Gemini agents were the central theme at Google I/O 2026, as Google outlined plans to move AI out of a single chat window and into Search, apps, shopping and wearable displays, the company said at its keynote.
Core focus: Gemini agents
Google described a unified system it calls a Gemini agent architecture, built from new models, platform APIs and product integrations intended to let AI act proactively across multiple services, Google said.
That strategy has two technical pillars: the new Gemini 3.5 family, led by Gemini 3.5 Flash, and a world model for video called Gemini Omni. Google said both will be embedded across Search, the Gemini App, Workspace, Shopping and Maps to give those services agentic capabilities, where the system can act without waiting for a user prompt.
1. Gemini 3.5 Flash, the new engine
Google said Gemini 3.5 Flash is the first model in the 3.5 family and that it is optimized for sustained, multi-step agent tasks and software development workflows.
According to Google, Flash outperforms the prior Gemini 3.1 Pro on coding, long-horizon planning and other high-value benchmarks, and sits in the companys “high intelligence, low latency” quadrant on internal Artificial Analysis metrics. Google said Flash outputs tokens about four times faster than other leading frontier models, which lets the company support more concurrent agents on the same hardware budget.
Google said it will make 3.5 Flash the default engine across first-party products, including the Gemini App, AI Mode in Search, the Antigravity developer platform and its enterprise offering, Gemini Enterprise. That means many of the new agent features users see in Search or the Gemini App will be powered by Flash.
2. Gemini Omni, a world model for video
If Flash is the systems brain, Gemini Omni is the sensory layer Google said it built to understand and rewrite real-world visual content from text, images, audio and existing video.
Google announced an initial video-focused release called Gemini Omni Flash. Users will be able to upload a phone video and ask in natural language to change lighting, swap backgrounds, alter camera framing or shift the story timeline, while Google said Omni will maintain character consistency, plausible physics and visual continuity across edits.
Google said Omni Flash will be available first to Gemini AI Plus, Pro and Ultra subscribers via the Gemini App and Google Flow, and that creators can use it for YouTube Shorts and the YouTube Create App. Google said Omni-generated video will carry a SynthID watermark, and that Content Credentials checks will be available within the Gemini App, Chrome and Google Search. The company plans later API access for developers and enterprises.
3. Gemini App, Daily Brief and Spark turn assistants into 24/7 agents
The Gemini App received a major redesign, Google said, adopting a new “Neural Expressive” design language with more fluid animation, updated color and layout, and Gemini Live for real-time voice conversation, all intended to make interactions feel closer to a live exchange than a text box.
Daily Brief is an agent Google described as an AI morning briefing. With user permission, it will monitor Gmail, Calendar and Tasks in the background and surface urgent emails, key meetings and prioritized next steps in a concise morning digest, Google said.
Gemini Spark is a personal agent Google said will run in Google Cloud and use Gemini 3.5 with Antigravity harnessing to perform extended multi-step jobs, such as flagging new subscriptions on a credit card statement, summarizing meeting notes into Google Docs, and, via the Multi-Connector Platform, interacting with third parties like Canva, OpenTable and Instacart to complete tasks on a users behalf.
Google said Spark will roll out first to trusted testers, then to Google AI Ultra subscribers in the United States as a beta, and that it aims to ship an “agentic browser” version in Chrome later this summer.
4. AI search: information agents and generative UI
Search remains the core of Googles ecosystem, and Google said it is moving Search into what it called an agentic era by running AI Mode on Gemini 3.5 Flash to deliver faster, longer-context and more complex reasoning.
Google described new “information agents” that let users create a dedicated agent for topics such as house hunting, tracking stocks or following athlete sneaker drops; those agents will monitor information in the background and notify users when updates meet set criteria. Search will also produce dynamic, generative user interfaces that can include interactive visuals, tables and simulation tools, and can persist as dashboards or trackers tailored to a users long-term tasks, Google said.
Google said generative UI features will be provided free to all Search users this summer, while custom dashboards and Antigravity-built experiences will be available initially to U.S. Google AI Pro and Ultra subscribers.
5. Universal Cart, UCP and AP2 reshape online shopping
On e-commerce, Google introduced Universal Cart, which it described as a single smart cart that can collect items found across Search, the Gemini App, YouTube and Gmail. Google said users can add products from any of those surfaces into one cart.
Google said the cart will automatically search for discounts, track price history, monitor stock levels and, for complex builds such as custom PCs, check component compatibility and suggest alternatives. Built on Google Wallet, the cart will factor in credit card rewards, loyalty points and merchant offers to recommend the lowest-cost payment option, the company said.
The underlying standard for agent and merchant communication is the Universal Commerce Protocol or UCP, while the Agent Payments Protocol, AP2, sets clear payment boundaries so users can limit agents to specific merchants or amounts. Google said Universal Cart will launch in Search and the Gemini App in the United States this summer with one-click checkout at merchants including Nike, Sephora, Target, Walmart, Wayfair and many Shopify stores.
6. Smart glasses and Project Genie bring AI off screen
Hardware updates include a pair of audio glasses for voice assistance and display glasses that can overlay information in the users field of view, Google said. Both device classes will connect to Gemini so users can ask for directions, look up information or receive contextual prompts while keeping their hands free and eyes forward.
Google said audio glasses are expected to ship in the fall. The company also previewed Project Genie, which uses Street View data combined with generative AI to turn real neighborhoods into editable three-dimensional sandboxes developers, planners and designers can use to prototype changes to buildings, streets and public space.
Google said the company envisions workflows where a designer edits a street in Genie on a desktop, then wears display glasses on site to view a future overlay, or where game studios prototype levels in Genie and use Omni to generate promotional footage.
7. Content Credentials and SynthID, infrastructure for transparency
To make synthetic media easier to identify, Google said it is expanding SynthID watermarking and Content Credentials checks into Search and Chrome. Google said SynthID, introduced three years ago, has already been applied to more than 100 billion images and video frames and to audio material equivalent to 60,000 years in length.
Google said it will provide detection tools in the Gemini App and extend Content Credentials so users can check whether an image or video was generated or edited by a generative tool. Google said partners including Nvidia have already adopted the standard, and that OpenAI, Kakao and ElevenLabs announced plans to adopt SynthID as part of a cross-platform effort to set AI content standards.
How much of your life will you hand to Gemini agents?
What emerged from I/O is less a pile of discrete features and more a single, integrated agent system: Search that tracks information long term, the Gemini App with Daily Brief and Spark to handle day-to-day chores, Universal Cart to advise on purchases, and smart glasses and Project Genie to overlay those agent capabilities on the physical world, Google said.
The practical question for users is how much they are willing and comfortable delegating to a persistent cloud agent. That judgement, Google suggested, will become clearer once users begin trying AI Search and the new Gemini App features in real workflows.
