect, winning different games: Google wins distribution, ChatGPT wins consumer mindshare, and Claude wins enterprise depth.
It is worth holding that framing in mind throughout the rest of this recap, because nearly every product Google announced is an attempt to convert its distribution advantage into genuine, agentic engagement.
Part 2: What Google Actually Shipped
The keynote unveiled a torrent of new products and features, all powered by the latest Gemini models and a renewed focus on agentic capability. Here is a breakdown of the announcements, grouped by what they actually do.
Core AI Models
Gemini 3.5 Flash: Launched May 19, 2026, optimized for agentic workflows and coding; it became the default model in the Gemini app and AI Mode in Search.
Gemini 3.5 Pro: Announced at I/O, with general availability expected in June 2026.
Gemini Omni: A new multimodal “world model” introduced by DeepMind on May 19, 2026. The first release, Gemini Omni Flash, generates and conversationally edits high-quality video grounded in Gemini’s real-world knowledge, and is rolling out through the Gemini app, Google Flow, and YouTube Shorts.
On the flagship model itself, the headline is price and speed rather than raw benchmark dominance. Gemini 3.5 Flash is meaningfully cheaper and faster than Anthropic’s Claude Opus 4.7 and OpenAI’s GPT-5.5:
Note: Gemini 3.5 Flash is a Flash-tier model, not a direct flagship peer to Opus 4.7 or GPT-5.5. It outperforms last year’s Gemini 3.1 Pro on coding and agentic benchmarks at roughly 4× the speed and a lower price, but 3.1 Pro still leads on pure reasoning and long-context retrieval. Pricing and benchmarks change quickly.
On the multimodal front, Gemini Omni arrives at an interesting moment. OpenAI sunset its Sora video app in April, and Google is positioning Omni as a unified “any-to-any” architecture that folds text, image, audio, and video into a single model. The most credible competition now comes largely from Chinese labs — Seedance, Wan, and MiniMax among them — making multimodal video one of the more globally contested corners of the AI race.
Agentic Tools
The most ambitious agentic announcement was Antigravity 2.0, aimed at developers, alongside Spark, a consumer-facing agent. Google also embedded lighter-weight agents directly into Gemini and Search, including “The Daily Brief” and “Search Agents”.
Antigravity 2.0
Antigravity 2.0 is led by Varun Mohan, the former CEO and co-founder of Windsurf, who joined Google DeepMind in 2025. It is a standalone, agent-first development platform — not merely an IDE — shipping with a desktop app, a CLI, an SDK, and an “agent harness” that lets teams of AI sub-agents tackle complex coding tasks in parallel. In one keynote demo, Antigravity’s agents built a functional operating system core from scratch for under $1,000 in token cost.
A candid note from hands-on use: the experience still involves a fair amount of babysitting. Running multiple agents in parallel often means jumping between them to click “yes” and approve each step — the autonomy is real, but so is the supervision overhead.
Spark — Your 24/7 Personal AI Agent
Gemini Spark is Google’s 24/7 personal AI agent: it takes action on the user’s behalf, under their direction, and keeps working in the background even while a phone or laptop is turned off. Built on Gemini 3.5 and the Antigravity agent harness, it runs on dedicated cloud VMs, integrates natively with Gmail and Workspace, and can browse the web through Chrome — executing long-running tasks without tying up the user’s device.
Spark is rolling out to trusted testers first, with a Beta arriving for Google AI Ultra subscribers in the U.S. Notably, Google repriced the AI Ultra plan at I/O 2026 from $250 to $100 per month — and Spark is included with it.
The Daily Brief
A new agent in the Gemini app that produces a personalized morning digest, synthesizing key information from a user’s inbox, calendar, and tasks. Available on paid personal accounts, it is especially useful for busy people juggling family and work commitments.
Search Agents
A feature that lets users deploy autonomous agents to monitor the web around the clock for specific information — from financial data to apartment listings — and deliver synthesized updates as things change.
Multimodal Across the Ecosystem
Google is weaving multimodality — voice, image, and video — throughout its products, including Flow, Search, Docs, YouTube, and its eyewear line.
Google Flow: The creative platform for artists is updated with Gemini Omni for advanced video editing, a new agent for multi-action tasks, and custom “Flow Tools.”
Intelligent Search Box: A redesigned, expandable search bar that uses AI to help users formulate complex, multimodal questions across text, images, and video.
Generative UI in Search: Search can now use agentic coding to build custom, interactive interfaces and visualizations on the fly, creating a unique experience for each query.
Docs Live: Lets users verbally brainstorm and dictate content while Gemini structures, formats, and assembles a Google Doc in real time.
Ask YouTube: A conversational feature that answers complex questions with digestible summaries and jumps directly to the most relevant segments of a video.
Intelligent Eyewear: The next frontier for Android XR, in two forms — Display Glasses with a small in-lens display for glanceable information, and Audio Glasses (launching this fall, with designs from Warby Parker and Gentle Monster) offering hands-free, spoken Gemini assistance for navigation, ordering coffee, and AI-enhanced photos.
AI for Good & Science
SynthID & Content Credentials: Google’s AI watermarking and verification tools are expanding to Search and Chrome to promote transparency, with new partners — including OpenAI — signing on.
Code Mender: A security agent that automatically finds and fixes vulnerabilities in software code.
Gemini for Science: A suite of tools to accelerate scientific discovery, including AI simulations such as Alpha Earth (a “digital twin of the planet”) and Weather Next, a high-accuracy hurricane forecasting model.
Taken together, I/O 2026 reads as a single coordinated bet: one model family (Gemini 3.5), one agent harness (Antigravity), powering both a consumer agent (Spark) and a developer platform (Antigravity 2.0), all riding the distribution Google already owns. Whether that bet pays off depends on the question the opening chart raised — whether built-in reach can be converted into genuine preference before ChatGPT’s opt-in momentum and Claude’s enterprise depth close the gap.