Google I/O 2026 made one thing clear: artificial intelligence no longer lives only in a chatbot. Google wants to bring it to Search, Gmail, YouTube, Android, developer tools, science, shopping, and audiovisual creation.
The company presented 100 announcements, including new models, autonomous agents, workplace features, and platforms that promise to turn a written or spoken instruction into an app, a video, a report, or a personalized experience. So, what matters most in this entire rollout?
Gemini 3.5 Flash and Gemini Omni expand Google’s capabilities
Google launched Gemini 3.5 Flash, a model that combines advanced reasoning with a fast response. According to the company, it outperforms Gemini 3.1 Pro in programming and agent-based tasks, such as Terminal-Bench 2.1, where it scored 76.2%, and MCP Atlas, with 83.6%.
The goal is to reduce one of the usual decisions you face when using AI: choosing between quality and speed. Gemini 3.5 Flash aims to offer both, at a lower cost for extended tasks such as developing applications, maintaining codebases, or preparing financial documents.
The model is already available through Google Antigravity, the Gemini API, Google AI Studio, and Android Studio. Google also confirmed that it is working on Gemini 3.5 Pro, which is expected to launch next month.
The other major announcement is Gemini Omni, a multimodal model capable of turning text, image, video, or audio references into a coherent output. Its first application focuses on video generation and editing.
Omni aims to better represent physical elements such as gravity, kinetic energy, and fluids. In simple terms, Google wants generated videos not only to look good, but also to better reflect how objects behave in the real world.
Content created with Omni includes an imperceptible digital watermark called SynthID. It can be checked through Gemini, Search, and Chrome to help identify whether an image, audio clip, or video was generated with AI.
Search becomes a more conversational and personalized space
Google Search also received one of its biggest transformations in more than 25 years. The new experience lets you search by combining text, images, files, videos, and open tabs in Chrome.
Google is bringing together AI Overviews, its AI-generated summaries, and AI Mode, so you can start with a question, review a linked summary, and continue the conversation without jumping between different experiences.
AI Mode will use Gemini 3.5 Flash as its default model and, according to Google, already has more than 1 billion monthly users. The company says queries in this mode have doubled quarter after quarter since its launch.
Information agents and personalized mini-apps
Search will begin allowing the creation of information agents. These agents will be able to monitor specific topics throughout the day, checking news, blogs, social networks, and updated data about finance, shopping, or sports.
Imagine you want to track the price of a flight, news in an industry, or the results of a tournament. Instead of repeating the search, the agent could monitor changes and send you a summarized update with possible actions.
The feature will begin rolling out in the Northern Hemisphere summer to Google AI Pro and Ultra subscribers. Google is also preparing a feature for generating personalized interfaces, such as tables, charts, simulations, and dashboards, directly from Search.
The idea goes beyond answering a single question. You will be able to create small applications for recurring tasks, such as organizing a move, planning a wedding, or managing a personal project.
Shopping, Gmail, and Gemini get agents for everyday tasks
Google introduced Universal Cart, a shopping cart that will bring together products found in Search, Gemini, YouTube, and Gmail. The system will be able to find discounts, monitor price drops, notify you when an item is back in stock, and show price history.
It will also try to detect incompatibilities between products and recommend alternatives. Because it is connected to Google Wallet, it can consider card benefits, loyalty programs, and retailer offers when suggesting a payment method.
Universal Cart will begin rolling out in Search and Gemini during the summer, with YouTube and Gmail joining later. The Universal Commerce protocol will allow some purchases to be completed within Google or sent to the retailer’s website.
In the Gemini app, Google introduced Gemini Spark, a personal agent that can work in the background on your phone or computer, even when they are turned off. Its job would be to carry out tasks based on your instructions and request confirmation before performing important actions.
Spark is still in an early stage and will begin with trusted testers. Google plans to add direct messages, subagents, and authorized payments with budget and retailer limits.
Another feature is Daily Brief, a personalized summary that reviews your email, calendar, and tasks to help prepare your day. It is initially available to Google AI subscribers over 18 in the United States who have connected their Google apps.
A more visual and natural Gemini
Google redesigned the Gemini app with an experience called Neural Expressive. It includes fluid animations, new colors, typefaces, haptic feedback, and responses that appear in a more structured way.
Instead of receiving a wall of text, you might find interactive images, timelines, and visual elements built into the response. Gemini Live was also redesigned to open immediately and understand your voice better, even with background noise.
The company said that choosing regional dialects will soon be possible. This detail may seem minor, but an AI that speaks in a more natural and familiar way is much more useful for people with different accents and ways of expressing themselves.
Google Antigravity brings agents to software development
For developers and entrepreneurs, one of the central announcements was Google Antigravity, an agent-oriented development platform. Its new version allows several agents to coordinate and work in parallel—for example, while one programs a website, another can prepare the brand’s visual assets.
Antigravity 2.0 will be a desktop application, while Antigravity CLI will let you work from the terminal. Google also introduced an SDK for customizing agents and hosting them on your own infrastructure.
The platform includes subagents, asynchronous tasks, connections with Android, Firebase, and Google AI Studio, as well as native support for audio models. Google says certain engineering tasks that once took days can now be completed in hours or minutes, although the result will still depend on supervision and the quality of the instructions.
Prototypes and Android apps from AI Studio
Google AI Studio will let you turn an idea into a functional prototype. The tool can connect with Workspace to create dashboards from Sheets data, organize Drive files, or work with a team’s documents.
You will also be able to create Android apps from the build tab, test them in an emulator inside the browser, and publish them to Google Play’s internal testing channel.
Users who start in AI Studio will be able to deploy their first two applications on Google Cloud at no cost and without a credit card. They can also export the project to Antigravity along with the conversation history, files, and configuration secrets.
Google also announced Managed Agents for the Gemini API. With a single call, the service can create an isolated Linux environment where an agent reasons, runs code, manages files, and retrieves current data from the web.
More AI for work and creativity
Gmail is expanding AI Inbox, a view that prioritizes important tasks and messages. It can now generate personalized draft replies, show the Docs, Sheets, or Slides file you need to review next to a task, and help you clear out suggestions that are not useful.
During the summer, Pro and Ultra subscribers will be able to chat with their inbox through Gmail Live, without having to manually search through long email threads.
Google also introduced Google Pics, an image creation and editing tool based on Nano Banana. It includes precise object selection, text editing and translation, as well as Workspace integrations. It will first be available to a limited group of testers and will then reach Pro and Ultra subscribers.
With Docs Live, users will be able to dictate a document and ask AI to organize ideas, structure the content, and consult relevant information from Gmail, Drive, Chat, or the web, always with authorization. In Google Keep, a similar feature will turn spoken ramblings into organized notes and lists.
Video, music, and design with Google Flow
Google Flow, its image and video creative studio, received an agent capable of carrying out multistep tasks. It can help develop dialogue, suggest script changes, create variations, edit several assets at the same time, and organize a project’s files.
The platform also adds Flow Tools, which lets you create custom tools using natural language. You could describe a video editor, an image resizer, or a visual effect and let AI build a first version without writing code.
With Flow Music, you can edit specific parts of a song, change lyrics into another language, modify the genre, or adjust instruments. In addition, Pomelli adds features for creating branded content and designing websites, while Stitch improves collaboration between spoken instructions, designs, and existing codebases.
Gemini Omni Flash is also coming to YouTube Shorts Remix. Users will be able to select an eligible Short, request changes, and add their own image or a visual reference to create a new version.
AI to accelerate scientific research
Google introduced Gemini for Science, a collection of experimental tools for different stages of research. They include:
- Hypothesis Generation, which uses several agents to propose, debate, and evaluate hypotheses with verifiable citations.
- Computational Discovery, which generates and scores thousands of code variations to explore problems such as solar forecasting or epidemiology.
- Literature Insights, which analyzes scientific publications and organizes findings into tables, reports, presentations, infographics, and audiovisual summaries.
Access to these tools began opening on May 19 through Google Labs Science. Google also released Science Skills, a package that connects agents with more than 30 scientific databases and tools, such as UniProt, AlphaFold Database, and InterPro.
The goal is not to replace researchers, but to reduce the manual work that consumes hours. Human review remains essential, especially when a hypothesis could influence medical, environmental, or technological decisions.
Learning, smart glasses, and content verification
YouTube is testing Ask YouTube, a conversational search feature that brings together long videos and Shorts to answer complex questions. For example, you could ask for tips on teaching a child to ride a bike and receive an organized answer based on different videos.
The experiment will begin in June 2026 for a group of desktop users in the United States who search in English.
For Android XR, Google confirmed that it is working on two types of smart glasses: some focused on audio, offering spoken assistance, and others with a display, capable of showing information at the right moment. The first audio glasses, developed with Gentle Monster, Warby Parker, and Samsung, will arrive in the fall and work with Android and iOS.
Finally, SynthID is expanding its verification to Search and soon to Chrome. Google says the technology has already been used 50 million times worldwide. It will also add support for C2PA content credentials, which let you know whether a file has retained its original state or was modified, and with which tools.
The big news from Google I/O 2026 is not a single application. It is the integration of agents into nearly every digital space where we already work, shop, learn, and create.
The important question will be how much control users retain when these systems act in the background. An agent that organizes your day, shops for you, or modifies an application can save time, but it needs clear permissions, understandable limits, and simple ways to review its decisions.
Google is betting on AI that does not just answer, but also observes, plans, and executes. The challenge for everyone will be learning to use that autonomy without confusing convenience with blind trust.
Original source
https://blog.google/innovation-and-ai/technology/ai/google-io-2026-all-our-announcements