Google doesn't do small announcements anymore. At I/O 2026 in Mountain View, the company dropped more new products in a single two-hour keynote than most companies release in a year — and a significant chunk of them are already live. The thread running through all of it: Gemini is no longer a chatbot. It's becoming the operating layer of everything Google makes.
Sundar Pichai opened with a scale check that few companies on earth could match. Two years ago, Google was processing 9.7 trillion tokens a month. Last year that climbed to roughly 480 trillion. Today the number is over 3.2 quadrillion per month — a 7x jump in a single year — with over 8.5 million developers now building with Google's models monthly. That trajectory is the context for everything announced on stage.
Gemini Spark: Your First Real AI Agent
Gemini Spark is described as "your personal agent" that takes actions on your behalf to help "navigate your digital life." Google calls it a big shift — transforming Gemini from an assistant that answers questions into an active partner that does real work under your direction. It integrates with Gmail, Docs, and other Google Workspace apps first, before expanding to third-party tools via MCP over the summer.
Spark runs on private Google Cloud servers in the background and keeps working even when you're not actively using your phone. The live demo showed it planning a block party — creating an RSVP tracker in Google Sheets, auto-updating Docs, and sending follow-up email reminders to people who hadn't responded.
You can work with Spark however is most convenient: in the Gemini app, or soon through email and chat. Later this summer, Spark will also operate directly within Chrome, acting as your agentic browser assistant. The beta rolls out to Google AI Ultra subscribers in the US next week.
Android Halo: The Status Bar That Knows What Your Agent Is Doing
Android Halo provides at-a-glance visibility into what your agent is working on at any given time, with subtle communication at the top of your phone screen — meaning you can see an agent's progress without having to stop what you're doing or switch apps.
The only visual cue shown so far is a glowing circle morphing into the Gemini sparkle in the upper-left corner of a Pixel phone's status bar. It's coming in Android 17, with more details expected later this year when new Pixel hardware arrives.
Search Gets Its Biggest Upgrade in Nearly 30 Years
By combining with Gemini Spark, questions you ask in Search can now be agentic — instead of returning a snapshot of information at that moment, Search can give you ongoing updates in the future. Searching is now an AI function, not just a text input.
Google is also combining AI Overviews and AI Mode into a more unified experience, letting users move seamlessly between traditional results, AI-generated answers, and follow-up conversations without losing context. The deeper you go into a conversation, the more relevant links and sources become. The updated experience is rolling out globally today.
Gemini 3.5 Flash and Omni: The Engine Room
Gemini Omni is a new series of models that combines Gemini's reasoning with creation — accepting image, audio, video, and text as input and generating video output grounded in real-world physics and knowledge. Omni Flash is live today for Google AI Plus, Pro, and Ultra subscribers, as well as YouTube Shorts and YouTube Create users at no extra cost.
Antigravity 2.0: Coding by Agent
Antigravity will now use Gemini 3.5 Flash and allow for faster development cycles. At I/O, Google announced a standalone desktop application dubbed Antigravity 2.0 and a new command-line interface for developers who prefer staying in the terminal, along with new Google Cloud standard privacy protections.
The platform now includes native voice support and integrations with Android, Firebase, and Google AI Studio, and is described as "unabashedly agent-first."
CodeMender: The AI That Patches Your Security Holes Automatically
This is the announcement that didn't get nearly enough attention in the keynote recaps. CodeMender is a new AI-powered agent developed by Google DeepMind that takes a comprehensive approach to code security that's both reactive — instantly patching new vulnerabilities — and proactive, rewriting existing code to eliminate entire classes of flaws. Over the six months Google spent building it, CodeMender already upstreamed 72 security fixes to open-source projects, including codebases as large as 4.5 million lines.
The tool was built using Google's learnings from BigSleep and OSS-Fuzz. It relies on Gemini for root cause analysis, after which it produces security patches that are peer-reviewed by specialized "critique agents" before reaching a human reviewer for final sign-off.
CodeMender is now available as a powerful AI security agent through Google's Agent Platform, letting any developer — not just Google's own teams — benefit from autonomous vulnerability detection and patching. Alongside CodeMender, Google also announced a dedicated AI Vulnerability Reward Program with bounties up to $30,000, and Secure AI Framework 2.0, an updated set of industry standards for securing autonomous AI agents.
Google Pics, Stitch, and Pomelli: The Creative Triple Threat
Google Pics is a new image creation tool inside Google Flow that lets you generate images on the fly with AI, all automatically watermarked with SynthID. Stitch lets people create and launch websites using AI, with simple voice inputs to make changes and guide layouts in real time.
Pomelli goes a step further — introducing AI agents that can help you design your brand book and launch a full website, adding new ways to build brand content and entire web presences from scratch. Together, the three tools represent Google's most direct move yet into the territory occupied by Canva, Squarespace, and similar creative platforms.
Universal Cart and the End of Tab-Switching While Shopping
Google has partnered with Amazon, Shopify, and Walmart on a new open standard called the Universal Commerce Protocol, designed to unify digital commerce so AI agents can browse inventories and handle entire purchases without hard-coded integrations for each merchant.
The Universal Cart follows you across Google services, notifying you when items go on sale or come back in stock. In one demo, it flagged that a chosen motherboard and processor were incompatible and recommended a replacement that actually worked — then added everything to the cart automatically. It arrives in the US this summer.
Gemini for Science, Project Genie, and the Long Game
Gemini for Science will bring together powerful AI tools to assist researchers and help scientists model complex concepts, framed as "a force multiplier for human ingenuity to usher in a new age of progress." Project Genie is being connected with nearly 20 years of Google Street View imagery, letting people create new virtual worlds anchored in real locations. These two announcements sit at the edge of the consumer keynote but point to where Google's deepest research bets are heading.
Daily Brief, Ask YouTube, and Docs Live
Daily Brief is a personalized digest that sifts through your Gmail, Calendar, and Tasks to prioritize your day and suggest next steps. It's rolling out today to Google AI Plus, Pro, and Ultra subscribers in the US. Ask YouTube expands more widely this summer, letting you use YouTube search as an AI chatbot that points you to exact timestamps in videos that answer your questions. Docs Live, coming this summer to AI Pro and Ultra subscribers, lets you verbally brain-dump whatever's on your mind — complete with "ums" and mid-sentence changes — and Gemini converts it into a finished, structured document.
The Bigger Bet
Google is looking to spend six times more on AI in 2026 than it did in 2022, with an estimated $190 billion in capital expenditures for the year alone.
That's not hedging. Everything shown at I/O 2026 — from Spark running quietly in the cloud, to CodeMender patching your repositories overnight, to Halo glowing in your status bar — is Google betting that the next platform isn't an operating system or a search box. It's an agent that already knows what you need before you type it.