What launched at Google I/O 2026 (30-minute day 1 recap)

Welcome back to How I AI. I'm Clare Vale, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today was the first day of Google I/O, Google's flagship event where they launch so many products, so many features, so many names of products, some of which are live and some of which we're going to try live on today's show. In this mini-app, we'll start from most technical to most fun, talk a little bit about the releases that caught my eye today at Google I/O and see if the promise of some new consumer-grade creative features really live up to the hype of the event. Let's get to it. This episode is brought to you by Magic Patterns. Today's engineers use cursor and cloud code to ship features in hours that used to take weeks. If you're a designer or PM, you've probably felt a shift too. The pressure to move faster, validate sooner, and keep up with a team that's operating at a completely different speed. You've already tried AI prototyping tools to close that gap, but if your prototypes don't look like your actual product, it doesn't matter how fast you can build. You still end up redrawing it by hand. Magic Patterns takes your product team from idea to production and works from your real design system. When you build a prototype, what you get back actually looks like your product. You'll validate faster, get alignment sooner, and when it's time to build, engineers can connect your prototype to cursor or cloud code with the Magic Patterns MCP to pick up where you left off. Your eng team has their AI advantage. Make Magic Patterns yours. Try it today at magicpatterns.com slash HowIAAI. There was so much interesting product launch today, but I want to start with the foundation stuff, the models. Today, Google announced Gemini 3.5 family of models, including Gemini 3.5 Flash, their fastest, smartest coding model. What's unique about Gemini 3.5 Fast, it is both rivals the intelligence of some of our favorite coding models, 5.5, Opus 4.7, even 4.6, but it is four times as fast as those models. And so if you look at 3.5 Flash, you're getting, according to Google's benchmarks, a super smart model, sort of a Codex 4.7 model, if you like 4.7, at the speed of something much more like their 3.1 Flash model. So it's super, super fast and super smart. If you look at the benchmarks, they're really focused on the agentic capabilities of this model. And if you trace this all the way through the announcements today, you'll see that Google is really going full bore into agents. It feels a little bit like Catch Up. Some of the features that they released, as you'll see, are things that you're used to seeing in some of the products from Anthropic and OpenAI, but applied, I think, in two different aspects. One is with the speed of the Flash models that we know and love, and two, with much more of a creative, consumer bent to it. And so while we're going to start this episode talking a little bit about the 3.5 models and the coding products, we're going to end this episode testing this model and a couple other models on more creative use cases. But again, Gemini 3.5 available across the product portfolio with Google and really focused on coding and in particular agentic capabilities at speed, as well as the thing that we all know Gemini is really good at, which is multimodal. I have always told people, if you are working with files, videos, any sort of transformative work where you have to go from one modality, maybe document, to another modality, Gemini models are really, really good at handling files. I love it for handling videos. And you can see here, the benchmarks compared to its peers, both the previous Gemini 3 family of models as well as its peers from the Claude and GPT models themselves, really exceed the benchmarks in multimodal and, according to them, some of their agentic capabilities. And then it's fast. We love something fast. Now, how are you going to use this model? Well, if you're a developer, you're going to use it in an IDE or in an agentic coding harness. And so, you know, you don't hear people talk a lot about anti-gravity, but when I do hear people talk about anti-gravity, they say it's quite good. Anti-gravity is Google's IDE, agentic IDE, and they announced several features. So when you read about these features, you're really going to feel like Google's playing catch up to, in particular, Codex. You can see a lot of the concepts that were built into anti-gravity are concepts that we've seen in particular in Codex. But let's go through them one by one. First, I want to pop up anti-gravity so you all can see it here. A couple changes. This is called anti-gravity 2.0. A couple changes they made in the desktop app. One is they've brought the idea of projects into anti-gravity. So these are sort of like folder constrained environments or workspaces that you're working on. I pulled in the chat PRD website app here. And then they've also added in scheduled tasks. So this UI looks very similar to what we've seen in the Codex app projects along the side, scheduled tasks along the side. And scheduled tasks are exactly what you would think they would be. A name, a project that you're working on, a schedule, and a prompt that runs on a regular cron. So nothing mind blowing here, but again, you're going to see that Gemini 3.5 flash, high reasoning and low reasoning, both very fast for a limited time, are available here in the anti-gravity IDE. The other thing that was announced was the anti-gravity CLI. So again, a cloud code or codex CLI style interface for coding. This is one where you can open up your terminal and work with it outside the IDE. Very similar form factor to what we've been working with. So again, you're going to kind of feel like this is a little bit of catch up to what the other providers have been doing in terms of agentic coding, but it looks nice and we're defaulted again to Gemini 3.5 flash high. And so it should be a fast experience. Let's just test this really quickly. This is on my website and I have a feature for the blog that I wanted to work on. And that is our blog generator for our podcast. So as I said at the beginning, the Gemini models are very good at video. And so one of the things that we do is we put the videos from these podcasts into an admin tool on the marketing site and it generates blog posts for us. But I want to be able to do that agentically. So let's see how fast Gemini 3.5 responds to that request. And I'm just going to use, I'm going to type in and I'm going to say, we have a blog generator UI at admin tools. I want to turn this into an API that an AI agent can use instead of a web UI for our team. Please build. So that should go ahead, just very similar to these other tools that we're used to. Look at directories, ask me for permissions. I'm going to say yes and always allow. And it's going to run through and hopefully write some quick code. Now I'm not noticing it is particularly faster, but let's see how long it takes to get to an outcome. You know, it doesn't feel that much faster to me, but we'll see how long it takes to generate. I want to show you a couple other features of anti-gravity that was released today. So again, going on this theme of playing a little bit of catch up with cloud code and codex, the core agent features that were released today in anti-gravity are sub-agents. Again, the ability for the main agent to spawn off a sub-agent to do a specific task. One use case that people love of anti-gravity, especially coming from Google is the browser sub-agent, which has always been available, but now different sub-agents could be spawned by the main agent to work on coding tasks. This is something that you've seen in cloud code and codex again, but now available in anti-gravity. There's also hooks. So you can use hooks at different parts of the life cycle of your agent. If you don't know what hooks are, there are little events emitted every time your agentic harness kicks off a tool or every time it completes a turn or every time a new session is started and you can hook into those events and do something on demand. And so now anti-gravity has the ability to configure those hooks. We have the idea of projects, which I told you about and showed you in the desktop app. There are native Git work trees and local local development environments. Again, these are all things that we have seen in codex. So nothing super surprising here. The ones that I really liked that I wanted to just spend a couple minutes on while we're letting anti-gravity in the IDE cook are these slash commands. Now I love some slash commands, especially in codex. My favorite one has been slash goal, the ability to define a goal and basically have your agent, you know, bash his head against that problem over and over until it solves, solves the problem or meets the goal. So anti-gravity has shipped a slash goal slash command, which will allow an agent to do a long running task against a goal. But there are actually a couple other really fun slash commands that I want to call out that I think I'm going to be testing over the next couple of weeks and probably give you all my feedback on. So this first one is this grill me slash command. I love this because, you know, cloud code has this question and answer tool where it'll like clarify for you. Buy ThoughtSpot. Product leaders know the struggle. Your users want data insights, but they don't want to leave your app to find them. ThoughtSpot Embedded solves this by putting analytics directly into your product. Your users can search in plain English and explore data instantly, right where they work. No separate tools and zero contact switching. What sets ThoughtSpot apart is that it's not just another bolt-on dashboard. It's a search-driven, AI-powered experience that feels native to your app. Developers can embed it with just a few lines of code and then fully customize the look and feel. The result? More engaged users, faster decisions, and a product that delivers more value every time someone logs in. If analytics is becoming core to your product strategy, visit go.thoughtSpot.com slash HowIAI for more information and try the free trial at go.thoughtSpot.com slash HowIAI/trial. Now, one of the things that you'll notice when you go into the new Gemini is there is a redesign. They made a big deal about this redesign. They called it something ridiculous that I can't remember now, but we'll put it in the show notes. It's like supposed to make you feel good. Look at this glow. Everything's redesigned. You have a bunch of prompts and examples. So they really are trying to up level and upscale the consumer experience of using Gemini. But even more than that, they have added some pretty cool features to Gemini. So if you're not familiar with it, Nano Banana is one of the best image gen models. You can now create images with Nano Banana using a bunch of templates. So they're really trying to make it easy to inspire you with what to do with these models. I use Nano Banana quite a bit actually for our podcast thumbnails. So let's try it really quickly. I'm actually going to grab a screenshot of myself so I look cute. Okay. I drag this in and I say upscale and beautify because I want to be pretty, the image of this podcast host. Change the background to a professional podcasting studio. Okay. So I'm going to press enter. Sorry to the people in the comments that keep saying, stop typing on your laptop, type on your keyboard. It's making the video shake. I know I'm just used to doing it. Okay. It took a little bit, but here it is. This is not my face. This is horrifying in every way possible. But some of the things that I notice about the image gen that has changed is, again, all these image gen models want to get a lot better at generating text. You see that it looks a little bit photorealistic, even though it doesn't look face realistic. And it generated pretty fast. But more fun than the image gen models are the video generation models. And so Google announced a new video gen model called Omni. Omni has a bunch of capabilities that we're going to show in a minute in the flow app, but it's able to create longer, more consistent, lower photorealistic videos using AI and it's able to use reference materials to create those videos as well. Okay. I have this image that my kid drew me of this little guy. I am actually going to take a screenshot of it while I hold it up to this camera. So just bear with me while I do that. I'm going to drag this over into Gemini video creation. And we're going to use this new model to animate this superhero. Animate this superhero breaking a kid out of class to go have fun. Now Google has made all of these experiences more agentic. So you saw it generate a plan and then it's going to create a video. It's going to take a couple minutes. So while we're letting that generate, let's just talk a little bit more about how Google is describing Gemini Omni. They're comparing it to nano banana for video is now Omni for video. And it's going to combine a reasoning model with a video creation model. And what they say is you can create video from anything. So, which is why I picked the example of taking this very cute drawing that I keep from my kid over here on my desk and seeing if we could create something from this again, going back to what I talked about, the Google models, the multimodal capabilities of these models is very high. So when you want to do conversion of video to text, image to video, all those sorts of things, you can ground what they say is you can ground Gemini in real world knowledge and create video components based on that. There's a couple of really cool features. So one of the things that they let you do is conversationally edit videos. So let's say you have a video of a structure. You can change that video of that structure to change the structure into bubbles. And so you can take real life video components and change them, edit them sort of Photoshop style with Omni by just prompting it. I think that's very cool. You can take a video and have Omni describe it. So again, this like multimodal part of it. And so you can have it describe it and then you can edit the description to generate a new version of that. I think that's pretty cool. You can refine the same video. So again, one of the challenges with these video generation models is that when you create them, they take very long and then you can't really edit them and then you get inconsistent characters. And so the ability to change the environment, angle, et cetera, but keep characters consistent is going to be really powerful when doing sort of production level video gen. Let's see. My video is ready. Let's get out of here and go have some real fun. Hold on tight. Here we go. This is really going to like this. Again, this was 10 seconds long. And so it really is much longer than like, I think the six or seven seconds that Sora was. So again, we're like extending this bit by bit. And then the ability for me to like conversationally edit this video or change the school to like an academy and have it be snowy outside, all those sorts of things are here with the Omni model. But you can see that it's going to be a really powerful video editing model. Now, if you want to go deeper into video editing, you can use Google Flow, which is a much more prescriptive step-by-step video editing tool. They embedded Omni into this new tool Flow. And one of the things that you'll see here is they're really doubling down on cinematic quality production quality editing. And so they're looking at cinematic realism. Is this hyper realistic? You can blend multimodal references. So if you have a person, which I'm going to get to, a situation and environment, you can use that to seed the video. And then you can kind of edit videos using conversational information. One of the things that you'll see in Flow that they've encoded is the ability for you to define characters. So let's say I want to take this guy and call him Captain Escape a School. I can use this design a character in Google Flow and then at mention that character in any future video moving forward and create videos with that. And then you can actually create yourself as an avatar. So in addition to being able to create kind of like a unique new characters, you can actually take a video of yourself and create an avatar. Maybe we'll try that. There's a lot of custom tools built in, brainstorming, how to scale, how to organize. So you see this really coming for production grade AI video gen. But let's jump into Flow really quickly and see if we can make an avatar of myself and how it works. How you do this, you go to flow.google, it will redirect you to Google Labs. You click your name, you create an avatar. We're going to get started. Okay, I scan this QR code. I pull it up on my phone. It's selfie.app.google. I agree to it taking my face. I'm going to allow camera access. I'm going to move this mic out of the way. Give me a sec. BRB. It's telling me to read numbers out loud. 25, 47, 56, 87, 18, 52. Okay, it's telling me to turn my head right. And then I've turned to, okay. And then it's telling me to turn my face to the left. Okay, that was it. Please don't steal my identity. All right, so it's uploading and it is creating an AI avatar of at me. And then I should be able to use that for any video moving forward. No, no, sir, I couldn't. So again, here's where we're really on the struggle bus with Google, which is they've announced a lot of stuff and it hasn't really worked. So I did it. I gave them my identity. I trained their model. I subjected my human face to their DeepMind engineers and it didn't even create the avatar. So the promise is really good for some of these things, but the reality is if you're not able to use them or they're broken on the day, then people are going to lose patience for some of this. And so, you know, a couple of my challenges that I've had with the Google announcements is one, I haven't been able to really find where they are because the products are named hard. And two, even when they announce a feature, even if it's live, even if I can get to it, it hasn't quite worked yet. So that's a real bummer. You know, the last ones I'll show you, I was hoping to do videos, but we'll close it off with something a little bit more accessible, which is there are two tools that I think are really interesting for designers and marketing. There is Pomelly, which is their brand product. And then there is Stitch, which is their design product. Again, we've got the anti-gravitys. We've got the AI studios. We've got Google AI slash Gemini. We've got Flow. We've got Omni. We've got Stitch. We've got Pommel. So like Pommely, Pommely. We've got a Okay, and not to wrap on a total dud, but maybe it's the summary of the evening. I had time to do a diaper change, come down and check out my Pamily generated website, and it's fine. So I think a lot of interesting releases today from Google, a bunch of stuff to go play with. I think the video piece is the part that I'm most excited about, probably second is just the speed of coding that comes from these flash models. But there are some sharp edges, some things that definitely need to be improved, and some things that actually need to be released. I cannot wait to see what you build. I can't wait to hear your feedback and what you're most excited about from Google I/O this year. Thanks for joining How AI AI. Thanks so much for watching. If you enjoyed the show, please like and subscribe here on YouTube, or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at howaipod.com. See you next time.