By Radhika Rajkumar, Senior Editor / May 19 at 2:45 p.m. ET

Hassabis returned to the stage to close the keynote with a plug for Google's progress on AGI, announcing a new suite of Google AI tools designed to advance scientific discovery. Scientists can use these tools to create digital twins of the Earth, for example, simulating real-world scenarios for better climate change and weather predictions. 

Like other AI labs, Hassabis also mentioned drug discovery as a fundamental frontier of AI for science that can actionably change lives. He said we'll look back on this moment as being in "the foothills of the singularity." Over the past few years, OpenAI has also invested heavily in scientific applications of its AI. 

By Kerry Wan, Editor in Chief / May 19 at 2:35 p.m. ET

Finally, some hardware! Here's our first look at the newest Android XR glasses, made in partnership with Warby Parker and Gentle Monster. The branded wearables, which Samsung and Google call intelligent eyewear, are the first Android XR glasses that consumers will be able to buy later this year.

From the looks of them, there's a style, finish, and budget for different users.

By Radhika Rajkumar, Senior Editor / May 19 at 2:30 p.m. ET

Developers got a few new toys for using agents in web apps. Google announced Modern Web Guidance, currently in preview, a new set of skills and resources that guide coding agents to build websites, as well as new Chrome DevTools for agents that are designed to help programmers diagnose issues more seamlessly. (Look for details later this afternoon.)

An honorable mention: the newly available WebMCP lets developers turn their web pages into toolkits for agents, giving them more autonomy (WordPress just launched a similar feature). 

By Radhika Rajkumar, Senior Editor / May 19 at 2:30 p.m. ET

Google Flow, a filmmaker-forward video tool released at last year's I/O, has expanded into a fuller creative studio featuring several new agents accompanying various processes. Google demonstrated how users can interact with Flow Agent, for example, like a creative collaborator, asking it for recommendations and input on dialogue and plot development. With the new Omni Flash model, Google said Flow now preserves character consistency across frames, comparing Omni for video to Nano Banana for images. 

Pics, Google's new image tool, leverages Nano Banana and lets you create all-new graphics or edit existing photos, flyers, and more.

On a related note, Canva launched its Connected App for Gemini, allowing users to create in Canva from within the Gemini app. Just connect Canva to Gemini and type "@Canva" to get started; the integration can even use your chats as design context. Canva already has similar access set up in Claude, ChatGPT, and Copilot, so this move just brings access up to speed for another major chatbot. 

By Radhika Rajkumar, Senior Editor / May 19 at 2:15 p.m. ET

The Gemini app got a "complete redesign," said Woodward, in a language called Neural Expressive. Alongside new typography and colors, the app now features Gemini Live, letting you switch from typing to talking, and supports more languages and regional dialects. You can also pause the mic to talk through a longer request without interruption. Plus, no more wall of text: as of today, Gemini will piece out responses to your queries, complete with graphics and imagery, similar to how the Generative UI feature appears. 

By Radhika Rajkumar, Senior Editor / May 19 at 2:02 p.m. ET

Building on the new AI Search for consumers, Google now lets you customize how your search results appear. Using Antigravity and 3.5 Flash, Generative UI in Search "plans the ideal response from scratch," providing a personalized, interactive version of the info you're seeking. 

The demo showed AI Mode in Search creating a visual, interactive weekend itinerary based on context Search already had about what a user likes, what's already on their calendar, and other personal data (similar to how Gemini Personal Intelligence tailors query responses for you). 

I can see this feature making information sharing more actionable and visually appealing between users, taking what's helpful out of a search query and leveling it up.

By Kerry Wan, Editor in Chief / May 19 at 2:01 p.m. ET

Google's been integrating AI shopping tools for a few years now, but Universal Cart may be the company's most ambitious one yet. As the name implies, it's a shopping agent that follows you throughout Google Search, Gemini, YouTube, and Gmail, and stores, researches, and adds products to a virtual cart when prompted.

Google says the agent will even keep track of price history, deals, and stock, and make suggestions based on products currently in your cart. If all of these claims are as real and reliable as the company says, then the Universal Cart may be a smash-hit AI tool.

By Kerry Wan, Editor in Chief / May 19 at 1:56 p.m. ET

Just a year ago, Google announced the $250 AI Ultra plan, tailored to the most involved developers, technical leads, and enthusiasts. Today, the company is launching a much more affordable package with its $100 tier, which includes the following benefits:

• 5 times the usage limit in the Gemini app and Google Antigravity

• Gemini 3.5 Flash integration for quicker testing

• Priority access to Google Antigravity

• 20TB of cloud storage for larger datasets, codebases, and other assets

• YouTube Premium individual plan, which includes access to YouTube Music streaming

Also: Google overhauls its AI plans - which one should you now choose?

By Radhika Rajkumar, Senior Editor / May 19 at 1:50 p.m. ET

Bad news for those of you annoyed by the AI-ification of Search: there's more! In addition to getting the new Gemini 3.5 Flash as its default model, the Search box itself is getting a substantial upgrade, the biggest in 25 years, according to Google. The new multimodal box, rolling out today, will expand based on the length of your query, reshaping the search experience you're used to around the chatbot experience instead. It'll also give you AI-generated suggestions and pull in tabs, files, text, and other formats as inputs. 

Also: Google's new AI Search box is here - along with agents and 5 more upgrades

That said, AI Mode has tallied up over a billion monthly users in the first year of its life, according to Google, so maybe people are warming up to the next era of search. 

But what's an AI feature without accompanying agents these days? Google also launched Search agents, which can "work for you" in the background like research assistants. Other agents will help you book events and tickets based on Search (both are coming this summer).

By Radhika Rajkumar, Senior Editor / May 19 at 1:45 p.m. ET

Rumors of Google's new "personal AI agent," known as Gemini Spark, leaked a few days ago, but Pichai officially announced it onstage today. Powered by Gemini 3.5, the new always-on agent is designed to proactively take on tasks for you -- with direction, of course. 

"Yes, you can close your laptop," Pichai assured to laughs from the audience. 

In a demo, VP of Google Labs Josh Woodward walked through how Spark helped him organize logistics for a neighborhood party based on predetermined tasks he shared vocally. Some "trusted testers" are getting access to Spark today, but AI Ultra subscribers will get the beta next week. It's unclear why Google chose to build these agentic capabilities into a sub-feature of Gemini as opposed to upgrading the entire assistant. 

By Radhika Rajkumar, Senior Editor / May 19 at 1:27 p.m. ET

Following its new video model, Google CEO Sundar Pichai announced that C2PA content credentials will now be available in the Gemini app to better identify AI-generated content. This joins Google's existing SynthID watermark, which denotes whether content has been created using Google AI (but, importantly, is not the same as a broad-spectrum detector for any AI-generated content).

Also: OpenAI's new image watermarks make it easier to spot AI fakes - here's how

Pichai also announced that Nvidia, OpenAI, and Elevenlabs will be adopting SynthID as well, a rare moment of collaboration amongst competitors. 

By Radhika Rajkumar, Senior Editor / May 19 at 1:25 p.m. ET

Google dropped its latest model family, Gemini 3.5, which the company framed – unsurprisingly – as central to building better agents. The first model available today is Gemini 3.5 Flash, which Google says outperformed Gemini 3.1 Pro on several coding and agentic benchmarks. Like all Flash models, it optimizes the strengths of the overall model family for speed and a cheaper, more lightweight user experience, without sacrificing the ability to handle "long-horizon" agentic tasks, Google noted. 

You can access 3.5 Flash in the Gemini app and AI Mode in Search. It's also ready to use in Antigravity and the Gemini API, as well as the new Gemini Enterprise Agent Platform, which launched at Google Cloud Next last month. 

By Kerry Wan, Editor in Chief / May 19 at 1:20 p.m. ET

Much like Google AI Search, YouTube is also getting conversational search capabilities, so you can type in complex queries (including follow-up ones), and YouTube will put together long and short-form videos with the most relevant information.

The feature will be limited to Premium members for now, but Google says it plans to expand access to all YouTube users in the future.

By Radhika Rajkumar, Senior Editor / May 19 at 1:17 p.m. ET

Demis Hassabis joined the stage to talk about Omni, a new world model that Hassabis likened to progress on artificial general intelligence (AGI). A follow-up to the ever-successful Nano Banana, the new Omni model lets you combine audio, video, text, and image inputs to create and edit videos. Google said Omni can create "anything from any input," setting expectations high for creatives and other users alike, especially considering simple image and audio outputs aren't available yet. 

Also: Google's new Omni AI tool will let you video clone yourself - I'm intrigued (and concerned)

Google said you can count on Omni for realism because it's grounded in Gemini's knowledge bank, pulling from scientific and historical context. If it lives up to its own hype of making natural-language editing seamless, Omni could fill video tool gaps left by the sunsetting of OpenAI's Sora. Like its 3.5 family rollout, Omni Flash is the first iteration available in Gemini, YouTube Shorts, and Google Flow. 

By Radhika Rajkumar, Senior Editor / May 19 at 1:13 p.m. ET

Sundar Pichai announced Docs Live, a new feature that lets you conversationally pull information across notes, emails, docs, and other material into a doc on the fly. In a recorded demo, a user asked Gemini to create a last-minute speech from odds and ends across his Google suite, and got a ready-to-edit document in seconds. The feature will be available for Pro and Ultra subscribers this summer.  

By Radhika Rajkumar, Senior Editor / May 19 at 1:01 p.m. ET

Google opened the keynote with a video heavily highlighting the transformative potential of AI, setting the tone for the rest of today's announcements. 

By Kerry Wan, Editor in Chief / May 19 at 12:00 p.m. ET

Press badges acquired, and we're seated early for today's keynote. It feels noticeably less crowded this year, though we also arrived much earlier and avoided the stream of Waymo traffic. If this year's I/O is like past events, a DJ should show up any minute now.

By Kerry Wan, Editor in Chief / May 19 at 6:44 a.m. ET

At last year's Google I/O, I demoed the company's first pair of Android XR reference glasses. Since then, I've gone through several demos of the mixed reality platform's capabilities, including AR navigation, Gemini prompting, and content capture.

So, what's next? I'd like to see how the company upsells its youngest Android platform to developers, especially given that Apple may be on the cusp of releasing its own pair of AI-infused smart glasses.

Google needs to convince developers that building for Android XR will be worthwhile, and convince consumers that there's more we can do with the form factor. The answer, much like the Googlebook strategy, may be to tap into the expansive network of manufacturers who also want a piece of the smart glasses pie.

By Kerry Wan, Editor in Chief / May 18 at 9:32 p.m. ET

Ahead of Google I/O, the company held its annual "The Android Show" last week, announcing a slew of Android and Chrome features set to roll out later this year. 

The news included Gemini Intelligence, which refers to more advanced AI capabilities such as background task automation and enhanced context tools, more accessible voice inputs, AI-generated widgets, and general video improvements in Android 17.

Also: First look at Googlebook: A premium Chromebook alternative for Android users

Google also announced a new series of laptops called Googlebook, which is separate from Chromebook and Google Books. 

Googlebook is described as more premium, with a more natural integration with Android phones than competing systems. That includes better app mirroring, faster file transferring across different services, and intuitive Gemini tools.

That all said, if Google does announce more Android news at I/O, expect it to be less than usual.

By Kerry Wan, Editor in Chief / May 18 at 8:52 p.m. ET

The ZDNET team has settled into Mountain View, where we'll be reporting on all the latest developer and software news during Google's big event tomorrow. Expect some pre-keynote shenanigans, media members scrambling to find the best angle for photos and videos, and a lot of mentions of AI, again.