Loading summary
Transcript56 lines
- [00:00]
Host
We are here.
- [00:01]
Logan Kilpatrick
It's Google I O. Logan kilpatrick from the DeepMind team. Friend of the pod. Been on the pod a few times. Logan, by the end of this episode, what are people going to learn?
- [00:11]
Host
You're going to hear about all of the new releases from Google that just happened at Google I O. And also like specifically we should deep dive on like how to build new AI agent native products, which was sort of the thread of Google I O this year was Agents, agents, agents. So we should talk in depth about everything that we launch and what it means for builders and developers.
- [00:39]
Logan Kilpatrick
Okay, cool. So I mean let's start off by okay, what was launch and why does it matter?
- [00:45]
Host
Yeah, there's so much, so much new stuff and I want to get your reactions to this too because I think you have a grounded perspective of sort of the technology. I think one of the highlights was Gemini 3.5 flash. It's the, the best model we've ever shipped and we've made available. Really sort of. If you look at the history of Flash, I think Flash started as sort of this like smaller workhorse model that was really great for chat and was sort of the, you know, very cheap to use, very cheap to run. And I think we've sort of. Flash continues to evolve to meet the era of what people are actually trying to use the models for. And I think the era that we're in right now is people are trying to use the models to do agentic sort of long running tasks. And I think we want Flash to sort of be the workhorse model for the agent era, for agentic long running tasks, for coding, for all that stuff. So you see a model, a Flash model that's like actually really great at coding, that's sort of competing with a bunch of our sort of ecosystem competitors with very large models. The Flash model is sort of pulling its weight.
- [01:50]
Logan Kilpatrick
How should people think about, you know, flash 3.5 versus the competition?
- [01:54]
Host
Yeah, I think it's, it's probably like a more like sonnet level model. I think OpenAI with just mainline GPT and then GPT mini is sort of, they don't. I feel like Sonnet is not like a small model, if you will. It's obviously it packs a punch and I think it's definitely smarter than the Mini models. So I feel like it's, I think we're anchoring more on the Sonnet level intelligence. So if folks are using that model and want to try Flash, please let us know and send us the feedback. Feedback of how it stacks up. Yeah, and the reasoning, all the agentic tool use stuff is all incredible. So I think that was one of the launches that models available actually for the first time, like available to all the users in search, available across 900 million users in the Gemini app, available to developers in the API so many other places. So I think it's like the most widely distributed model launch on day one that we've ever done, which has its own set of challenges that we could talk in depth about. The other thread which I think folks are very excited about is Gemini Omni. So sort of this new model that we've created, actually sort of somewhat of a world model I think is how Demis framed it when we announced it on stage yesterday. Being able to take in any type of input and create any type of output. And I think to give context for folks like Google, you know, we had Veo and Veo was state of the art and sort of pushed the frontier for video generation. We had nanobanana which could do sort of image generation and editing. We have all these audio models that do TTs. We have a new lyria music model that's actually really, really capable of. And the idea is how do you fuse all of those models into a single thing so that A developers lives are easier B we don't need to train nine different models and actually get this really interesting cross pollination of capabilities so that the same model that can actually benefit from Gemini's world understanding and the ability to generate text can also make a video, can also edit a video and you see these really interesting things. And we saw this with nanobanana of like what happens when you give world knowledge to an image generation and editing model is really interesting use cases and we get feedback all the time about what that's unlocked for customers. And so I'm really excited to see Omni starting out in the gemini app and YouTube and in Flow. And then very soon already we're kicking off a bunch of early access tests. Hopefully as soon as I get out of I.O. and get back to the office and get feedback from developers and sort of continue the iteration and bring it to developers in the API so that folks can actually build products on top of Omni.
- [04:17]
Logan Kilpatrick
Yeah, I mean it's products building products, but also like building content that gets seen. One of the hardest parts about you know, vibe coding or building any business is getting distribution to the thing.
- [04:29]
Host
Yep.
- [04:29]
Logan Kilpatrick
Right. So what when I was watching the Omni demo and just seeing like how amazing it actually is, like what was going through my mind is like, okay, I. How can I make a commercial? How can I make an ad? How could I create an Instagram account using this model? And I think what you're going to see is a lot of people generate millions of followers, generate millions of views every single month if they know how to storytell well and use the model.
- [05:01]
Host
I actually also think. And what I would love to see, I don't know, we'll give some work to your editor team, make the intro of this video have a bunch of, like, crazy stuff happening and you can actually, like, change the intro of the video. We did this for some of the podcasts that I was doing actually with the Omni team, and that. That hopefully it'll create a bunch of new creators. I also think it is a fundamental accelerator for existing folks who are producing content. Like, editing video is hard. Like, I'm sure you have an amazing team that's doing really hard work and, like, there's not enough hours in the day, there's not enough editor time and hours and storage on SSDs in order to do all the stuff that I think could be done. I think Omni will sort of fill this really interesting gap for maybe the creator who wants to tell an interesting story that didn't have the means to go have a whole team sort of support them in telling that story. And I'm really excited to see that happen. And I've already been seeing a bunch of examples on X and other places of folks starting to be able to tell that story and bring content to life in new ways and actually, like, repurpose existing content. So it's going to be crazy. And this is only. This is the, like the. This is the first iteration of the Omni model. This is the Flash variant. So, like, it only gets better from here, which is really exciting. So I think we're going to see some really cool stuff throughout the rest of this year.
- [06:19]
Logan Kilpatrick
And from an API perspective, like, how do you see. You think people are going to, like, integrate with Omni in terms of the products they build?
- [06:25]
Host
Yeah, yeah. I think there's so many, like, interesting creative suites. I think a lot of people will also just be doing what you described, which is like, doing content with it and using the Gemini app or Flow or YouTube in order to do so. But I really do think it's an example of, like, this video remixing editing capability is, like, not something people have built into products thus far. So it is going to open up a new category. And I think if you're building businesses and, like, want to help, you know, actually if you want to build a business and help the next thousand, a hundred thousand million creators go and tell their story, I think there's a bunch of really unique opportunities because this is a fundamentally different way of, of interacting with video.
- [07:02]
Logan Kilpatrick
Totally. Yeah. I mean, it reminds me like I remember when social media was coming up and social media agencies were just popping up. Right?
- [07:08]
Host
Yeah.
- [07:08]
Logan Kilpatrick
There's sort of a similar opportunity right now where, you know, you can build an Omni agency and sort of deploy this for small businesses as an opportunity. Right?
- [07:19]
Host
Yeah. And there's so many ways to like, make the. My takeaway from seeing this, like on my own content was it makes the content more engaging. Like you could like, do all types of like, goofy things that like, don't actually dilute. You know, people are like, the tongue in cheek thing is like the subway surfer, you know, engagement farming and videos because people don't have attention. I think there's actually like, I'm, as a, as a consumer of content, I'm actually really hopeful that there'll be more people doing like the much more thoughtful version of that in a way that like, you could actually only do with Omni or like a really, really advanced, like VFX studio that was able to pull this stuff off. So I think we'll actually see that and I think it will like, it's like move us away from this like, slop sort of era into something that's like a little bit more tasteful. And I think that's maybe corollary to I think some of the narrative of like AI tools sort of just generating more bad stuff in the world. I think Omni actually might be a really cool tool to help people tell a story with a higher degree of taste than they've been able to.
- [08:22]
Logan Kilpatrick
What wasn't launched today that you wish was launched today? I know it's a question that you probably don't like getting, but one thing I like about you is you're not afraid of criticisms. And I see you. When I ever see criticisms of any of the products, you're the first person in the replies to be like, we're working on it, we're taking feedback. So anything that you wish to see over the next coming months?
- [08:53]
Host
Yeah. Well, first of all, I do love the feedback. I think actually, I think folks look at sort of criticism in such an interesting way. I see it as like, it actually makes my job easy. Like if you come say, you know, hey, we wish the models were able to do this thing, or we wish the API was able to do this thing. Very easy to like, go and take that feedback. The hard thing is actually trying to identify these things when people don't tell you what they're, what they actually want. So I love the feedback. I appreciate it. Please, please keep it coming. I think what I would love to see and what the team's actively working on is the other set of Gemini models. So we, Obviously we're launching 3.5 Flash. 3.5 Pro is in the works. It's cooking. There's many iterations and runs happening behind the scenes right now. I think we said, we announced yesterday it'll be available early next month or hopefully sometime next month, maybe not early next month. And so, yeah, I'm excited. It would have been great to sort of bring the whole 3.5 model family out at the same time. But it is also fun because I think 3.5 flash sets the bar high. And I think we, we continue to. I describe it as like pulling the rabbit out of the hat. We continue to take the pro level intelligence and then some and stick it in the Flash model. And I was talking to Oriole and Jeff yesterday, who are some of the leads for Gemini, who actually invented distillation, which is the technique that allows us to take the pro level intelligence model and stick it into the Flash model sort of every single time. And it's just crazy that it, that it keeps working and it brings the cost of intelligence down, which is really interesting. Even though lots of feedback about the cost of Flash so far, the important thing actually is that the cost of intelligence goes down. The model is one implementation of the skew, capturing the cost of intelligence. And I think, folks, it's a lossy way of expressing the cost of intelligence. But I think net, net cost of intelligence has gone down with Flash, which is really exciting.
- [10:54]
Logan Kilpatrick
What else was launched that a founder, someone who's trying to be more productive or make money on the Internet should know about?
- [11:01]
Host
Yeah, I think 20. I mean, obviously the open claw revolution that sort of like took the world by storm is exciting and is an opportunity. And actually, you know, I love the folks who are working on OpenClaw. It's an incredible open source project. Peter's awesome, but actually, for folks, if you've tried that product experience before, it's really tough. Like, you really do need, like, you need to be a confident person who's like, willing to take risk and let things break. I think Gary Tan sort of described this as like, you know, it's a Ferrari, but you have to also Then be your own personal Ferrari mechanic and sort of make the tools work.
- [11:34]
Logan Kilpatrick
And so, I mean, I studied computer science at school and like, same, you know, and I didn't feel great, you know. Yeah. Love what the team has done and stuff like that, but I was kind of nervous. My heart rate was going up when I was installing and doing things.
- [11:49]
Host
Yeah, yeah. And you have to take the leap of faith. And so I think for people, building stuff like that is in itself an opportunity. And so I think there's sort of two angles to this one. Inside the Gemini app, we sort of launched our sort of flavor of this, like always on 24. 7 agent to sort of help you run your business or come up with your next idea. And sort of it's rolling out to trusted testers this week and then next week it'll start rolling out to Gemini Ultra customers, which is exciting. So I think sending work, throwing work over the fence and having an agent do it is the best thing. And if you want to build experiences like that. We just launched managed agents in the Gemini API. So using the same harness that's actually powering the same model that's powering Gemini Spark in the Gemini app, we also have that ability for developers to go and build those experiences themselves. And I'm really excited about this. Like the friction to build agents historically and choosing a framework and all of this sort of like iteration loop that you have to do in order to like get quality good enough, etcetera, etcetera. Trying to shortcut that for people who want to build and just like not have to deal with the infrastructure, not have to deal with any of the problems. Send a single API call. I demoed this on stage yesterday doing like an AI radio show and I think we're calling like seven different models and doing all of this. I didn't write any orchestration code. I literally just like wrote skills and markdown and was able to have it sort of go and orchestrate this whole show. So I think it also, like, even if you're not the most technical developer and you didn't study computer science and you're just using these tools off the shelf, like managed agents in the Gemini API, I actually think hopefully will lower the barrier to entry for people who
- [13:32]
Logan Kilpatrick
want to build agents, I think MCPS are coming soon to that product.
- [13:37]
Host
I think we will have MCP support for it to do tool calling. I think right now it is just like skills in order to sort so you can describe and it can do some of these things. But yeah, there's a whole this IO is sort of like step one of the managed agent story and there's like 50 other things that are on the roadmap that we need to land in order to make the experience rich and feel featureful and all that's coming.
- [14:01]
Logan Kilpatrick
I mean that was my one big takeaway from being IO is, you know, I always thought of this time as the AI era. And you guys have been saying it's the agentic era.
- [14:10]
Host
Yes.
- [14:10]
Logan Kilpatrick
And a lot of the new products that I've been seeing that you guys have launched over the last 24 hours has been agentic this, agentic that. Can you just. Yeah. I'm curious your thoughts on this agentic era. And you know the person who's listening to this is an idea person. Right. So they're listening to this and I'm like, okay, how can I use this in my business or create a new business? Do you have any requests for startups or any ideas given that we're now in the agentic era, that we can be using Google products? Go build.
- [14:42]
Host
Yeah, no, it's a great question. I think two things come to mind. I think historically there was this like one to one correlation of you spending your time and actually an actual work happening. And I think the exciting thing for somebody who has ideas and wants to build stuff is like asynchronous agents, agents running in the background fundamentally changes. That dichotomy of like you don't. Doesn't require you actively in the driver's seat every moment that there's actually useful work happening. And the important thing is like useful work. I think we're all sort of um. I was having a separate conversation about sort of every three to six months this like expectation reset that you have to do with somebody building in the AI era. I think a lot of people like tried. I'm trying to think of Auto GPT is one example like three years ago. And that team I think at the time like did something really interesting but like it didn't really work to do anything useful. It was like, it was a really, really cool and interesting demo. And so I think a lot of us, like, and myself included, I was like, okay, we're you know, agents are exciting. It's the future. The future was not there at that time and it feels like we've really crossed the chasm and that future is right now. So if you sort of written off that there's a bunch of opportunities, reset your priors, There are opportunities and folks should be actually building those products because the customer base this Is the interesting thing is that customer base does not yet actually know that they need an agentic product. And so I think there's a, there's some alpha in like how you storytell this. Like, it's not clear that like the average person, some of them maybe are looking for an agent. I think there's a lot of people who are like, they have a problem and they just want that problem to be solved. And you can solve it in a way that like really brings time back into people's day instead of like requiring them to be actively in the driver's seat the whole time. So, and I'll add one more comment which is doing that in the form factors that people are already familiar with. Still feels like there's so much alpha in this. Like trying to convince the small business owner to go and adopt some completely new thing that they've never heard of before. That's really good. They're gonna have to teach all their employees about and their family about. It's a high bar to pull that off. Teaching the small business owner how to text with an AI assistant or how to send an email. You don't have to do that. They already know how to do that. And so I think you can really use this like existing technology to meet people where they are at the same time that you like fundamentally introduce new technology to them.
- [17:18]
Logan Kilpatrick
Anti Gravity got a huge overhaul.
- [17:20]
Host
Yes.
- [17:20]
Logan Kilpatrick
We should talk about that.
- [17:21]
Host
Yeah, yeah. So I think the whole Anti Gravity suite actually is coming together. I think we introduced Anti Gravity, I think like six months ago, something like that, sort of as an AI powered ide. And I think if you look at where it is now, it's actually an entire ecosystem. So you have Anti Gravity sort of the Agent manager. If you don't even want to touch the code itself personally and you want the agent to do all that for you. There's the Anti Gravity Agent manager on web and desktop that you can install. There's the ide, the same antigravity product that you were using before continues to work as an ide, you can do that. There's Anti Gravity, the CLI product. So if you're a developer and you're like, I love the cli, it's the best thing in the world, they've got that for you. If you want the SDK so that you can actually build agents on your own infrastructure structure, the SDK exists. If you want Anti Gravity sort of powering experiences in the API because you don't want to manage the infrastructure, we have that in the Gemini API. And so I think the story is like antigravity sort of as this agentic coding layer meeting you wherever you are, however you want to build products. Actually integravity, sort of the agent harness powering the always on Gemini Spark experience for consumers in the Gemini app. So everywhere you go, actually even going to search. So everywhere you go you sort of access Anti Gravity. Now it's another layer sort of bringing the Google ecosystem together just like Gemini is as well, which is really exciting.
- [18:52]
Logan Kilpatrick
How should people think about Google AI Studio and antigravity?
- [18:55]
Host
Yeah, you're making my job easy.
- [19:00]
Logan Kilpatrick
I mean that's a question I have. Like, honestly, you know, it's like I can tell you how I see it.
- [19:04]
Host
Please.
- [19:05]
Logan Kilpatrick
You know, to me it's like AI Studio just feels way more comfortable if I'm non technical. But an Anti Gravity just feels like more of a fully featured developer focused product. But that's also changing a little bit. So yeah, I'm curious how you see it.
- [19:27]
Host
Yeah, I think it's changing in both directions, which is interesting. I think AI Studio does a lot of different things. We have sort of the playground experience. If you want to test our latest models and agents, you can get your API key to build with the Gemini API if you're a human or an agent, which is exciting. And we're now also, we started building this Vibe coding experience probably I think actually 12 months ago, which is crazy and to see the progress of where we are today. You can now natively build Android apps and sort of share them with users and download them onto your phone and you can natively integrate with Google Workspace without ever leaving AI Studio, which is also really exciting the way that I frame this. And actually Andrej Karpathy has the best framing of this, which is it's actually two different things. There's on one end of the spectrum is Vibe coding, on the other end of the spectrum is agentic engineering. AI Studio is going after Vibe coding. We want to make it so that you can bring your idea to life, go from prompt to profitable company without ever actually having to see any code. You could never look at a single line of code and we should help you do that. You should be able to deploy, get a database in the future, add payments, get a mobile app, do all those things. Never look at a single line of code on the Anti Gravity side. I think the problem they're trying to solve is like very much like production quality code. You want to work in a million line code base, you want to build Google. We're using Anti Gravity to build Google. You want to build Google. The bar is quite high. We are using AI Studio in certain ways to build Google, but not contributing to the Google search code base as an example. And I think there's a flexibility control trade off versus batteries included. I think Anti Gravity can do anything. You can use other models. You could be using the OpenAI API if you want, as part of Anti Gravity. AI Studio is really focused on the Google ecosystem and so we're trying to make a bunch of opinionated decisions for you, make it so that you don't have to think about those things actually. And it just all works. You can sign in with your Google account, it works out of the box. You can build an Android app natively for free, install it on your phone,
- [21:32]
Logan Kilpatrick
which I just found out yesterday by the way, that you can do that.
- [21:34]
Host
It's crazy. I mean, we just launched it yesterday, so it's a brand new feature that I'm excited about. And Paige and I were talking off stage before yesterday with a bunch of folks and like neither of us had ever actually built an Android app before. And it's like opened up this whole new ecosystem that we probably would have. I mean, I love Android and Android's awesome. Like, I was probably never going to build an Android app. And so to see, literally in the first day, like tens of thousands of people go and build their first Android app in AI Studio, literally for free. I think it's really cool.
- [22:06]
Logan Kilpatrick
I mean, so much alpha and just that alone. Right. If you think about it, you know, Android is the number one operating system.
- [22:13]
Host
There's like billions of people using it. Billions of people.
- [22:15]
Logan Kilpatrick
It's crazy. And you don't need to be technical and you can ship your Android app and you need a good idea, you need a good niche. You have to understand distribution. But there's no reason that if you want to build a mobile experience that you're not just trying it out.
- [22:33]
Host
Yeah. And one of the intentional decisions on Android that we actually made was to make it so that these are native Android apps. So this isn't like you're not using one of the frameworks that sort of lets you do like all of the different ecosystems with a single app. It's intentionally a native Android app. And the reason for that is because we want to make sure you can actually build for all the different form factors. So you can actually. And this doesn't work perfectly well today, but we have this sort of groundwork in place and this is where we're incrementally getting towards, but you can build for our Android XR wearables. You can build so like glasses in the future, in the fall when the glasses land, you're going to be able to build in AI studio, build a native app for XR and build XR apps without ever actually writing any code. You're going to be able to do the same thing for the other wearables like the watches, Android auto. You can build apps for people in the car when they're on the go. All of those things using the same sort of setup in the same agent in AI Studio is really exciting. And those form factors you actually don't get using any of those other components compatibility frameworks. And so I think it'll be cool to sort of build these frontier type applications in AI Studio with native Android apps.
- [23:45]
Logan Kilpatrick
For people who listen to this, we'll end with this. For people who listen to this, what are your hopes with how you know there's hundreds of thousands of developers and builders who are listening to this. What are your hopes for these people who are watching now to go and build?
- [24:00]
Host
Yeah, that's a great question. I think the, the most exciting thing is like what, what YouTube did for creators I think is what's happening now for software. I think there's going to be this entire generation of like software creators where you can just like go by yourself or with a small team like go and build a business using all of these AI coding tools. And I'm very excited to see, similar to what YouTube has done where like you don't need this like historically to build software, you needed like either get really lucky or raise a bunch of money from people. And in order to raise a bunch of money from people, the market had to be massive. And so there was all these really, really interesting problems that people have just decided they're not going to solve because the market's not big enough. And I think those markets are now in play. And I think it's just like it's alpha for the observant person who's willing to do the diligence to go try to find some of those opportunities. And I think why I love the content that you do is because I think you're actually helping people find those opportunities. And like that is the next generation of like value creation that's going to happen is people going after these things that you wouldn't have obviously gone after five years ago when the cost to build software was so high and you needed a 40 person team to build something. Now it's like one person or a small group of friends and some of these tools and anything is possible.
- [25:23]
Logan Kilpatrick
You're off to the races.
- [25:24]
Host
I love it.
- [25:25]
Logan Kilpatrick
All right, thanks a lot, Logan.
- [25:26]
Host
This was great. Thank you for.
- [25:28]
Logan Kilpatrick
Thanks for having me and first IO
- [25:30]
Host
first IO hopefully we'll have you for next year as well. And thanks for blazing the heat of Northern California.
- [25:40]
Logan Kilpatrick
I appreciate you.
- [25:41]
Host
Yeah.
- [25:41]
Logan Kilpatrick
Take care of.