Google will quickly unleash a wealth of latest synthetic intelligence-powered instruments and methods, together with an AI assistant that can assist customers by proactively performing duties on their behalf.
“Agentic” AI, the current buzzword of alternative for tech companies, was a central focus of Google’s annual builders convention, Google I/O. The upcoming AI agent, Gemini Spark, was one in all lots of the firm’s bulletins from the convention Tuesday.
“We’re firmly in our agentic Gemini period,” Google CEO Sundar Pichai stated Tuesday earlier than a packed amphitheater close to the corporate’s Mountain View, California, headquarters. “I’ve performed round with all kinds of brokers and you’ll actually see the potential, however it’s nonetheless early days with regards to making brokers straightforward to make use of, tremendous safe and really useful.”
Google and its company mum or dad, Alphabet Inc, have poured billions into AI improvement. Its high finance government stated on a name with buyers in late April that this yr’s capital expenditures might climb as excessive as $190 billion. However the funding appears to be paying off, with its quarterly earnings exhibiting robust progress. The inventory has climbed one other 11% for the reason that report final month.
Pichai stated in the course of the keynote tackle that the Gemini app had 400 million month-to-month energetic customers final yr, however that usership has now surpassed 900 million, greater than doubling in a yr.
Google’s newest household of fashions, Gemini 3.5, is rolling out Tuesday to billions of worldwide customers starting with Gemini 3.5 Flash. The Flash mannequin is targeted on velocity, and Google says 3.5 Flash is its strongest agentic and coding mannequin but, however it’s additionally about 4 instances sooner than some opponents.
This mannequin is now the default for the Gemini app and “AI mode” on Google search. The corporate can also be engaged on the three.5 model of Gemini Professional, which it says it is utilizing internally and expects to launch subsequent month.
Gemini 3.5 was developed with new, extra superior security coaching and mitigations, which means its fashions are much less more likely to generate dangerous content material or to mistakenly refuse to reply secure queries, the corporate stated.
Google additionally introduced a brand new mannequin, Gemini Omni, which is able to allow customers to create high-quality video by making a question with any enter, be it textual content, photos, movies and audio. The video Omni creates can then be edited simply although a dialog with the mannequin. Customers will finally be capable to create photos and audio with Omni, however there have been no particulars about when these options will likely be rolled out.
The corporate stated Omni’s movies will seem extra real looking than movies created by different fashions due to its understanding of forces like gravity, kinetic vitality and fluid dynamics.
Gemini Omni Flash, the primary of the Omni household, is launching Tuesday for Google Al Plus, Professional and Extremely subscribers by the Gemini app and Google Circulation. Starting this week, will probably be obtainable for gratis on YouTube Shorts and YouTube Create App.
All movies created with Omni will embrace Google’s imperceptible digital watermark, SynthID, however Google can also be including content material credentials verification to the Gemini app. This software determines if content material like photograph or video was created by AI or captured with a telephone digicam and edited with AI instruments. It is going to be obtainable in search in Chrome within the coming months. Google additionally introduced AI corporations Open AI, Kakao and Eleven Labs are adopting its SynthID expertise to extra of their AI-generated content material.
Powered by Gemini 3.5, Gemini Spark will be capable to full mundane, routine duties like sorting by assembly notes, emails and chats after which making a doc with the most important takeaways and to-dos. In contrast to different obtainable brokers, Spark is predicated within the cloud, so it continues working within the background even when customers shut their laptops or lock their telephones.
The proactive nature of AI brokers is what differentiates them from chatbots, and that has additionally led to some anxieties concerning the expertise’s energy. Gemini Spark is designed to ask for permission earlier than performing “high-stakes” duties like sending an e mail or making a purchase order, the corporate stated.
Choose testers can have entry to the agent starting Tuesday, and the corporate plans to roll out the beta mode to U.S.-based subscribers to its Google AI Extremely tier.
Later this summer season, Gemini Spark will function immediately inside Chrome, the corporate stated.
Shahram Izadi, of Android XR Platform & Merchandise, speaks about clever eyewear whereas talking at a Google I/O occasion in Mountain View, Calif., Tuesday, Might 19, 2026. (AP Picture/Jeff Chiu)
Among the many many AI-centric bulletins on the convention was an replace on the long-awaited sensible glasses from Google, of which there will likely be two sorts: audio glasses that supply spoken assist in your ear, and show glasses that present data visually. The audio glasses will come first, with the corporate anticipating them to reach later this fall. Customers will be capable to say “Hey Google” or faucet the facet of the body to entry Gemini, which is able to then help with navigation, managing communication on their telephone, real-time translations and different duties.
Google partnered with Samsung and eyewear manufacturers Light Monster and Warby Parker to create the glasses and gave the primary take a look at two designs on Tuesday, with sun shades from Light Monster and glasses from Warby Parker. These designs will launch as a part of the eyewear manufacturers’ full collections later this yr, Google stated.
Ultimately yr’s convention, probably the most talked-about improvement was the introduction and rollout of “AI mode” on Google’s search engine. The characteristic provides customers a extra conversational reply to their question earlier than offering related hyperlinks, constructing on beforehand applied modified how customers expertise and work together with the platform.
AI mode queries have greater than doubled each quarter since its launch final yr, and the software lately surpassed 1 billion month-to-month customers, in line with Liz Reid, Google’s head of search.
The brand new default mannequin in search will now be Gemini 3.5 Flash and the corporate is introducing what it calls an clever search field. This variation, which Reid says is the most important improve to the search field in 25 years, means the field will adapt to accommodate longer queries and it may well assist customers write out their questions with AI-powered ideas as a substitute of conventional autocomplete.
Customers may also search utilizing a number of modalities, utilizing textual content, photos, video, information and even Chrome tabs as search inputs. The brand new search field is beginning its roll out Tuesday in all nations and languages the place AI mode is presently obtainable.
The corporate additionally introduced a brand new software, the Common Cart, which it known as “a really clever buying cart.” It really works throughout retailers and throughout providers so customers can add issues to their cart whereas shopping Google search, chatting with Gemini, watching YouTube, or studying emails in Gmail. The cart then runs on Gemini fashions to go to work as quickly as an merchandise is positioned within the cart, searching for offers and value drops, offering value historical past data and alerting customers when one thing comes again in inventory.
The Common Cart software will likely be obtainable to customers on search and the Gemini app this summer season, with YouTube and Gmail to observe.
——
Related Press Author Barbara Ortutay in Oakland, California contributed to this story.
© Copyright 2026 The Related Press. All rights reserved. This materials will not be printed, broadcast, rewritten or redistributed with out permission.
