Google I/O 2026 Live Recap: Everything Announced

May 19, 2026 1:37 p.m. PT

The Google I/O keynote has now wrapped! But our I/O live blog is not over yet. We’ll continue updating this as more announcements are made throughout the developer conference.

Google said it’s bringing together AI tools to help accelerate research. Google’s new Gemini lab prototypes can help with daily scientific tasks, such as staying up to date on newly published papers. Google’s hoping to use AI for medical advances as well as weather.

After spending several minutes showing off the look of the audio glasses, it doesn’t seem to me that this demo shows the glasses really fulfilling much of a need. The glasses seem like a glorified microphone that just connects to your phone, and that phone’s doing all of the actual “work” for ordering a cold brew. I could just… pull out the phone and probably do this faster.

Google’s Android XR glasses demo started with a little stagecraft. Gemini used what the glasses were seeing to play entrance music for Nishtha Bhatia, the product lead of AI & Glasses at Google, kicking things off with some Charli XCX.

But the more practical demo came a few minutes later, when Bhatia used the glasses to order coffee. Gemini navigated her to a cafe, opened DoorDash on the phone in her pocket and prepared her usual nitro cold brew order for pickup.

She still had to approve it before Gemini placed the order, but Gemini did most of the the annoying parts: opening DoorDash, finding the drink and tapping through the screens while her phone stayed in her pocket.

Google’s first audio glasses will arrive this fall. During the I/O keynote, Google’s Shahram Izadi said they’re “designed to give you all-day help with Gemini that is spoken into your ear privately rather than shown on a display.”

Using Gemini Omni, Google says creators can now transform a video’s environment, add visual effects and introduce new characters while preserving the original performance.

In the demo, Google started with raw footage of a man walking down a hallway, then moved him through different scenes without changing his movement or pacing. Google also showed the same video reimagined with a puppet-like character with button eyes, showing that Omni can change the look and cast of a scene without starting over from scratch.

The Google I/O keynote’s moving over to Android XR, including two different types of AI glasses.

Over 100 million UI screens have been generated with the collaborative design tool Stitch, and Google adding new ways to design websites and more by using natural language. New features include real-time design and steering. 

Google Pics is a new image creation and editing tool rolling out this summer in Google Workspace. You can use Pics to create party flyers or infographics, starting with a base image, resizing or removing elements and adding text.

As Gemini Omni’s video editing features continue to be unveiled, this demo shows how a single photo can be used to generate 16 other videos featuring a variety of different angles interpreted by AI from the scene.

Google showed off a new Gemini for MacOS feature that lets you select files in Finder, press and hold the function key and dictate what you want Gemini to do with them.

In the demo, Google selected a mix of dog boarding documents, invoices and images, and then asked Gemini to turn the information into a friendly email for a kennel, using a new voice feature. Gemini pulled details from the selected files, organized them into a table and generated the message in Gmail, even correcting the request when the speaker changed the boarding date mid-dictation.

Google says the new voice capabilities and Gemini Spark are coming to the Mac app this summer.

Gemini Spark is going to get even smarter this summer, with features including proactively setting up your Instacart order and more. 

CNET Editor at Large Scott Stein wore Google’s many new smart glasses during demos at Google’s campus for I/O. His takeaway? The smart glasses he tried could be the best out there.

Scott demoed an advanced AR glasses setup, Project Aura, coming this year. Additionally, this fall will see the arrival of wireless AI glasses made with Samsung, Gentle Monster and Warby Parker. You can read about Google’s new smart glasses here.

One of the new Gemini Omni demos shows a video being edited to add special effects to a woman playing guitar.

The Gemini app’s getting a redesign, using a design style that Google’s being called Neural Expressive. This includes how Gemini will look when using the conversational Gemini Live mode.

Google announced Universal Cart, an AI-powered shopping cart that works across Google services, including Search and the Gemini app, with YouTube and Gmail support coming at a later time.

The idea is that you can add products to one cart while browsing Search, chatting with Gemini, watching YouTube or reading Gmail. From there, Google says the cart can track deals, monitor price drops, show price history, alert you when items are back in stock and even flag compatibility issues, like if you’re building a custom PC and a certain part isn’t compatible.

Universal Cart is rolling out in the US across Search and the Gemini app this summer.

Google is boosting shopping and announcing a series of partnerships, including Amazon, that will use the open-source Universal Commerce Protocol — and it’ll expand to new regions outside the US soon. 

Google is updating Search so people can create and manage multiple AI agents directly from the search box.

The company showed how users could ask Search to keep track of ongoing tasks, like monitoring biotech stocks, apartment listings or sneaker drops. Instead of returning a single set of results, these “information agents” can keep working in the background, watch for updates across the web and send alerts when something changes.

Google says the new Search agents will start rolling out this summer.

Google showed how Gemini Spark, its new personal AI agent in the Gemini app, could help plan a block party by working across apps in the background.

In the demo, Spark pulled together RSVPs, tracked who was bringing what, drafted follow-up emails to neighbors who hadn’t responded and created a live RSVP tracker in Google Sheets. It also generated a Google Slides “hype deck” for the party, complete with details like a bounce house and neighborhood rules pulled from a file in Google Drive.

Essentially, you can hand Spark a messy planning task, let it work in the background and approve any important actions before they happen.

The Google I/O conference is meant for developers, and the Antigravity 2.0 section is a clear example of that. It’s tossing around a lot of terms like tokens, agents and subagents — but it does ultimately mean that it’s capable of generating a lot of assets simultaneously. It helps that the demo includes an example of the classic game Doom, including using Antigravity to make a custom operating system and ultimately play the game. That said, for those of us wondering how we’d use these as consumers, that’s ultimately for developers to build upon first.

Google’s Varun Mohan announced an update to its agentic development platform, Antigravity. Antigravity 2.0 is a new standalone desktop application that delivers “a truly agent-optimized experience.” In a demonstration, Mohan asked the agent to look up fun facts and stats about the 1993 video game Doom, and the agent generated various assets, including an infographic using Nano Banana Pro.

Gemini Spark is a more “personal AI agent,” and Pichai is saying that it can integrate with other tools like email and chat.

Google is introducing Gemini 3.5 Flash, which CEO Sundar Pichai said is faster than the previous Gemini 3.1 model and built for more complex agentic tasks, longer workflows and real-world developer use.

Pichai said it can generate output tokens at roughly four times the speed of other frontier models, making it useful for AI agents and coding tools where waiting on responses can slow everything down.

Sundar Pichai cued up a “behind-the-scenes look” at TPUs training for this year’s I/O. In an animated sequence, the microchips lifted weights and tried on outfits in the 24 hours before I/O.

Google’s adding the ability to ask Circle to Search if a photo was generated by AI. 

Google Docs has been infused with more Gemini with Docs Live. You can now ask Gemini (or verbally brain dump) to create a document based on a prompt and make adjustments accordingly using natural language. 

The demo showed someone talking into the microphone using a lot of different pieces of information, asking Gemini to pull in a doc from Drive and asking to reformat the doc on the fly. The document was created and edited in real time per the instructions.

Gemini Omni has been announced, meant for creating graphics and videos based on prompts. This can include asking it to create an explainer in the style of claymation.

Google says its AI infrastructure spending is ballooning. After spending $31 billion on capital expenditure in 2022, the company expects that number to hit $180 billion to $190 billion this year.

A big part of that is Google’s custom AI hardware: its Tensor processing units, or TPUs. These are not the Tensor chips inside Pixel phones, but data-center chips used to train and run AI models. Google says its newest generation splits the work in two, with one chip built for massive AI training jobs and another designed for fast responses when people actually use those models.

To show what that speed looks like, Google generated a Chrome dino-style game from a prompt using an upcoming Gemini Flash model, with the output appearing in real time at nearly 1,500 tokens per second.

Pichai is discussing the increase in “tokens” — these are small bits of written text that AI language models use to process responses to prompts. Check out our AI glossary — there will certainly be more references like this.

Google CEO Sundar Pichai is onstage, kicking off the 2026 Google I/O keynote. Pichai said AI Overviews in Search now has over 2.5 billion monthly users, and AI Mode has surpassed 1 billion users.

Google CEO Sundar Pichai arrives onstage, from our vantage point in the audience.

Google’s I/O 2026 keynote is now starting, with the livestream transitioning from jellyfish to animations and AI prompts.

Before we even get started, Google has thrown in a QR code into the livestream intro to play a game called Infinite Scaler, because apparently even the preshow needs an interactive side quest now.

Players move left and right to hit jump pads and climb through various levels, avoiding laser beams, collecting tokens and performing chain jumps to multiply their score. If you keep playing long enough, you can get a chance to add your own unique level to the tower. 

It’s a small touch, but a very Google one.

Google I/O is just moments away, and we’re expecting a ton of new announcements from the tech giant at this year’s developer conference. Expect to hear more about Android, XR Glasses and Gemini. 

The All Things Mobile Live watch party for Google I/O 2026 is now streaming, starting off with a preshow that will run right until Google begins its I/O keynote at 10 a.m. PT (1 p.m. ET). Join in and drop your questions into the YouTube chat, and participate in polls that will drive discussion throughout the show. Then stick around as after the keynote, the watch party will continue with postshow reactions from our ground team, who will react live to all of the announcements.

We’ve arrived at Google I/O 2026, collecting our media badges from the registration area. And there’s a band playing!

Google’s latest health announcements lay out a clear vision for where both the company and the industry are headed. Wearables, like the new Fitbit Air, are becoming gateways to Gemini-powered software that helps people understand years of biometric data, including medical records.

The rebranded Fitbit app, now called Google Health, includes an AI health coach that can personalize training advice using data pulled directly from a person’s health history.

Google is not alone in the race; companies like Whoop and Oura are already building their own AI health coaches. 

But first, there’s a trust problem to solve. Letting Gemini draft an email is different from letting it interpret your medical history. Health data has a long history of being exposed, shared, or sold, and even strong privacy promises have failed before.

Before this AI-driven health future becomes reality, companies like Google will need to convince their customers that their most sensitive data is actually safe.

For years, we’ve been hearing about the rise of AI agents — super assistants that would help us get things done across multiple apps and remember everything we’re forgetting. We’re getting closer, if Google is to be believed. At last week’s Android Show, the tech company showed off everything coming in Android 17, including drilling Gemini deeper into the phone so it can better handle tasks like smart scheduling, autofilling documents with data from across Google apps, and generating custom widgets.

The company has called this new iteration of assistant Gemini Intelligence, but aside from a handful of examples, Google hasn’t been terribly clear about how it’s bridging the gap between handling simple voice requests and blitzing across apps to make a hard task look easy. We’re looking forward to hearing more about how Gemini Intelligence will help shift Google toward fully capable AI agents at Google I/O 2026. 

Earlier this month, Google transformed the Fitbit app into Google Health while debuting the displayless Fitbit Air fitness tracker. The announcements made clear how Google plans to integrate Gemini into its health-tracking product, while also creating a competitor against the more expensive Whoop band.

While IO is unlikely to show off more products that would tie into Google Health, the company might use the developer conference to show more ways that Google Health is set up to integrate with other services. Google already announced that the new Health app will link with services like Apple Health, along with medical records, and the IO developer conference could provide a larger look at other ways the company plans to offer AI coaching. All the while, privacy will likely remain a high priority as these ambitions develop, as health data is by nature quite sensitive.

Read more: Google’s Biggest Health Announcements: New Fitbit Air, Goodbye Fitbit App and Hello ‘Coach’

As part of last week’s Android Show, Google unveiled Googlebooks as the next generation of laptops powered by the company’s services. While no devices were announced, Googlebooks look to be an Android-powered successor to Chromebooks that will heavily feature Gemini’s AI capabilities. 

While last week’s reveal marked the debut of Googlebooks, Google I/O might include more details about the company’s plans for this newer desktop operating system.

Read more: Googlebooks Could Be the Ultimate Laptop for Android Fans

This week is about Google’s developer conference, but Apple is making news with an announcement about its equivalent show, WWDC. The company on Monday announced its WWDC 2026 schedule, including its own keynote on June 8 at 10 a.m. PT.

And for those who like to speculate on the graphics that Apple creates for these events, the WWDC 2026 logo appears to be the one from its Swift Playground programming tool, with a metallic, reflective look. Maybe it’s meant to evoke the Liquid Glass theme?

Google I/O will kick off a week after the company showed several new Android and Gemini features at this year’s Android Show. Many of these AI-related announcements included the ability to command Gemini to book appointments on your behalf, an enhanced auto-fill that can pull your personal information from other places on your phone and the ability to create your own custom widgets that can automatically refresh with new information as needed.

While we expect Google to make more Android announcements over the course of the I/O conference, the show provides an initial jumpstart to the changes we’ll see when Android 17 launches later this year.

Read more: Android 17 Is Smarter Than Ever, Thanks to Gemini Intelligence