Google AI Makes Building AI Apps Too Easy⏐Ep. #234

I want the product experience when you land on my website to be an AI voice agent. All right, it looks like it's done. 66 seconds. Let's try it. Hey, how's it going?

It's going well, thanks for asking. Could I get your full name, email address, and company name?

Okay. My mind is buzzing with the possibilities. How big is the opportunity to have, like, an agency that helps implement these tools into small businesses?

I was at a an event in Santa Barbara last week. The top executives in the world from Fortune 200 companies and even these top executives don't know how to AI tools. You're 10 prompts away from the home run.

Could you imagine how arrogant that Advice would sound three years ago, just launch 50 different things. Now it's like, no, that's 50 prompts. You can do that.

You're literally going from, like, I have that idea to deploying the thing into literally 90 seconds. And, like, the cost and barrier to entry is literally free.

All right, guys, I just recorded a banger of an episode and I feel the need to issue a public apology. What apology? Well, in the past, I've talked about a little bit of crap about Gemini and or Google's AI products. I'm a changed man. Today I chat with Logan Kilpatrick. He is from Google's DeepMind team. He knows a ton about all of Google's AI products and about what ideas you should vibe code, what AI tools you should use to vibe code, what types of ideas. He shared a screen and we vibe coded two different apps with one prompt each. This is not a sponsored episode. Google paid me $0 to publish this. But I am a shareholder of Google and a personal fan of their company. And I think there's a lot that we can learn and a lot that we can build on top of Google's A products. So Today we covered VO3 Nano, Banana, Gemini, AI search, and more. So if you are here for ideas and tactics and tools, this is the episode for you. Please share with a friend and enjoy.

My name is Logan Kilpatrick. I lead the developer product team for AI Studio and the Gemini API at Google DeepMind. And we're really focused on how do we take the latest AI capabilities coming from Google and coming out of DeepMind and put those in the hands of people who want to build stuff and the whole sort of spectrum of builders these days from everyone from, you know, developers at enterprises who are like, building massive companies and products to like, hey, I just have this idea and I want to build my first AI App or build my first app at all and you know, share with a friend or someone in my family. So we're building for all of those people, which is challenging but also a lot of fun.

That's exciting. What of like all the products within Google's AI Studio? What are like the main ones? Why don't you just list them off?

The fun part about Google is we have a massive spectrum of products. You get everything from if folks have heard of the Gemini app. So the Gemini app is like Google's personal assistant powered by Gemini to, you know, all of the Gemini integrations that you might find in like other Google product services. So that's everything like Google Docs and Sheets and Drive, et cetera. All have Gemini integrations to search with AI overviews and AI mode folks haven't tried is like an awesome experience inside of search to as you mentioned, like actually custom built, like almost like AI native products. And that's NotebookLM, which is sort of like a research assistant that takes all this like content that you have and makes it infinitely repurposable to AI Studio, which is our sort of AI builder platform. If you want to create software and you want to build something powered by AI and put it in the hands of users, is the platform for you to Google Labs, which has all these like early stage bets and things like Opal and others. So there's, there's a huge spectrum. The simple answer to your question, Chris, is you can go to Google AI and it has sort of like the suite of all of these different products that are available and you can sort of look through and find what maybe is most exciting to you.

Okay, so you've got a really unique insight into all of Google's AI products. If you were forced, heaven forbid, to have to pick one of them to build a business with or to start a business with only one of them, what would you choose and why?

Yeah, it's a good question. I'm biased because I spend my time thinking about AI Studio. So probably AI Studio. If you want to build a product and build a business, I think my pitch to you would be build AI products. And the reason for this is like AI is both extremely monetizable. There's edges everywhere you look like there's all of these emerging interesting product experiences that you could potentially build. And actually the barrier to entry is pretty, pretty minimal. The beauty is you don't need to be an AI expert to go and like build AI products. You just need to have some problem that you're trying to solve. And like there's a probably a good chance if it's at least a digital problem that AI can help solve that for you. It is this like unique combination of all these things. Plus like consumer interest is increasing, consumer willingness to pay for AI products is increasing, costs are going down, the technology is improving. So there's like all of these very unique tailwinds that are happening at the same time which I think makes this like such a exciting moment to build AI products in AI Studio.

You've probably seen a lot of people building cool stuff with Google products. What are like some interesting stories or case studies of people that have used your products to build something awesome or something simple or something profitable or any and all of the above.

I'll give sort of two ends of the spectrum from, from an answer perspective. So on one hand, like in AI Studio today, teams across Google are like prototyping the next versions of A Flow, which is our video editing creation product for creatives. Like experiences around Google search and YouTube and other things like that. So it is really like becoming sort of the engine room for Google to prototype and like build next gen AI experiences. So I see like one end of like, you know, these experiences might end up impacting billions of users too with the Nano Banana launch. A few weeks ago on day one of the Nano Banana launch, users inside of AI Studio were vibe coding literally tens of thousands of AI powered apps using that new model and like much thinner experiences. Things that are like kind of goofy but also like interesting like haircut AI and like all these like photo restoration products. These like very, very vertical use cases. But like you're literally going from like ah, I have that idea. I could see why people would want that to deploying the thing into production and like putting in front of users in literally 90 seconds and like the cost and barrier to entry is, is literally free. It's paid when you deploy and you have to like, you know, service external users and stuff. But to actually build the thing is, doesn't cost anything. Which is just such a unique moment that I don't think like if you look at the arc of what's happened in AI historically, there's this lag of like new capabilities relative to products on the market actually being available. And I think what Vibe coding is enabling is this like compression of like new capability to there's a thousand new products built on top of that capability on day one literally being deployed, which is a really unique combination.

Now if someone's hearing that, you know, tens of thousands of products being tested or Deployed on day one or week one. Give me reasons why. And I'm. I'm in your camp here. Give me reasons why people should not feel discouraged by that. Like, oh, it's. I'm already too late. Oh, there's haircut. I had that idea. Like, what are you seeing out there?

Yeah, that's such a great question. I think a couple of things. Like, one, the thing that gets me excited about this experience that we're building in a studio is how easy it is to go and remix. So, like, yeah, maybe somebody has, like, generally the same idea, but, like, you're like, oh, actually, you know, I am a hairstylist and, like, I have some unique insight or, like, I get my haircut and I hate these four things about that experience. And, like, therefore, I'm going to put a unique twist and a unique spin. And, like, that's literally like one that's a prompt away. So you're a prompt away from, like, yeah, maybe, like, roughly. There's other people doing the same thing, but, like, you have some unique, differentiated perspective. So I think even if you take, like, all of those, like, very basic verticals, there's like, this massive, like, tree explosion at the bottom end of, like, how to make a difference and how to keep adding features and, like, discoverability and how you position and how you make the experience fun. So I really think it's like every. At every one of those different thousand, ten thousand things, there's, you know, maybe a thousand other additional options that, like, could be interesting and fun and exciting products. Yeah.

And I think it's important that people start with what they know. Let's say you're a hairdresser, Right. And you just hear anecdotally, you know, what's that going to look like? What's it going to look? Do you want your hair like this? What is it going to look like? And you have a unique insight into the market that makes you want to do haircut. AI Right. But if you're just listening to this and you have no background in working in a salon, you're probably not the person to do this. But if you do a quick audit of your own life and experience, you could have ideas that are 10 times more valuable than that. Right. Like, when we solve our own problems with these vibe coding tools, it has the greatest likelihood of succeeding because we know it more intimately than anyone else.

Yeah. It is also fun, I think, while at the same time that that's true, Chris. It is also fun to just, like, take random shots at things that I know nothing about.

I'm like, oh, I do it all the time.

Yeah, that it's like, both things are true at the same time. That, like, while the value of your expertise and your edge is, like, very real, it's, like, also so much more approachable to, like, go into domains in which you're not an expert. And, like, maybe you actually just get lucky and stumble upon the thing that's, like, really interesting or you have some, like, random unique insight that actually is super valuable. So it is just another. Another plus in my book for this vibe coding moment, being, like, such an interesting experience.

Have you ever read the book Range?

Yes. Yeah.

Yeah. That's what I'm thinking of as we talk about this, because they did a study, and a lot of times the experts have, like, the worst insight because they're so biased and they look at something a very specific and narrow way, whereas an quote unquote ignorant outsider comes in and they, you know, it's like the midwit meme. They look at it like a caveman and they have this big unlock because they're not biased by their previous history. So think there's. There's a range there, right? Between, like, what you know, what you recognize, you being able to see opportunities more clearly than others, and, like, you being way too biased, right? You being like the old fogey in the room that's like, oh, that would never work. Because I've been in that business for 10 years. Like, I've had people tell me that, and I've launched anyway, and it's gone great, right? So I think it's a matter of, like, finding where your place is on that spectrum and just taking wild home run shots on goal just because you either win or you learn.

You know, I think you're 100% right. And it's also because of how accessible this is now. Literally, you could be doing and taking a shot at building something every single day, which is awesome. And, like, that's how I feel. And, like, I. I wish I had more time in the day, but, like, there's a thousand things I want to build. And, like, you actually could be building one AI app every day with, like, probably 30 minutes of work or something like that. And, like, again, Maybe the first 50 things you build aren't interesting and no one cares, but, like, the 51st might. And, like, again, you spend $0 in that whole process to get to the point where you're, like, you actually built something that you're really excited about. So such a special moment again from that Perspective.

Well, and could you imagine how arrogant that Advice would sound three years ago? Like just launch 50 different things. It's like, what? Like I don't have unlimited time and money and now it's like, no, that's 50 prompts. Like you can do that.

Yeah. There's something interesting about if folks follow like levels IO and some of these like people online who talk about like, you know, they had like 20 ideas and only three of them work. And the three that worked are now like, you know, a hundred thousand dollars a month businesses or whatever. Like that experience probably doesn't perfectly generalize across everyone who tries to do this. But like that is now like more in reach than it's ever been. Oh yeah, you can literally take a hundred shots at it and like, yeah, you should be thoughtful and deliberate about what shots you want to take. But like the barrier is like basically non existent, which is really exciting. Like it really is just like, do you have agency? You know, are you interested? Are you spending your time doing this? Or is like there's something that's more high value in the world that you're spending your time on instead?

Yeah, that's beautifully said. And I ask you this next question more as like someone who I know is a student of the game and like in the startup community ecosystem, seeing what's happening, how are you seeing people find success at actually marketing these apps that they vibe coded and not just building them?

Yeah, one of the interesting things is this goes back to like, what is the edge that you have? Like, I'm somewhat biased because I spend a lot of time on like seeing what people are building on social, which I think is like one version of this, of this like flywheel of like how people are coming to market these days is like you could basically for free social first engage and go and do that. I think there's another angle of this which I've seen more and more startups and this is like more in like the indie hacker bootstrapped space. But people are building these like packages of not just a single tool, but it's like, hey, you're a student or hey, you are X, Y and Z Persona. Here's the package of like 10 AI tools that you need. So it's not like I'm trying to make the tool to rule all tools. It's like actually no, you have 10 different needs as a student or as a, as a lawyer or as a whatever your your Persona is. And like I'm going to package all these things up. I think some of the Challenges like how you go to market with that package looks very different than how you would go to market with one of those individual tools. Like the search volume, for example, for like, you know, haircut. AI might be extremely thin, but you sort of package together the like seven different tools in some specific domain and then all of a sudden, like you've actually reached a critical mass where there might be a business that you can build. So I think that's one of the things that I've been most interested in. And I don't know if like, I haven't seen a lot of people doing this and could be an interesting edge to like, try to, to try to come up with what that package would look like for like a getting to a wider audience.

Well, I just interviewed a guy, John Rush. Do you follow him at all?

I don't know. I'm not sure.

It was episode 228 and he, he makes 3 million bucks a year from a dozen AI tools. Maybe like 20 AI tools, and it's definitely 80. 20, right. Like, most of them don't make much money, but he has like an, an AI SEO tool and a marketing tool and all these other tools, and it's all like he's selling them to startup guys, to indie hackers, et cetera. But he's able to cross sell and cross promote between all of his email lists and he's able to triple down on his same, like meta ad accounts and he's able to, you know, share all these resources, his AWS account and all these things. He just has a lot of surface area for learning what people want and what they're asking for or not asking for. And he's able to take all these shots on goal and, you know, if I might have signed up for this, you know, AI SEO tool for a hundred bucks a month but never actually paid for it, or I churned out, but then I end up spending $200 a month for this other tool that he offered me. I think that's a really smart idea. Like, there's no point in building something and not hitting publish on it. Right. Or telling people how or why you're building it. Like, it's very little incremental friction to publish what you're doing compared to actually doing the thing right.

Yeah. I think there's also something, which is an interesting point that you're making, Chris, which is, I also think there's just like a level of complexity in what you have to build. And if you think about, like, if you're just getting started Doing this stuff, like, again, building the like full end to end AI marketer that does SEO and outbound and ads and five, it's just like really daunting. And I think for a lot of people, like your experience is like probably not broad enough that you would have enough domain expertise to like really have a point of view on like what's the right product experience for all those things. But like, you can maybe build the like, sliver, again like thin version of like three of those products pretty easily. It makes me ponder what the future is going to look like as, as people sort of, you know, there maybe will be these like all the like big companies or maybe not all of them, but like all these companies are trying to build the Everything app and at the same time that everyone's trying to build the Everything app, like it's now becoming easier and easier to also build the like just one Thing app. And I think users actually appreciate that sometimes. Like, I don't want all this complexity. I just want to stick a picture of myself into an app and like, see what I look like with different haircuts. I don't need a bunch of other stuff.

Like, I mean, Elon Musk isn't even really finding success at building an Everything app, but you're, you're out there trying to build the next HubSpot, you know, like, what, what do you just build? A tool that puts furniture in people's empty rooms and go from there.

There's like very specific niches of this. Like, I love the furniture one, but like, I was helping my girlfriend with this. She wanted curtains. That was the only thing that she cared about. She was like, I don't need to redesign my whole room with AI. I just want to see seven different types of curtains and like going deep on that experience. Like, again, it's maybe, it's maybe the funnel to your furniture AI business. But like really the thing that people are trying to solve is like some very specific problem and you can win on the niche and like then get people back into the larger funnel.

I love that, like, I just wish people had more of a bias towards like that's not niche enough as opposed to that's too niche. AI curtains, you know, like, that's not too niche.

It's not too niche. But I do think, and it's interesting, I think this like, bias comes from like the historical cognitive dissonance, dissonance of like the cost to build software being high. So like all the conventional wisdom, the way that humanity has been programmed is like, you have to take a big enough swing for it to be worth it to build software. Because the cost of building software is high. And I think it is like, not trivial. I think for a lot of people that go through the like, flipping the switch that like, actually the cost of building software is now trivial and like the strategy in which you should employ in that world looks very different than the conventional wisdom. I genuinely think the way that the flip happens is you, like, experience at firsthand. Like you go build 15 different furniture apps powered by AI in a single day and you're like, wait, holy crap. The limitations are all different than what I thought they were.

Yeah, you spoke on a different interview about, I forget which product it was that watched like a 30 minute museum video and pulled out a list of exhibits. What are you seeing with AI tools that can, quote, unquote, watch video, even if it's just frame by frame? And what applications do you think are exciting around that technology?

Yeah, that's a great question. Do you want to build stuff or. I can, I can answer this question, but we should also. I'm happy to. I'll sort of give vibe coded answers to all of these questions and then, and then we'll try. And I'll also talk while it, while it writes the app.

So. And this is a fairly selfish question because this is something that I've been trying to do personally. Like, I want a tool that can like watch short form videos for me and like extract interesting points from it and where I can ask it questions. It can go out agentically and find interesting short form videos that I might be able to post content about. Right.

Yeah, I love that. So I'll kick off building something and then I'll give you my answer. So we want to build. Let's do. And maybe I'll try. I'll see if it's speech to text on a podcast. Does it work? Let's see. Build me an app that takes in a video and then parses out perhaps the most interesting stuff from the video, picks up on the trends, gives me sort of maybe an entire analysis of what happened, what the gist of the video is, what the takeaway should be, 10 insights. But make it so that I can actually use AI to customize that report that's being generated in addition to using AI to analyze the video as it's processed.

What a prompt. It's like watching Picasso in his prime.

Yeah, my, my word vomit of watching that, doing that prompt live without being more thoughtful about it.

But no, that's great.

Yeah. So to answer the question, like that example that I was referring to about a tool that can like take videos. And you know, in the example I was giving, it was a museum and you can sort of extract all of the exhibits in the museum. The beauty is it's just Gemini, like literally that is just like the Gemini model out of the box. So you can actually do this in AI studio. So if you just go to AI Studio, you upload a video you asked, the prompt is like, extract all of the different scenes from this video. It will literally go and do that for you. So it's like very simple, that works out of the box. The question then is like, what is the product experience that you sort of build around that model? And I think that's where the magic is. I've just seen, like, Gemini has state of the art image and video understanding. It's one of the only models that has native video understanding as well. And it's more than just like the frame. Like, it's not just if you were to like pull, sample a specific frame. It's more than that. It's like the interleaved frame and audio coming together and like the context that it's shared on a, like a frame and audio by audio basis. So it's a little bit more complicated than that.

Oh, man.

All right, video analysis engine. Now I need to see, do I actually have a video on my desktop that I can use for this? All right, 270 megabytes. Let's see if it says I can do supposedly up to 1 gigabyte. So let me drag this in and analyze video. Let's see. And so just to give context on this, you know, the prompt was build an application that takes a video, parses it, application should identify the most interesting, etc. Etc. You can see it in the top left hand corner. Gemini 2.5 pro. Went and, you know, did a plan, wrote all the code for this and actually made me an AI video analysis engine. I just copy and dragged this and was able to get an analysis, was able to get 10 key takeaways. And then the cool thing is we can actually, again, and this is powered by AI in real time. We could change what we want from this analysis. So, Chris, I don't know if you, if you have suggestions of like, how we would potentially change this analysis, but we can change it if you want.

Yeah. What about videos that have no text or no audio?

It's just, just, just like pure video.

Yeah, huh?

Yeah, yeah, we can also do that. So I think it's it has. This goes to like, Gemini has like, native image understanding and video understanding. So it can. If there's no audio, like, it's perfectly fine. We could find a video that has no audio and it'll actually, like, look, okay.

My mind is just like buzzing with the possibilities. I don't know why. I've never had much success with like an AI watching videos, but I just haven't. But this is.

I think it works really, really well. So this is. We have like tons of customers who are super successfully doing, like, AI watching videos. And there's this interesting loop now where it's like AI watching video that is then AI generating video with VO3, which is then like using Nano Banana to do something. Like, I've seen a lot more of this activity where you can sort of take this full suite of models, stitch them all up together and we have some examples of this where it's like, can you. Yeah, can you stitch all these things up together? And these are just a bunch of the, like, random examples in our, in our gallery right now. But yeah, lots of cool stuff. And like, I genuinely think you're probably three prompts away from something that works really well. Again, it's like critically, the Gemini model that's enabling that analysis.

What are some other use cases for watching videos that you've seen people use?

That's a good question. I think one of them, one of the most interesting ones is like, there's all of this intelligence and videos and it's just like, how do you extract intelligence from videos? Like, I talked to a lot of companies that just like, have lots of video data internally inside of the company and they're like, yeah, wouldn't it be nice? And I think like, content creators actually is like another great example where like, I, you know, very inactively do my own podcast and like, for that I'm like, you know, if I wanted some montage, I did this in the past where like, I basically uploaded all of the data and I built this pipeline that would just like go through the video and I would ask a question and it would like search across all of my videos, completely vibe coded and like pull out interesting insights that like, I didn't even know that I talked about that on one of my podcasts. It was super cool.

What are mistakes that you're seeing people make when it comes to prompting and. Or vibe coding that are easily fixed?

Yeah, I think the big one is like, people. And I actually think, like, building software is very much like this as well. It requires persistence, you know, in this example of that we were just looking at, like it kind of worked, it like kind of didn't. Like maybe it kind of like missed the point of like the second part of what I wanted. But like the beauty is the cost to continue to iterate is literally just like putting something in the prompt box. And we even added a bunch of, you know, suggested downloading the report as text or clearing error messages or refining report sections, a bunch of pre made prompts that will like show up here based on the context of what you're building. So I think people oftentimes if that first one doesn't just like hit the home run, they're like this thing doesn't work. But I think it's like really, you know, you're 10 prompts away from the home run and the important part is that you do the 10 prompts to get there. I think that's the biggest mistake right now.

With everything changing so fast, what skills or habits do you think will stay valuable throughout all these changes that we're seeing with AI?

Yeah, it's funny and like back to the similar thread of the last question you asked. I think everyone talks about like building soft, you know, being a software engineer is this like diminishing thing that you should be doing. And it's funny because I think about the skills that I've learned as somebody who's like traditionally trained in building software. As a, you know, computer science undergrad and someone who was formerly a software engineer. I think about almost none of that of what I learned having to be like typing, typing letters into a key into a keyboard that then translates into code. And like, I think almost all of it is like how to think about solving problems, how to be resilient. And I don't know about you all or other folks who have built software, but like my experience of building software is error messages. And I think that's like spending years of your life just seeing error messages. I think like things basically not working. I think trains you in a very different way of like how to be resilient. And I think the reality is like how to be resilient and how to have agency and like go and take initiative I think are like the two determining factors in all of this and just life in general. The challenge is like how do you put yourself into an environment where you can do both of those things. The nice thing is like for AI is sort of enabling you to be, have more agency. Like you now have all these tools at your disposal. The floor is sort of is leveled up so that you don't have to worry about, like, the drudgery of a lot of the tasks that you were doing before. I think it's a beautiful moment from that perspective.

Yeah. I mean, more agency can make or break people depending on the person. Right. It could drive us to be our best selves or more of our worst selves. But I think the people listening to and or watching this are, are in the former category. What are some overhyped, like, AI use cases that you're seeing right now that.

You'Re, you're not bullish on travel? I hate the idea that I'm going to be booking my flights with. And this is as somebody who travels too much. We were talking before this off camera. I travel too much. I wish I wasn't traveling. But I just think it's like the most quintessential like, demo use case that is, like, not actually valuable that much. I think it's also because the form factor that I've seen, like, I do think there's something interesting in AI, you know, enabled travel planning. But I think the product experience that most people are building is like, AI to book your flight ticket. I'm like, I don't know about you, but the United website is like, pretty simple to use. Like, I don't actually need AI to book my flight. I need AI to like, orchestrate the, like, actual complexity of traveling. And like, most of the products are not doing that. So it's just like the easy thing they're actually helping me solve. And I'm like, I don't need the easy thing to be solved. I need the hard thing to be solved.

Yeah, people are trying to solve a problem that's not a problem. And the cost of AI getting it wrong is quite high. I have AI book me a flight to Vegas with a layover anywhere. You know, it's really messed up. There's like 50 nonstop flights from DFW to Vegas every day. But that would happen, like, something in the logic would happen to save me $3. It would, it would have me layover in Miami or something, and that's to save me three minutes. What are some travel AI use cases that you think could be useful, if any?

There's something interesting about, like, how do you bring together all of these, like, disparate sources across the Internet? Like, you know, I think there's some amount of useful signal that you get from, like, main travel websites as an example, but there's also, like, you kind of want to discount the opinions that you get, like, depending on what experience you're looking for. If you want like a nuanced travel experience, choose your favorite like, rating website that, that rates like places on travel. Like, oftentimes you might actually get like, you know, it's four and a half stars, but it's like very biased because it's like a bunch of, you know, US travelers or it's like some, you know, specific Persona that is like maybe not the Persona that you're looking for, even though you might be a US traveler yourself, which is the irony. So I think there's something that, like, how do you sort of connect all of this stuff together and have like real sort of like travel intelligence of like, where you're going and like the places that you should be going to that like, maybe are. Haven't bubbled up to the top level of like, you know, the, the hottest place to be and also like discounting a lot of those places, like if you want to avoid those things. And also just like the logistics, I think about this, like traveling places that have like public transportation systems that are like new. I was in London a few weeks ago and hadn't used the like London train system to get from the airport before. And I was like halfway through this process of like walking around through the airport, I was like, I sure hope I'm going the right way. The information about how these systems work is like out there on the Internet. You just need to like build an experience that make. If I had gotten like a email from my AI travel agent saying like, hey, by the way, you're in London. Here are your three options for how to travel. And like, here's literally the map of how to do that. Like that wouldn't even be that hard. I could have probably vibe coded it beforehand, but I just didn't set it up.

What if it was like a travel companion that texted you? Like it was, it tracked you, right? Like Uber might and texted you as you got to different spots along your journey, like, hey, I see, I see you landed. You're going to head to Terminal 4 because that's where the new, you know, subway is, right? Oh cool, you found the subway. Awesome. I know you love Lebanese food. There's a five star rated place right at this stop, right? Like just kind of like texts along the way and you could tell it to stop at any point, but it's, it's tracking you text as the form.

Factor is something that I'm like overly, I'm overly excited and bullish on.

100%, 100%. I don't want more notifications but I can handle it if it's all within.

My messages app and like you're traveling with your friends and like all of a sudden now it's multiplayer in the conversation. Like I want to put my sort of travel companion into the group chat and like I'm traveling with my family in a few, in a few weeks and like I would love to be able to like just not have to be the arbiter of information about like what's our plan and our like all this like conversation toil that you have to have like put my companion in there, drop the little thing and then whoever forgets to check the Google sheet that has all the list of the itinerary, just text in a group chat and like the companion is going to answer like technologically like it's actually not that hard to build that which is awesome.

Yeah, absolutely. It's a Twilio integration. What other cool things like are you like did we not get to that? You're just excited like a proud parent to show me.

Yeah, that's a good question. So there's two that are top of mind and I'll show one of them and the other one we're working on like a great showcase for I think like voice AI agents and like computer use agents I think are two things that are really exciting. Computer use agents, we just launched the time of this recording like literally yesterday. A state of the art model that's able to sort of like navigate browsers and do all this stuff for you. And it's like a really helpful if you, if you're somebody who's like thinking about building agents, it's really helpful as like a supplemental mechanism to like go and gather information or to like automate repetitive like UI tasks that you're potentially doing. So really excited about that. The other one which is like adjacent to this is like voice agents. I think there's a huge amount of alpha in this. Like if folks have like tracked the voice AI market they'll like amount of companies, the model quality, the like investment, the use cases are all like up and to the right. There's like a huge amount of potential impact and like it's actually not that hard to build. And the fun thing is like you can literally go and vibe code voice AI experiences inside of AI studio which is really cool. So I'm going to show you this demo and then we'll, I'm trying to think of like what do, what do we want to build that's like voice AI powered.

I will Say I had this lead magnet that was just a form for this initiative I have and I just swapped it out with a voice AI agent and I'm up front that it's a voice AI agent and I see every day like the transcripts of whoever having a conversation with my agent. And I've talked to them like afterwards and they love it. Like they know it's AI. It's a great experience. I think we have this perception that a voice AI agent is taking over and people aren't going to like it. And it's more frustrating than being on hold with AT and T. But it's a better experience and if you know it's AI and you don't feel tricked, it's just superior. Extremely bullish on that.

Yeah, yeah. Well, that's actually a great thing. Let me try vibe coding this. So I'll, I'll describe what I understand of what this is doing and we'll see if the model can. We'll see if AI Studio can actually build me an app that does this out of the box.

Okay.

All right. I want a landing page for my website where we sell, we actually sell AI services. And I want the product experience when you land on my website to be an AI voice agent powered by the Live API. And basically the whole experience is going to be a. It's, you know, it's a veneer on top of a lead gen form behind. So the user will talk to the Live API and behind the scenes the model will actually just be filling out a form. And then once the model has the information that it needs, which let's say that it's like the name and the email address and the company that the person works at, the model will say thanks, it'll end the conversation and then it'll kick off a workflow behind the scenes to email that person and like create a lead gen and some database. So let's do all that. Let's powered by the Live API and let's make the website look futuristic. I want a little bit of. I want to feel like I'm in the future for this and my nickname for this is Yap yap to app. So we'll see, we'll see if. We'll see how good it is. This is a fairly involved. So like, this might be an example to, you know, level set my own expectations. This might be an example where like it doesn't work the first time because we're actually. I didn't think deeply about my prompt and I was truly just yapping, but we'll see whether or not we get anything interesting from this. But, yeah, I'm. I'm not surprised. There's that, like, not you as you were describing, Chris, that, like, experience that a lot of the users have. I also think there's this, like, novelty effect where, like, you might be able to, like, it's. People are surprised and interested in this because they don't see that experience everywhere. And, like, I do. Maybe there's a. There's a time window of, like, when that's true, but, like, it feels like it's right now where, like, people are open and willing to try because they know how AI is changing stuff.

Well, and the bar is pretty low. Like, if I need to call customer support for anything, my day is ruined. I'm. I'm yelling at my kids. It's. I'm on hold. I'm going to get hung up on. Like, that's what AI has to be. We will never be on hold ever again. Right. It's either that or I'm calling a restaurant to make a reservation and they don't answer. Right. Like, that's kind of what we're up against here.

Yeah.

All right.

It looks like it's done. 66 seconds. I have low confidence it's going to work, but let's try it. Hey, how's it going?

It's going well. Thanks for asking. I'm here to help you with Stellar AI. Could I get your full name, email address, and company name?

Yes, My name's Logan Kilpatrick. My email is emailyz.com and I work for Google.

Thank you, Logan Fitzpatrick, for providing your full name, email address, and company. I have noted that you work for Google. Is there anything else I can help you with today?

Nope, I'm all good. Can you contact Stellar AI and let them know that I'm interested? We were so close. Almost. Almost zero shot. Experience accepted.

That's amazing.

Yeah.

I'm curious, but I think it.

You sort of close. Like, I'm like, two prompts away from this thing. What I need.

Is it like a walled garden? Like, if you said, what's the capital of. Of Kansas, Would it answer you? Or would it be like, I'm not trained on that, you know?

That's a good question. So I think my. My hypothesis is like, the guard, actually. So this is where looking at the code is actually helpful. So if I go to code and then I go to. Let's see, not voiceagent. Tsx. Oh, no, here it is.

Yeah.

Actually, system instructions. You are a Helpful, friendly and personal AI agent for seller AI. Your primary goal is to collect the user's full name, email and company name, be conversational and be conversational and engaging. Once you successfully collect all the pieces of information, you must call the Submit Leads form function. Do not ask for other personal information. Give the user an introduction yourself before starting. So my guess is like this system instruction doesn't say anything about like don't answer other questions. So my guess is it'll answer right now. We can try and then we could try adding in like only talk about stellar AI. Do not talk about anything else and see how much it works. But my assumption is that works out of the box as well.

The pause, the latency is so much better than even a year ago. Right? It was, it was awkwardly long. I didn't notice it at all there.

Yeah, you don't. It feels, it feels extremely natural. It is actually always surprising to me how quick it is. But yeah, it is, it's, it's very fast. So I'm going to keep playing around this. I might maybe I'll add a voice AI agent to my personal website and see if people want to talk to my voice AI agent and I'll just have it make up random stuff and confuse people just to throw them off my trail.

Oh man, that is really cool. What are some things that you're excited about that feels like still very underexplored, like unexploited by the everyday person?

Yeah, that's a good question.

Let me ask a follow up which is related.

Yeah.

How big is the opportunity to have like an AI automation and or implementation business or agency that helps implement these tools into small businesses? Because you and I toy around with them and it's easy. You one shot. Prompt it. Cool. It doesn't matter how easy it is, right? It doesn't matter how easy Google Ads or Facebook ads or building websites are. Like there will always be business owners that just want someone to do it for them. It's not about the complexity, it's about the brain space and like the, the perception of difficulty. Do you see that as being a big opportunity?

Well, so I'll give my bias which is like I don't know what the actual numbers look like. My intuition and like my personal experience engaging with people. And I'll share. I was at a, an event in Santa Barbara last week and it was like some of the top executives in the world from like all the like Fortune 200 companies and even these top executives who I think are like obviously extremely smart, like in the know, they. They're. They have teams that are telling them how important AI is. If you, like, actually get into the nitty gritty, like, a lot of folks don't know how to use AI tools and, like, they would need to go and delegate to, like, go and do a bunch of these things. And they're like, extremely well paid, like at the frontier of sort of knowledge.

They can talk the talk, but that's about.

Yeah, exactly. And they're like. And it's like, really interesting to see this. And then they actually, like, need a lot of handholding in order to make progress on this stuff. So I think, like, you take small businesses that are, like, just trying to get by and like, go and make, you know, move the needle on, you know, their family business or some small business with only a few people, like, I think there's just not enough time in the day to be deeply embedded and, like, you show them these three experiences. Like, I was on a call with a clothing brand and they were talking about how they, like, want to do all these ideas with mood boarding and like, using AI and virtual try on. And I was literally on this call just like, vibe coding the things that they're saying. I was like, yeah. And like, we were just like, here. Here's the app that actually does that. And they're like, how is that possible that you already have this thing ready? And, like, you can give people, like, the timeline, the like, fact that it took 66 seconds to build this thing. And again, it wasn't perfect. But, like, you get the point across of, like, this is within reach, which I think is a lot of the cognitive dissonance that folks have right now is like, they don't actually believe that it's in reach. And I think there's a world where, like, you can actually just go. And the experience sells itself so much because you can go and take this vibe coded experience. You can go find this company website that you want to work with. You can take their products or their website, reimagine it with AI and like, that's your lead gen. And like, the cost to you is like basically zero. But you can do that outreach. And instead of being like, hey, I'm blah, blah, blah, and I do AI stuff, you can be like, hey, I redesigned your website using AI. Here it is. If this seems interesting to you, let me know. If not, like, your marginal cost was basically nothing. So it's like on both ends of the spectrum of building this type of, like, AI agency, you can do really targeted, really Rich, really deep outreach and at the same time, like, you have a customer base that's like potentially really excited and interested. And again, my caveat to this, I don't know how big the actual market is, but like, I would imagine it's large. And there's a lot of people who want this as a, as a service, so.

Well, I mean, there's. There's 400 million small businesses around the world. All of them need AI to some extent. Maybe not all of them can afford $1,000 a month, but I'm insanely bullish about that. What do you think VO3 is going to look like in five years?

I think it'll look like a toy, but also like an important, like to me, VO3. And this is from, from conversations that I've seen, like, VO3 was the GPT4 moment for generative Media. I think there's like all of these companies that are engaged in that space have now realized like, oh no, this technology is actually possible and going to be good enough. And like, it's basically here right now. The exponential increase in like, sort of, maybe not exponential, but the increase in like, capabilities and how good the models are becoming I think has made like, lots of folks across the ecosystem aware. So I think there'll be like a huge amount of investment over the next three years and it will be like a massive accelerant of like the media and entertainment industry, which is, which is exciting and sort of like the outcome that you would want as, yeah. As a, as a fan of the.

Space, will we be in, in a Cinemark Theater in five years watching A, a Tom Cruise movie that cost 500,000 to produce with VO3, is that realistic?

It's a good question. I don't know. My, my sense is like, actually what's more realistic to me is that there's this like infinite canvas of media that I think is really interesting. So I think, I think my, my expectation is like, the Tom Cruise movie will still have a high, you know, maybe something like roughly the production cost will go down, but it'll still be expensive. It'll still, you know, they'll still really deeply like art and craft put into making this. An experience that both like the people producing it and also the audience is going to love. The thing that will be net new is the, the content lifespan with this like infinite repurposing and adventure is going to dramatically change. So the way you experience content is like you'll have a base video and you'll be able to be like, wait, this is fa. I love this story, but what would happen if, you know, Tom Cruise decided not to hang out on the side of an airplane and to do something else instead? And like, you can go and see what, what's possible. So I think there's, there'll be, and I don't know, it'll be interesting to see how like, the platforms adopt this. But like, I think this infinite content, remixing and like exploration is going to be very much possible, which is exciting.

I completely agree and I'm not of the mindset that like VO3 is going to replace Hollywood. Right? The Kindle didn't replace books. There's going to be a world for both. But I just think it's going to look so different. And what we see on VO3 today.

Is going to look laughable in a few years.

And today it looks amazing.

You know, it's pretty good today.

What, what's like a Google product that you wish everyone knew existed that is just completely slept on and ignored and underrated?

It'll sound counterintuitive, but AI mode in Google Search, despite it being such a big product surface, like, has actually gotten like the, the user journey of like, I ask a query, it's sort of, I get an AI overview. I want to sort of go deeper and like, look at the, you know, look at the links and have a deeper conversation. One click into AI mode. The AI mode experience is like so fast. They're like constantly rolling it out into new countries and languages and all that stuff. So I think also specifically like the US where like the prevalence of like all of these AI tools is so large. Like, we don't think about AI mode that much in the us but actually like, it is likely the first experience that billions of people are going to have using AI. And it's actually like comparable against the rest of the AI ecosystem. Like, it's a frontier AI experience. It's like the quality is really great. It's super fast. I'm proud for the Google search team who was able to build that because, like, I think they've actually built like a really top tier experience. And it's great that that also gets to be the experience that like billions of people who haven't used AI before will, will get to feel.

Yeah, I feel like those that are on like the online tech bubble or the Twitter bubble think that ChatGPT is king and then that is secondary. But when I talk to my friends that are not on Twitter, which is honestly most of them, like my real world friends, they're not on Twitter, they're using Gemini and they're using AI search through Google as they should. Like we're millennials and we've been using Google our whole lives. It's the most frictionless way to experience it. With that in mind and this, I don't know, this is kind of a different question, but what is your bowl thesis for Google as a company in an age of, you know, Google's Google Ads being eaten by other LLMs?

Yeah, it's an interesting question. I think my worldview is everywhere I look, I get excited about Google and there's, there's so many interesting places where like Google has built and like search, I think is one of these where like Google has this like steward responsibility in the ecosystem. And like, if you look at what that means from a product perspective, it means that like our obligation to the world is to not, not like snap our fingers. And then all of a sudden everything changes inside of Google. Because actually in a lot of ways Google is the front door to the Internet and the Internet economy. So I think like, how you approach building that product and the responsibility that Google has to build that product looks very different than in other domains where like, it's a complete, completely green field. Like Google's basically competing against, you know, all of these like other top companies and like, it's super competitive and like it's a, you know, no, there's no clear winner yet. And so I think it's been interesting to see us make bets across the board. I think the thing that gets me, like, I'll make two, like small comments about what gets me most excited. I think DeepMind helping solve all of these scientific problems is like a very real potential positive outcome for the world. And I don't know if you saw this, but sort of folks at Google, we just got our third Nobel Prize in two years, which is awesome. So like Google really is this like home of frontier science that's taking place. Demis and John John Jumper won last year and then we just won one for a bunch of the guy who runs our, the Google AI Quantum lab. And so there's like so much frontier science work that's happening. It's really exciting. The other angle is infrastructure and I think Google builds the world's best infrastructure. So for me as like somebody who, if you think about the future, like ultimately dependent on this AI infrastructure that's being built out, Google is like well integrated in that world and we have our TPUs which are sort of on the frontier of what's possible with, with computing hardware and like I actually have two TPUs behind me, if you can see, which is really fun. So, yeah, I think there's. Everywhere I look, there's reasons to be bullish in addition to, you know, Waymo and all these other experiences. Like, it's just, yeah, it's a ton of fun to be at Google these days. And I think it's. We're just getting started. Which is funny to say given how, how great things have gone for Google.

You only realize how early you are decades later, Right. It's easy to feel that you're late at anything, but usually when you look back on it, you are early, like most of the time. Right. And we're very early. Is there anything that I should have asked you that you just find super interesting that people would find helpful?

Before we go, I'll make one last pitch. That AI Studio build. Get started vibe coding for free. If folks have product feedback or questions or need us to build things for you, let me know. We're, we're very much like building AI studio in public, which is what I'd love to do. So you're on the Internet, on email, Twitter, whatever. Send us feedback, Send a suggestion, suggestions. And Chris, thank you for, for having me on and for taking the time to chat. I would love to. We gotta vibe code some stuff together and would love feedback.

I would love that. And where can we find you, Logan, if anyone wants to reach out?

Yeah, I'm on Twitter, LinkedIn, Logan Kilpatrick, you can find me, find me all over the Internet where people are talking about AI.

Okay, well, thanks, Logan.

Thank you.

Hey guys, if you're still listening to this, it's probably because you haven't had a chance to take your AirPods out. You're still mowing the lawn, you're still driving, what have you. If you're still here with me, I would really, really love and appreciate a five star review on Spotify, Apple or wherever you get your podcast. It would mean a lot if you want to go the extra mile. Share this episode with a friend that might have an interest in starting a business. It would mean a ton. Hope you have the best day of your life today.