Google wants to turn Gemini into a personal AI assistant

The main announcement of Google I/O is Gemini Spark, a cloud-based AI agent. It should parse emails, chats, and meeting notes, highlight what's important, create to-do lists, and continue working even when the user has closed the laptop or locked the phone. The company calls it a move into the "agent" era of Gemini.

But we're not talking about a mass launch for everyone just yet. Gemini Spark will first go to individual testers and then to Google AI Ultra subscribers in the US in beta mode. For high-risk actions, such as sending an email or making a purchase, the agent must ask for the user's permission.

Details

At Google's I/O conference, the company unveiled several updates around Gemini at once. The central idea is to make the AI more autonomous. If a regular chatbot waits for a question, the AI agent should help with routine tasks on its own: summarise the results of a meeting, find important things in correspondence, prepare a document, track a purchase or continue working in the background.

Gemini Spark will run in the cloud. This is an important distinction: according to Google, the agent will be able to continue the task even when the user's device is not active. Later in the summer, the company plans to build Spark directly into Chrome.

Google also unveiled the Gemini 3.5 family of models. The first to be released is Gemini 3.5 Flash, a model focused on speed, programming and agent tasks. According to AP, it will become the default model in the Gemini app and in AI Mode in Google search. The company is still using the Gemini 3.5 Pro version internally and expects to launch next month.

A separate block of announcements has to do with video generation. Gemini Omni is supposed to create and edit video on demand using text, images, video and audio. Google's official notes say Omni is launching in the Gemini app for Google AI subscribers 18 years and older, but feature availability varies by region.

Google promises to tag videos created by Omni with a SynthID digital watermark. The company is also adding content provenance verification to Gemini: the tool should help understand whether a photo or video is created by AI, taken with a camera, or has been edited by AI tools. Later, this check should appear in search via Chrome.

Search will also change. AI Mode will use Gemini 3.5 Flash, and the search bar will become smarter: it should better accept long queries, help formulate questions, and work with different types of input - text, images, videos, files, and even Chrome tabs.

Another household announcement is Universal Cart, the "universal shopping cart." Google wants the user to be able to add items to the basket from search, Gemini, YouTube or Gmail, while AI tracks prices, discounts, cost history and availability. The feature should first appear in search and Gemini in the summer, with YouTube and Gmail coming later.

Google also showed off a new generation of AI glasses. There will be two versions: audio glasses with prompts in the ear and glasses with a display. The audio version should be the first to be released - in the autumn. With Gemini, these glasses will be able to help with navigation, translation, messaging, and other tasks. Google is working on them with Samsung, Gentle Monster, and Warby Parker.

Why it matters

Google is trying to build AI not into one single service, but into the entire everyday digital environment: search, browser, mail, YouTube, shopping, documents, and devices. It's no longer just a competition between models in terms of quality of answers. The big question now is who will be the primary interface between humans and their digital lives.

If Gemini Spark works as Google promises, the user will not only ask questions, but also assign real-world tasks to the AI. This is convenient, but it also raises security requirements. The agent that sees emails, appointments, purchases and documents needs to be transparent: what it reads, what it does on its own, where it stops and when it's obliged to ask permission.

That's why Google's caveat about "high-stakes" actions is important. Sending an email, buying a product, or changing an important document is no longer a chat response, but an action with consequences. An AI error in such a scenario could cost more than a failed text generation.

Background

Google has been rebuilding its products around Gemini for a few years now. Chat and answer generation used to be the main showcase, then AI Mode in search, and now the company is talking more and more about agent-based AI. At Google I/O, Sundar Pichai said the company is in the "agentic Gemini era," but admitted that making agents simultaneously simple, secure, and genuinely useful is still difficult.

This turnaround comes amid huge investments in AI. According to AP, Alphabet's CFO previously told investors that the company's capital expenditure this year could reach $190bn. At the same time, Google says Gemini's audience is growing rapidly: according to Pichai, the app's monthly audience has surpassed 900 million users.

For Google, it's also about protecting its core business - search. The more people get answers through AI assistants, the more the usual model of link searching changes. That's why the company is simultaneously developing Gemini, AI Mode, multimodal search, and shopping features like Universal Cart.

Source

Based on Associated Press Kaitlyn Huamani's coverage of Google I/O 2026 keynote announcements and official Gemini Apps release notes dated 19 May 2026.