← Marketing Against The Grain
Loading summary
Transcript76 lines
- [00:01]
A
Gemini3 has been out for two weeks making a huge amount of noise. But just how good is it? Well, we bring on Logan Kilpatrick, the PM for Google AI Studio to blow your mind. Watch Gemini 3 completely replicate one of Google's big products with a single prompt. We go back and forth in some of the best ways that we're using Gemini 3 and some incredible data visualization ones that I am already using in my day to day work and has completely changed the way I look at some of the performance of our marketing. And not just that you want to stay tuned all the way to the end because we go through a Nano Banana Pro use case that I know you will use every day and really does showcase how incredible of an image model that is. This is probably been one of my favorite episodes that I've ever, ever done. It just sparked so many ideas of how we can be using the latest model from Google Gemini 3. And if you want to get all of those ideas for free, then stay tuned for marketing. Against the Grain with Logan Kilpatrick. Enjoy the episode.
- [01:09]
B
Did you know that most businesses only use 20% of their data? That's like reading a book with most of the pages torn out or paying for a coffee. That's 1/5. Full point is you miss a lot unless you use HubSpot. Their customer platform gives you access to the data you need to grow your business. The insights trapped inside emails, call logs and transcripts, all that unstructured data that makes all the difference because when you know more, you grow more. Visit HubSpot.com today.
- [01:52]
A
Logan, thank you so much for coming back on the show. I will say even though we were fans of Gemini and I use Gemini, I will like be really honest. I think it was like third in my stack rank of the kind of assistance I used. It just that's the way my kind of usage gravitated. The recent launches have been a total paradigm shift for me. I think Gemini 3 is my default model. It's just Nana Banana was really good. Nana Banana Pro is next level. I think it's the best image model. I think it's a huge feat that you have all achieved and for someone like me who is in the weeds working with AI, it's really exciting to see this kind of like leap in terms of functionality. How are you feeling about things and kind of since the launch, how are you feeling about the feedback you've all gotten?
- [02:35]
B
Well, first of all, I'm happy to see us moving up your personal stack rank. This is the eval that we're trying to hill climb is get you to use Gemini, which is awesome. I think one of the biggest differences with how we launched Gemini 3 versus how we've done even last year's, last December's model launch sequence was we did a lot of the iteration this time sort of privately versus doing it in public. And this was based on a bunch of feedback from the developer community. It is this balance. Like people want to get their hands on the latest model, but at the same time, us shipping a bunch of models and doing this whole iteration process, like causes a lot of churn for a lot of customers and you lose track of, you know, which version of this was better. It's sort of slightly better for this or slightly worse for that. So I think that's been the biggest difference. And it's also just chapter. It's like page one of the Gemini 3 chapter, which is also exciting. Like, obviously Flash has historically been the thing that we've been known for and that model's not even out yet. So I think it will be awesome to see like a real workhorse model to power intelligence across the Internet when we land that model, which is exciting.
- [03:38]
A
Yeah, I can't wait. So I think for our listeners and watchers on this show, we're going to go through some Gemini 3 use cases. If you're not using that model, you should be. We're going to show you why. We're going to go through some of the quote unquote, vibe coding things you can do with build. So that was in Google AI Studio. Google AI Studio is also phenomenal. I spend a lot of my time in there. And so that's your kind of like vibe coding tool. And we're going to build some fun apps to show people how you can use that tool. And then right at the end, you want to stay tuned because we're going to really show the power of nanobanano Pro, which is a whole part paradigm shift in terms of image models. And so we're going to go through those step by step. If you're listening along or watching along, we're going to get into each of those categories. So why don't we start with Gemini 3? I have a bunch of use cases or just fun things that I've done with it that I would love to show you, but I would love to kick it over to you and say, what are some of the ways that you're using Gemini 3 or you've seen folks use Gemini 3 and kind of anything you can show us live in Terms of the capabilities of that model would be amazing.
- [04:38]
B
So we're sort of early stages in making this happen. But inside it AI Studio we have this new build tab on the left hand side which allows you to go in and vibe code a bunch of stuff which is exciting. So we tried to capture like, you know, some amount of breadth of like the use cases that we've seen be really successful at 3 Pro. One of them obviously like everyone's building landing pages, it's pretty basic, but you can get some really, really cool and impressive stuff. Another one is sort of like combining the code generation capability with nanobanana Pro which we'll look at later on. A fun one is like games gaming is the largest media category in the entire world. And like as the cost to build like even simple games goes down, like I do think it actually unlocks like some really interesting opportunities that were like priced out of the market historically because of how expensive it would have been to build a game. So I think there's something there and that's what's really been capturing my imagination as far as like what the world will look like in the future as anyone can build games. So that's a good one. And then there's just like these sort of interesting capabilities of the model like SVG generation and voxel art, which I think are not super widely applicable, but I think can speak to the creativity of this model. And then my last favorite example is, I don't know if folks have seen this ball bouncing example, but it's like historically had been one the canonical examples of are models good enough at doing code. You'd ask the model to drop a ball in a sort of rotating pentagon or square and see if it worked. So I took that example and asked the model to make it like 20 times more complicated and in a single shot it was able to do that. And I think this is this example of we all as users of AI have been trained to don't ask for too much, say it in this way, try to make it very finite so that the model can easily do this. And I think 3 Pro was the first model where I was having to remind myself to be more ambitious. And I think that was my default stage. But as I sort of have spent so much time using these models, I've pared down what I've been asking for slowly over time. I'll say this ten times through this conversation, but I think you need to continue to be more ambitious with what you're asking the model to do. So that's My sort of like high level comment. I'll make one more comment which is as far as like the way that I've been using this new functionality the most in Gemini 3The most is actually this like product exploration. So I'll do this example live and we'll see how good we can make it. But I'm going to go create my first app and this is something that our product team and actually teams across Google, search, YouTube, et cetera, et cetera, are doing this. A ton of what they're doing is they're actually using AI Studio to build the prototype of the next generation of their product. So I don't know if folks know Flow, which is this AI video editing tool that Google has.
- [07:17]
A
This is what I'm using to build a sitcom. It's Flow. It's so good.
- [07:21]
B
Their team's incredible too. And like Vo Nana Banana, all that stuff like brings that product to life and they've built the sort of like next generation of that product like completely in AI Studio in the build mode. Again, there's other teams across Google that are doing this. So I took a screenshot of this existing UI and I don't know if you have suggestions of things that we can do to improve this, but usually I'd be like, hey, I have some idea for a feature that I really want. Clone this UI and then add these other features. I'll make this a really simple one, which is clone this UI and then add five really interesting new AI features to it.
- [07:59]
A
That's what I've been messing about with it. Just give it a, an app and tell it to remix it with five additional valuable features. And Gemini 3 is like kind of one of the first models, or maybe not the first, but it does it consistently where it's really surprising what it does. Models, for a long time they kind of just did the, I wouldn't say the average, but they did the best practice, if that makes sense. They did the thing you expected they probably would do. Whereas Gemini 3 and I have this fun thing I'm doing with it is like I have it working on a strategic problem for me and, and I say to it like, think outside of the box and then I say go 10 foot further from the box. Go 10 foot further from the box. And I tell it not to be anchored by what it's training said and like, you know, only give me brand new ideas that are so outside the box. People would be shocked if you told them. And it gave me two ideas. I was like, I would never have come up with them myself. Never.
- [08:49]
B
Yeah, this is the magic and this is the way that I've been. The perpetual sort of like, prompt that I'm using these days is like, whatever I want to do is. And then I just say, like, add five more features. I need to start saying, like, out of the box features or like, sort of giving some qualifying terms because sometimes if you don't push it in a direction, it'll sort of come up with a bunch of like, basic ideas which are not always helpful. But I'm like 25% success rate right now with this. You know, give me five really interesting new AI features and it just like continually comes up with really good ones. So it's building this app right now, and we'll hopefully be able to see sort of the start screen for AI Studio cloned as it was before, almost exactly, and then reimagined with these five next features. And again, I think, like, you know, for folks who are in the marketing world, like, there's something interesting about, like, what's the analogy of this in your specific role? So I'm curious, I don't know if.
- [09:40]
A
Yeah, can I take the screen? I want to show you some things that, that I did just for fun because I knew you were coming on. And I think what's incredible as well about Gemini 3 is its ability to, like, one shot things. And so just for our listeners, when we say one shot thing, you ask it to do something and it literally does it in one shot. Like, you don't actually have to ask it to make edits or anything like that. So let me just see if I can show you this.
- [10:03]
B
This is my obligatory comment because the. I get nerd sniped all the time about this. Supposedly what you're describing is zero shot.
- [10:11]
A
Zero.
- [10:11]
B
Okay, I agree with you. It should be one shot because it doesn't make any sense. So all the nerds who are trying to nerd snipe me anytime I say single shot or one shot. Like, I agree it should be one shot, but supposedly it's zero shot.
- [10:23]
A
Okay, well, it did this in zero shots.
- [10:25]
B
Because if it was one shot, supposedly you would have had to provided like a complete example of this. Like, basically this was the example you gave it and then you asked it to, like, remake this example or something like that. That would be one shot. I don't know, it seems stupid to me.
- [10:39]
A
Yeah, well, basically it did this in one go. And the concept here is, well, if code is cheap and we were having problems as marketers and keeping people's Engagement, then you could turn every land page into an interactive game. And so here is like a brand who they've created this ebook on like deep work. How do you focus? I think that's kind of appropriate because one of the not downsides of AI is like, for someone like me who likes to do a lot of things, it's nearly too democratizing of work and that I can do a lot of stuff. And so I try to do too much. And so they're trying to give me an ebook on how to do like de focused work. And so what this does, the game is pretty rudimentary, but I'm moving my little ball about it. You can't really see these things, but they're like, take a zoom call. You know, they're all tasks that I would get disrupted by. And so the thing is, you try to keep your ball away for 15 seconds and then when you've done that, you can download the ebook. There's like, you know, a fun thing for marketers to turn every page into an interactive game. You were talking about gaming, and I think you can actually do that. Like you can actually have a landing page, which is like a fun little game. And at the funnel, if you go through the fun little game, you get your thing, your prize. And that prize is whatever, whatever that company is trying to market to you. And so that was like one of many things I did. First of all, it took me a small amount of time, like one prompt, for Gemini to come up with three good ideas. And you know how it came up with the ideas? It came up with ideas based upon your most interesting ideas that you've shared on YouTube shows. And so the other thing I would just tell the audience is the integration with YouTube, to me is one of the most impactful things you have now as a marketeer, the Gemini integration to YouTube is amazing. And so it was able to go through and show me your top rated shows on YouTube. So it said, these are the shows you were on and got most engagement. And then it extracted some of the ideas you showed. And there was a podcast where you actually showed an interactive game. And then I said, apply Logan's best ideas to marketeers. And then it said, hey, you could turn landing pages into interactive games. And all I said was, show me.
- [12:41]
And then I did this. And I was like, that's 10 minutes of work. It took me 10 minutes. Insane.
- [12:48]
All right, if you want to build your own apps with Vibe code, we've got you. We've compiled 10 prompts you can use to Turn any business idea into a working app in just a few hours. This builds real lead generation tools that actually convert scan the QR code or click the link in the description. Now let's get back to the show.
- [13:10]
B
Yeah, there's something interesting too about like, the value of this is literally right now, if you're like doing this type of marketing work and you're like trying to find, like, how do you drive conversion? Or like, how do you sort of get people in an interesting way that they haven't seen before? Like, this is unique. Like, no one else is doing this right now. I've seen zero of this so far. So I actually really like this idea as like a 2026 holiday campaign. Another one that I saw that our, our team is actually chatting about doing in our own product is this, like, everyone does these like, unwrapped things. And historically it's actually been like, difficult to build people any amount of like, custom, nuanced, unwrapped experience. But, like, you can actually do that. Like, we were playing around with a prototype that we built internally of like taking everyone's like custom usage data for their user profile and then like building them their own unwrapped experience based on how they use the product. And you can imagine if you have like a diversity of users who are using your product in different way, the unwrapped experience maybe doesn't feel as unique for each person. So there's something cool about building that, like, ephemeral software on demand for people. Like, it's a. Your, like, total cost is probably pretty cheap, honestly, versus sending like a piece of holiday swag or something that costs like 50 bucks or whatever. It's like you could send them this really cool, personal, unique experience and it's like, you know, 40 cents of tokens.
- [14:30]
A
Oh, I love that. Actually, you've got me thinking. So I would have a company and I would probably gather some data around them and then based upon that data, it's like a ways to showcase, to bring the product alive. Is that what you mean? Could you explain what you mean by unwrap? Actually?
- [14:44]
B
Yeah, yeah, it's like wrapped the year. Like, you know, Spotify does that. Like you're wrapped that. Like, I see a lot of people who are like, trying to build that and it's like, it's actually kind of hard to, you know, set everything up the right way. But I think you could build that for all of your customers. Oh, God, that's completely bespoke in a really, really fast and easy way. And like, Give them a personal experience.
- [15:04]
A
Yeah. So basically, for HubSpot, that would look like we would have all this usage data and we could basically show these were your favorite workflows this year. And here's where your industry favorite workflows were within HubSpot. So you have inspiration on, like, how better to use the product.
- [15:18]
B
Yeah. And I think the fun part is, like, the thing I don't like about Spotify Wrapped anymore is like. And I do think there is something about, like, everyone's on the same playing field, which is what's interesting about it. But it doesn't feel personal because it's my songs. But then you look at the distribution of this and it's like everyone listens to Taylor Swift and these, like, five other things. Like, it doesn't actually end up being like that personal of an experience. I think the cool thing is, like, deeply personalized and like custom experience. So there's something interesting about that.
- [15:46]
A
The other thing I just want to quickly show, and then I wanted to just cut back to your UI remix that you're doing is just following along that little app I built. I know this is a kind of more basic thing that Gemini 3 can do, which is like the landing page. But I just, again, wouldn't undersell how great this is for marketeers. Look how quickly I can basically build a business as well, by the way, right? Like, I have an idea, which is this interactive landing page. It's a basically interactive game. And I said to it can. Okay, like, here's some details about a theoretical company. Can you set up a landing page so I can start to sell this product? The landing page is amazing. The copy is really good. I didn't give it really any information on the copy, the structure, but look what it did down here. I didn't ask for this and actually added like a little interactive game that's pretty basic. I can make it do something better. But again, I just found the actual page and the details. So good. So if you're a marketeer, you used to brief your design team on a landing page. You used to, like, brief them in language, right? Used to say, here's a JIRA ticket, here's some stuff I need. Every marketeer should be using Gemini 3 to, like, do a landing page, and then they can just easily send it to their designers. And, like, here's exactly what I want, right? So that language barrier that I think used to exist between teams and engineers and designers just doesn't exist anymore because you can prototype with these tools to actually show what you want. And I think in companies traditionally, the amount of cycles you spend because what's in your head is not what comes out the other end from the designers and engineers. None of that should ever happen anymore. None of it. And I just think the ability to code these landing pages is like incredible. Like, it really is incredible how good it is.
- [17:28]
B
Yeah, we should clip this out and also post it somewhere else. I think what you just described is so profound. And this is the shift that's happening because of AI. And like marketing is a great example of this. The bit is basically the same, which is like, historically product managers would describe the product that they wanted in some like, lossy way and then you'd have a designer try to bring that to life and then the engineers would try to maybe bring that to life. Now it's just you actually build the product. I mean, it's not the production ready version that's hooked up to all your infrastructure, but it's the product that you actually want to see.
- [17:58]
A
It's like AI has moved us from English language to visual language and that you can be much more visual in how you speak to people.
- [18:04]
B
Yeah, this was just the one shot example of this and it like redesigned and added a bunch of suggestions. I'm curious what they are. I could click around. I wonder what happens. Oh, interesting.
- [18:13]
A
Oh, you can create conversational voice apps.
- [18:15]
B
Hey, how's it going? Actually, I don't even know if it's picking up my mic. Let's see. And this is a great example. This is the beauty of Vibe coding. And all I have to do now is say, hey, the live voice part didn't work. Please fix. Thank you.
- [18:32]
A
This is in the demo. It build that. You're asking this question.
- [18:35]
B
Yeah, yeah, yeah, Nano Banana. Let's see if we can make it. I'm curious if this actually works.
- [18:39]
A
Oh my God, this is insane.
- [18:42]
Whole new functionality and we'll see.
- [18:45]
B
It's supposed to have added a bunch of new AI features and it looks like it just took like it added Nano Banana and it added this live API experience. But I actually think these experiences are so representative of like where the state of Vibe coding is. And like, I always, I love to keep stuff like this in because I think it just like makes you realize like, you probably will ask for something and then like you kind of won't know if it's working and you'll have to test it out and see. But I think the thing that makes this so powerful is that the next step is just like, hey, this thing doesn't work. I want it to work. Fix it and then it goes and fixes it, which is the exciting part.
- [19:22]
A
Just to be clear for people watching, we built AI Studio. It added some features I know from building apps with Google Build over the last week. You probably can fix those things within a couple of minutes. But I don't think we should get away from the fact that you asked it to rebuild an image of an incredible app, which is AI Studio, and it just built it all. I think that's pretty insane. This is how I do things with Flow. So I'm building videos in flow via VO3. The big unlock for me was I used to go text to video and now I go like text to image to video. So I go text to Nano Banana, Nano Banana to Veo. And so this here would be pretty cool app, actually. Just dropping in the images and you can just like automate them or animate them through VO3. That's a pretty cool feature.
- [20:07]
B
Yeah. And we have one other thing which has been super helpful, which is this annotate app feature so you can go in and do things like this text is a bit meh. Make it.
- [20:18]
A
Oh, I didn't know you could do this.
- [20:19]
B
Yeah, it's awesome. And then you just add this, then add to chat and then you can. We're working on this right now to be able to like queue up messages while something else is going. So it won't let me run this because I'm fixing the live API integration, but. And then it literally has the screenshot and all the details and it could just kick this off. So it makes this like visual editing process actually really fast, which is a ton of fun.
- [20:40]
A
That is so cool.
- [20:41]
B
Yeah. Your landing page example, I just want to show a couple of these. Like that blow my mind of how good some of the examples are. So this is like following my cursor right now and as I scroll through, like, again, this was a one shot landing page that somebody on our team made. But like a very interactive experience. It is kind of heavyweight. Like, it's not easy to make this happen, but just like so much subtle, like as you hover over it, the word view shows up on these things. So like so much interesting detail of putting this landing page together.
- [21:12]
A
And like, so cool.
- [21:14]
B
Yeah, so cool. This is like a prompt away from anyone, which I think is the really, really exciting part.
- [21:19]
A
Yeah. And everything can be different, right? Like for a long time now, the web is like you go to a party, you feel like you've done a really good job of purchasing your outfit, and you turn up and everyone's in the same outfit or different versions of the same outfit. Right. Like, that is the Internet, because it's all been templatized. And I think AI now allows you to really be creative and you can rethink the webpage. Right. I think how software is going to be rethought completely in terms of how we use it, but actually, you can really rethink even web pages and have these fun little things, like an idea for you all. And I won't give the exact name because you'll be able to find them, but there's like, these huge sites that get a ton of traffic online. And they are, you know, kind of template galleries, right. They have a ton of, like, templatized assets that the entire web uses. And, like, I feel like Google should just, like, own some of those. And then when you're going to like, pick your template land page, you can choose that or you can just build it via a prompt with Google Gemini. That's a great way to start to onboard people onto just building all those things uniquely instead of using the template through Gemini.
- [22:21]
B
Right.
- [22:21]
A
And if you own those, you own kind of the places that people are going to do that today. Because I just don't think there's a world where you should have to use the same template as everyone else when you can kind of build something from scratch and really make it much more unique and creative.
- [22:35]
B
I agree. And I, I actually, as somebody who is traditionally trained as a software engineer and has actually gone and like, bought a bunch of those templates before for, like, random websites ideas that I was working on myself. Actually, a lot of them are, like, exceptionally complicated. Like, you get this thing and, like, it's actually not easy to just, even as an engineer, to just, like, take it, rip out a bunch of things. It's like hours and hours to replace all the assets and change the text and make the colors customize. And like, if you really want it to be your own, it's a lot of work. And I'm doing this example right now, and I took the existing landing page and I said, change the color palette to match HubSpot's design, and we'll see if it gets the details right on this, which it probably won't. But, like, literally, it's just a prompt now to sort of customize this and you can sort of change all the detail and make it your own. And I feel probably for me, it was like 100x faster to do this than it was to, like, buy one of the existing templates and then refactor it.
- [23:27]
A
I also am from a computer science background. I was a software engineer. Not a very good one. That's kind of why I'm in marketing. I'm being honest. But now I'm a great one. Right. Because I was always pretty good on creativity and ideas and, like, understanding how to build things, but I couldn't build them. And I actually have so many different experiences of actually what you described, like building websites, trying to get templates, and just trying to customize the template and not doing a very good job. Whereas now I can't even really put into words how different it is. Like the fact that that landing page I showed and the one you showed, I would think it would take a month of trying to customize a template, and I don't think I could have done half this. I could never have done half the stuff that the template is even doing. So just the power you have in your hands with these models. I think the general world has not appreciated just how much of an unlock it is for ideas. Kip always says it's a bazooka for the creative mind, and I think that's one of the best descriptions for AI.
- [24:23]
B
I love that. Yeah. I also think it's special to that these role profiles are sort of evolving in such an interesting way. Like, now the marketer is the developer, is the designer, and if you have the interests, like, you can flex across so many different things. Which I think historically the sort of mandate was the way that you would create value in the world was by like, deeply specializing in some specific area. And now it's like, take that expertise that you have and sort of apply it across the spectrum, which is really, really exciting. And yeah, I enjoy it because I. I want to build cool stuff.
- [24:57]
A
It's done. It is it. Oh, it's actually here.
- [24:59]
B
Does this look like the HubSpot design language?
- [25:02]
A
Yeah, it's got our little orange and everything. And the fact that you didn't upload a style guide, you just said it just took that from the web, right?
- [25:08]
B
Yeah. I mean, in this context, I think it probably just like, kind of HubSpot's in pre training, so it probably knows, like, HubSpot's orange and stuff like that versus you would in some cases right now, if you're like random business, you want to ground this in like, providing some assets of like, here's what our style guide looks like. But yeah, it looks good. I like. I actually like this design more.
- [25:27]
A
Yeah. Yeah. I Like the kind of font effects that it's had and things like that. The other one I just want to quickly show for people watching, to show the power of this is the ability to do things like in the Canvas. And so if you are a worker today, and I'm going to focus on marketers, but this is agnostic of marketers, you can build everything you want as an internal app via Google in Canvas if you want. You don't even have to build it as an app. What do I mean by that? Right. So today you have to, like, use whatever kind of software your company is using. And I think what's going to happen actually is people have had to kind of use things in the same way as everyone else, because you have to have software that has like one too many users, but everyone actually has quirks and nuances and would like to use things in special ways. And now those folks can kind of just build things internally. They are the only ones using them. Right. And so I'll give you a couple of quick examples that I've shown in Gemini 3 and Canvas. And so Canvas is basically, you can switch it on as a tool and it builds these kind of interactive things for you. And so this one is a really simplistic one where I give Gemini 3 some. Now, this is synthetic data, but basically it's like synthetic data that would be basically be your sales calls. It created, like, fake sales calls, support conversations, like internal information you would have about your buyer. And then really how you do great marketing. And really being a great business is really tailoring everything to who your ideal customer profile is. But that ideal customer profile is usually like somewhere in a deck, right. And never gets updated and kind of gets forgotten about because it's like, hidden away. And So I had Gemini 3 create an interactive, like, customer profile, a little interactive app. And so it actually did a pretty good job. These are not specifically HubSpot, but it obviously used some inspiration from us. It's based on synthetic data. But basically what this has is it has four customer profiles that we were targeting, has a scrappy scaler. It gives you the employee size, gives you the ecosystem switcher, the overwhelmed solopreneur and the enterprise power user. So these are theoretical customers that someone is targeting. And then it basically tells me what is their primary risk. The scrappy scaler is a high price insensitivity based upon all the conversations we've captured on that person, all this support conversations. It tells me a little bit about them, their key motivations, their key pain points. I love this, it builds out the customer journey map. So when they're trying to buy a product, what are they trying to figure out at each stage of that journey? Because then I can optimize it for those folks. And then it has like voice of a customer. So it actually pulls out real quotes from that synthetic data. I give it. And then I can switch to like the overwhelmed solopreneur and it changes it down here. And so I get like a different set of motivations, pain points, primary risk. I get a different customer journey. I extract those from the voice of the customer and a little bit of a strategic opportunity. But here's like the value, right? So I can just create this, I can share this with the rest of the org, but anytime I get new information, I can just upload it here and say, upload the Personas. And so they're always in sync with my latest internal information. And now the entire company can just see it in a really visual way and make sure they understand who we're marketing to. Now what I actually do is I basically ask it to export this into a two pager, and then anytime I'm using AI, I upload it as a doc and say, tailored this thing to this two pager. So you make sure that all of the output is tailored. But prior to AI and prior to like Gemini and Canvas being able to do this, this was just like hidden away in decks, right? Like, this was not possible. There's no software that does this.
- [28:59]
B
Yeah, it is super cool. And I think the magic is we were looking at AI Studio before and now you're showing the Gemini app Canvas. The magic for the Gemini app Canvas is it's like so accessible to so many people. Like, the Gemini App has like 650 million users or something like that. And you can just go in and be like, make me this thing. And it does like a pretty, pretty solid job at bringing that thing to life. There's another angle of this too, which is, and I think this is actually doing a good job of this, which is this, like data visualization. Our growth. The woman who does our growth marketing usually sends out these, like, historically had been like, pretty boring, basically, like CSV data outputs of like the results of our emails. And she started vibe coding this experience now where she can actually just drop in the CSV and it'll like populate this like, world visualization over time of like the emails being sent. Kind of like Shopify does for their Black Friday where they're like showing the like, pings all over the world for Email like literally in a single prompt was able to do that now and then it just like it also updates based like if there's some like random new fields or whatever, it just like redynamically takes in the input based on what it is.
- [30:05]
A
I can just upload the CSV and then I can unplug the email performance.
- [30:09]
B
In a globe per time over time too is interesting. Yeah, it's so cool.
- [30:14]
A
So this is the other use case that if you're going to just do a couple of things from the show, being able to just upload data and visualize it in any way you want is just again an incredible superpower. So what this here is doing again based upon synthetic data, it's just taking a business's marketing campaigns from a quarter and uploading them and then showing you a bunch of things like their spend, the total revenue, your blend. Like I just cannot overemphasize how incredible being able to customize this stuff is for a marketer. Your blended cac, your total conversions. But look what it's doing here. So what I asked it to do is like every quarter what a marketer will do is they'll say, okay, well I have a quarterly budget, I need to assess what's going well, what's not going well, and where should I redistribute capital. And this is doing it for me. So it basically says, okay, here's how we're going to redistribute capital in the upcoming quarter based upon what is the best return on investment. So it looks at the LTV to CAC Pro channel and then it says, okay, like increase enterprise search, increase mid market search and then decrease these things. And then it gives me what's working, it gives me the kind of patterns and recollection and then it gives me a segment kind of breakdown for Q3. And again, this is all just from a CSV upload and it's incredible. But the fact that can really make strategic recommendations now, which is kind of like scary in some ways, but like you can move very fast if you kind of unlock this for your team.
- [31:34]
B
Yeah, I'll show one other really, really similar example to this. But it sort of has like one interesting flavor of difference, which is this data visualization I think is one of the most impactful things. Like instead of making a, you know, chart and a slideshow, like a slide deck to like present to other people, this was an example of I took the Gemini 3 eval result page, which was like not interesting to look at and was like actually really ugly, unfortunately. And I asked Gemini 3 to actually turn this into an interactive visualization. And the fun thing is, like, I can actually answer questions now, different questions in real time. Like, we go into these reviews inside of Google, and people are like, oh, you know, it's great we have these different models. But, like, actually what I really care about as a leader is I just want to see Gemini 3 versus 2.5 Pro. And you can now sort of do that in real time and see all the differences. You can actually, like, change, like, hey, as we toggle some of these knobs. How does this impact what model shows up first? There's, like, a chatbot at the end to ask questions about the raw data itself. So, so interesting and powerful to, like, see data brought to life in really, really different and new ways. And, like, again, this was not some, like, sophisticated prompt. This was. I literally took a picture of a table of data and said, turn this into an interactive visualization so that I can answer questions more deeply about this data that I'm looking at. And it just works, which is crazy. So I think there was historically this narrative of, like, you had to be sort of like a prompt wizard in order to sort of get the most out of the models. Today, just ask what you want the model to do in plain English, and, like, it should be able to do it, which is really cool.
- [33:09]
A
I'm glad you touched on that. I actually just want to make sure people listening, really understand that point. There was a time where I felt my ability to prompt, and I'm not, like, in any way like your prompt engineers or anything like that, but for someone who works in go to market and is working with AI, I felt I had, like, some leverage because I really was in the weeds learning how to prompt, learning how to craft prompts. And for a long time, all I did was share prompts on the show, and people just wanted prompts for me. And now I just think I don't know that I do anymore, because I just ask Gemini 3 to, like, either craft a prompt or just do something, and it just does. Just interprets what I want so much better. You see that? That's a notable change in Gemini 3.
- [33:49]
B
A hundred percent. Yeah. Yeah. And there is, like, the frontier of where, you know, having a really differentiated perspective on prompting I think still exists. But I think with every model that comes out, the smarter the models are, the more intuitive they are. It sort of, like, flexes in a different way. So I think people are now, like, figuring out, like, where is the alpha as far as you know what you can tell the model that will get it to do things. I think one of the pieces of alpha is what we talked about before that I promised I would keep reiterating, which is ask the model to do more. The model is like basically pent up with all of this creative capability. It just needs you to like tell it to go and do stuff and like do more than you would have expected it to historically have done. So I think that's actually the most alpha right now is just like ask for more stuff and you'll get more. And like even if you know 40% of it is garbage, like the 60% of like net new incremental stuff that you wouldn't have thought about can oftentimes be really helpful.
- [34:39]
A
Yeah, yeah. It comes back to the curious mind. Could we end Logan on anything? You could just show like Nana Banana Pro is just, I don't want to be one of those, you know, influencers, insane, mind blowing, blah blah, blah like get the YouTube clicks. But it is insane and it is mind blowing and I just. Could you just maybe end on showing a little bit about that with that model of that app?
- [35:02]
B
Yeah, let's do this. So again, we're in AI Studio. There's a couple of things at a high level to think about for Nano Banana. So one of them is this is the first time that we've been able to sort of combine grounded world knowledge with an image model together, which I think unlocks a bunch of really interesting stuff. So that's what we're going to look at here, which is basically Google search plus the power of generating images. And I'm going to say show me the bull thesis for why in person home marketing is a good idea for HubSpot. We'll just pick on HubSpot and we'll kick this off. And what it's going to do is generate a diagram and an infographic pulling in real world knowledge using Google search, about HubSpot, about sort of like business trends, all this stuff. Again, this is a vibe coded app that we're looking at here in AI Studio. You could fork this and build your own version of it if you want, or just use it out of the box and then it's going to take all that information and synthesize it into nanobanana. And again the cool thing is we'll see the fidelity of text generation, which is one of the big leaps for this model is not only its world knowledge and its understanding and its ability to access tools and reason, but actually also like Rendering text coherently. Incredible.
- [36:14]
A
Yeah.
- [36:15]
B
Is so difficult. And this model. This model crushes it. So I'm excited to see this example and I love this UI as a sort of a gimmicky ui.
- [36:24]
A
So, first of all, this app is awesome. I'm using Nana Banana at the moment to create characters that I'm doing and a sitcom that I'm using VO3 to create. And it's like, I assume, like, every filmmaker is going to be using this model because it's character definition, its ability to create scenes. The other thing you can do is, like, you can mix and match images. So I created, like two characters. I created neutral backgrounds, and then I created a scene, a setting, and you can, like, ask it to combine those reference images into another new image. And so it also feels like a. Just a huge step up in terms of capabilities. Do you see Nana Banana Pro becoming much more intertwined into, like, filmmaking, things like that because of just how good that is, like, how much it can actually accelerate the process of doing things like that?
- [37:10]
B
Yeah, it is a good question. You see this for filmmakers on sort of both ends of the spectrum. I think on one hand they're sort of like, what does this enable for, like, existing professionals? And I do think, like, giving them sort of scaffolding to bring their ideas to life in ways that were expensive before, I think is super exciting. And then feeding that to Veo is a flow that I think we're seeing a lot of success with. I think the magic and the thing that gets me most excited as sort of not a professional in that field is like, raising the floor for everyone to be able to create stuff. And I think there's, you know, there's a similar story actually, with like, YouTube as an example, and how YouTube enabled everyone to sort of become a creative storyteller and all the sort of generational talents that has been discovered and created through that. I think we're going to see that potentially with filmmaking and sort of some of these other categories, like coding, et cetera, just because the AI capability keeps leveling everyone up and making it so that this stuff is possible, which is so exciting.
- [38:02]
A
I was never really a gamer. I think I could be a gamer in another life because I just have a. I have a personality that gets, like, really obsessed by things. But I think this is my version of gaming. I was up last night and I was going like, nana banana to VO3, nana banana. And it's like doing something. I can't explain it. It's like doing something that I could never do before, I could never even imagine creating like video. Creating a little show that I can show my friends or creating like a world class ad. Right? Like being able to just like create a video ad. And so it's kind of like addictive because it allows people to do things that, as you said, level up things that you would never have been able to do in the past or I had even thought of. I never thought that I would be able to create video in that way.
- [38:40]
B
Yeah, 100%. I'm, I'm trying to get an example that's actually working. It's just taking a long ass time to do this research, which this better be the most impressive research task ever because all of a sudden as soon as I said that it started working, as soon as I threatened verbally that I was going to, you know, go report this, it is done now, which I didn't even do anything that impressive, which is interesting that it took so long. Show me the bull thesis for why in person home marketing is a good idea for HubSpot. Let's see the field in person. Data capture, the bridge. Automated nurturing sequence, the offense. I'm unimpressed. I'm unimpressed. I'll ask the model in a more generic way, but I'd love to see your use case.
- [39:19]
A
Yeah, let's see if I can do this one here. All right, so this may or may not work. And so basically what I did was I had a Gemini 3 critique the above the fold of our homepage and then basically give a bunch of feedback. And so what I was going to try to do is see if Nana Banana can actually apply that to our homepage. And so again, in this way, what a marketer would have to do in the past is like get designers to do iterations and see how it looked and things like that. Whereas now I think Nano Banana can actually just apply those changes to the model. And so if you're a marketer, you can create like many variations of a page and start to get feedback pretty rapidly without the need of a designer. I will say that I did not make it easy. I give it a pretty large amount of changes to make. So you can see here a lot of critiques of that page. I asked it basically to make it more appealing to sales. At the moment it's a little generic. And so look at this. So it actually does recraft it. So like basically I give it a Persona, a sales leader, and it recrafted it pretty well actually for the sales leader. The sales engine built for scale in teams, move beyond spreadsheets and rigid enterprise tools, get the visibility of VPD's and the usability of reps. Will love. It's pretty good. The ability to just quickly remix images is really interesting. Like you can critique things and remix. And again, I will say that I didn't even remix this. I didn't even ask it to make these changes. Just to make sure people understand. I give it an image and it critiqued the image itself and then applied the changes through Nana Banana.
- [40:47]
B
I love it. I think it opens up this idea of like image space being something that anyone can manipulate. And I think historically you have to be a designer or like have one of these professional tools in order to have image space be where you iterate. So now it's like text space, code space and like pixel space. Are these like 3 degrees of modality and freedom that you have Just to finish close the loop on this example, I just took this and sent it raw to the model. I actually don't know what was happening in that app. The UI was cool, but I'll have to debug what the system prompt is. And I said, make me an infographic for why out of home ads are good for HubSpot. I turned on grounding with Google Search. I popped this open. This is exactly what I want, which is like really dense, rich text sort of telling this product story. This is almost a little like too much text. I think I could do the iteration to sort of pare it back, add more data maybe for like different markets, like targeting different segments and stuff like that. But I think this gets to the point of like, what is powerful, literally with a single prompt you can just ask for the information in an infographic or some other format and the model does a really, really good job of like pulling in that information.
- [41:48]
A
Yeah, this is all text within an image. Six months ago, this would all have been wrong. Like the change here should not be understated how incredible this is.
- [41:59]
B
Yeah. And it's like pixel perfect. There are mistakes every once in a while for sure, just because that's how the technology works. But. But it does a really good job in general of like not making weird errors. So it is like almost production ready just out of the box, which is cool. So cool.
- [42:15]
A
Yogan, I would probably end up keeping you on the show for hours and hours and hours. So I better let you go because you've been really gracious with your time. Again, huge congrats to you. Huge congrats to the team at Google. This is an incredible set of tools for everyone who is listening and been working and trying to integrate into how we actually do work. And I think we've given a ton of cool use cases that you can go and start using straight after you listen to this show.
- [42:37]
B
Yeah, this is a ton of fun. My sort of quick ask is as folks build with Gemini, if you have feedback or things you need or like, whatever it is, like, we want to make this model sort of like the most usable model in the world for folks across every discipline. So please send us feedback. My email's online, I'm all over the Internet. The comment section of this episode, like, send us the feedback. We want to make the product better. We want to make the models better. And this was a ton of fun things for spending time jamming and hopefully folks had some good ideas from this stuff. Cool.
- [43:04]
A
Thanks everyone.