B (4:09)

Yeah, and also I think AI Studio, like building these apps is free, right? Like are you charging your money for this or no?

A (4:15)

Exactly. No. So it is free. So there are certain parts of the experience that require a paid API key. Let me find the other example that I kicked off. What did I kick off before? Oh, the cross platform social media content generator. So we kicked this off. It took in all the context that we put in it spent, you know, it's been running for 190 seconds. A couple of things that I'll I'll call out. So left chat panel, you can sort of continue to iterate on your app. You can see all the files that have been created and you can hover over and it'll give you like the quick tldr of what's happening in the constants ts file or what's happening in the app tsx file. You can then preview your generation and then you can actually go in and see the underlying code. So if you're a developer, you sort of know how to code and you want to make manual modifications. You can actually do that if you want to, though you could do everything without seeing any code. So for this one we're using Nano Banana Pro, which is why it's prompting me to put it in an API key. So I'll go ahead and I'll do that and then we'll hopefully be able to build something and we'll see. What was this idea again? Create cross platform social media content generator. So we'll see if this works. What's our idea? Peter, you have anything?

B (5:24)

Let's do a rant about product management. I'm glad that. Okay, let's rant about, you know, stop the cross functional meetings. Let's actually just prototype and show it to users. Yeah, something like that.

A (5:39)

I like it. So we're going to take this in and this is the post idea and we'll make it a witty tone and we'll do it 1k. Sweet. I'll kick this off and we'll generate it. We'll see what happens. So behind the scenes, this is using the nanobananapro model and I'm assuming it's actually going to use Gemini to then take in this input that we gave, put it into a format that nanobananapro is going to actually be able to use and then it'll generate that image for us. So it might take a few seconds to first generate the actual prompt for the model and that. Oh, I like this. Okay, so LinkedIn is. I have a confession to make. I love a color coded roadmap as much as the next product manager. There's something deeply satisfying about moving a JIRA ticket from backlog to in progress. Let's be honest. Yeah, this is interesting.

B (6:29)

You gotta tell it not to add the hashtags, you know, but you know.

A (6:32)

I know, I know. Well, this is, this is too much in pre training that everyone thinks there needs to be hashtags on X, which is funny but actually nice. So we also have images that go with all these things. I kind of like this one the most, to be honest, of all, of all three, the one for X is pretty engaging. So a good example of like this is like the fusion of three or four different things and you can actually go in if you. This is the one time where like the code is actually quite helpful. You go to the code. We have this Gemini Services file which sometimes has the prompts like actually yeah, this is what it has here. It's like you are professional social media manager. The tone, the requirements. So here you can actually go in and like do a little prompt tuning if you want to. This seemingly works like reasonably well, but here I could say like for X, never use hashtags, they are lame. And then you'd be able to sort of make use of it. So lots of cool customization and we can keep iterating, we can keep asking for additional features or customization in different ways.

B (7:31)

Yeah, and I will say that like I was playing with this the other day and you know, I try to use like nanobana to build stuff in cursor too and it just took more work. Like I think in AI Studio you have all the Google stuff hooked up already, right? So like it's just I literally had to click a button like make me a nanopane app and my prompt and it's work.

A (7:51)

Yeah, it does work. I mean this is the magic for us. Like this is part of why we wanted to do this too is like we have an interesting edge to be able to hook all the Google stuff up. And you can also do things like deploy directly to Google cloud so that you could like make this a production ready app if you wanted to. So there's, there's lots of, lots of interesting and unique things like that. But let me, let me go back really quick just to follow up on our. The, the other example. So I'd asked this to switch the ball bouncing example to more like water droplet Fluid Dynamics and it kind of is getting there, which is interesting. So it seems like there's a lot more. Maybe each ball is a drop of water or something like that. Yeah, yeah, but gets very interesting to see. Like you know, you can do a lot of remixing and refactoring of this stuff, which is super cool.

B (8:38)

Yeah, it's easier to remix the stuff, man. Like if I want to build a beautiful landing page, it's easier just take an existing landing page and just remix it. This episode is brought to you by Linear. Let's be honest, most project management tools suck. That's why I love working with Linear, because They share my obsession for craft and quality. Linear is incredibly fast, beautifully designed and comes with built in support for AI agents like Cursor. As a PM, I love using linear to capture customer feedback, draft PRDs and manage development and communicate across the organization. Product teams at OpenAI, Vercel, Brex and Cash App all use Linear to build complex products at lightning speed. See for yourself by visiting linear app BehindTheCraft. That's linear app BehindTheCraft. Okay, now back to our episode to

A (9:26)

your question before of like, how do we potentially use AI Studio inside of Google to like increase the product velocity overall from a Google perspective, but also from an AI Studio perspective. This is what we do. We use AI Studio to make AI Studio. So I clone the AI Studio UI all the time and I ask for like adding features or modifying and we can actually do that live if that's helpful. But it's one of my favorite, it's one of my favorite use cases for product development or pms to like take their existing product and like add features.

B (9:54)

Do you have to give us some sort of a style guide or something to make this?

A (9:58)

Yeah, so this is. Let me do it live and I'll move my tabs around. So I literally, I'm just going to go like this screenshot and capture and then I'm going to go start, I'm going to say clone this UI exactly. Use the same styles. So we'll kick this off, we'll see how quickly like one of the things that I'm always impressed by is just like how fast it's able to make this product experience work. And again like our team is literally doing this like dozens or hundreds of times a day of just like taking the AI Studio product experience, iterating on it and other teams across Google are doing this as well, which is, which is really, really fun. So I think you're seeing like a real speed up in how quickly we can iterate on some of these ideas

B (10:41)

while we're on this topic. Like you know, like in the old days we used to write the PRD and the design and everything first. Right. But I'm guessing now you just do the prototype first and then do you still do the PRD and design or you kind of just skip, skip that thing or like.

A (10:55)

Yeah, yeah, it's a good question. So I think for us right now the place that we're at is there is some detail in. It's like all three at the moment is basically the short answer. So we do PMs at least on my team are still writing PRDs. We still have design like actually doing a lot of this work. And the last part is like we also basically require now you go and build a functional prototype. So like it is a step part of the process is to have that interesting. But this is pretty darn good. I feel like it looks almost exactly like the AI Studio UI and we got in 68 seconds from an image to basically a fully functional version of the AI Studio ui, copying most of the details. And then the fun thing is like I can just iterate from here. I could be like, so what would it be like if I had an X, Y and Z feature instead? Or this thing was collapsed in this way or we presented information in this different way and do that iteration really, really quickly.

B (11:54)

That's awesome. Do you like the planning and development stuff? Like do you ask it to write a plan first or like I think if you're just doing iteration you can just tell it to make updates. Right? You don't have to make like a whole plan, a spec or anything.

A (12:06)

Yeah, I don't, I don't do that. The only time that I do that that's like common in my workflow personally is like taking a PRD actually. So oftentimes what I'll do is like I'll spend a little bit of time thinking about like, what should we actually build? I'll write that as sort of requirements and a bunch of cjs and stuff like that. And then I'll dump that into AI Studio with a screenshot of the UI and a bunch of information like that. And I'll say like, make me, make me this product experience now. And it depends on like if it's a brand new product experience. Like we were playing around with like what does design mode in AI Studio look like? Because we're going to add that. That was one of the examples where like I wanted to let the model sort of make a bunch of decisions and like show me some different ideas based on like the requirements that I put together. So. And I got some pretty good stuff which is pretty fun.

B (12:50)

And how do you get it to show you different ideas? You just make a bunch of tabs and give it different directions.

A (12:54)

That's a good question. One of the things that I do is I often say, and I'll, I'll do my favorite like yap to app feature. So I'll, I'll talk to the model so that way you don't have to me type. But I'll show you here. All right. I really like what we have. So Far. But I want to see maybe like five or six different styles of what the like, you know, design of this UI might look like. So make five or six different styles and then add a widget in the UI so that I can actually click through the different styles and see all of them without needing to like reload my page or anything like that. So I do this. I always have to do this in a very verbose way. I don't know like what the like succinct way of saying this is, but it's a super common workflows. Like I want to literally just like have something in the UI where I click a button.

B (13:39)

Yeah.

A (13:39)

And then as I click that button, it shows me five or six different UI options.

B (13:43)

Yeah, that. That's very smart. Yeah. I was thinking you would like open different tabs, but this is way easier.

A (13:48)

The tabs works as well. I mean, it depends what, like if I'm really getting like truly different options, then I will, I'll use different tabs. So it helps from that perspective.

B (13:57)

Okay, so let me ask you this. So you have. So when you go present to Josh or whoever, right. Like, like, like your boss. So do you. We have a product review with him. Do you guys look at the prototype first? Do you look at some Google Doc first? Or like, how does the product review actually work?

A (14:12)

Yeah.

B (14:12)

Or is there even not a problem? You just send the link. Hey, play with this.

A (14:16)

Yeah, that's a good question. We don't do a lot of product reviews with Josh, but maybe we should. So Josh, if you're watching, you know, I would love to do more product reviews of you. Josh is very busy. He's got a lot of, he's got a lot of stuff going on. We do product reviews internally and the demo is, is often like something we do. Sometimes people have slides and like the slides will just like give some random context and then we'll go in and look at a demo. But almost all the time we're looking at like live demos that like actually the team can play around with. And the most fun part is like, it's not just about the team playing around with it. Like everyone, like 10 people in a meeting could start iterating on that idea because we all start from the same base level and like add new features. So it is really, really cool to see, to see that play out, which is fun.

B (15:04)

Oh yeah, yeah. You can basically 10 people can all remix this thing and just add their own stuff, right?

A (15:09)

Yeah, exactly. Yeah, yeah. And this is maybe an example. Let's See, let's see. Let's choose different styles so when I click on them. Yeah, this is what I love to do. Like, I'm just like, I can't. I could never imagine what a pink version of AI Studio looks like, but, like, now I can. And it's actually really helpful to be able to see these different styles show up. And the brutal. Let me say I do this.

B (15:31)

Oh, nice.

A (15:31)

Okay. Yeah, yeah, this is. I mean, it doesn't even look that bad, to be honest with you. It kind of looks good, so it's nice. Yeah. And I usually do this more for, like, different feature decisions and stuff like that. Less for style, but it actually is really helpful for style. So we're also working on a way to, like, import your style guides and sort of, like, have that type of context in AI Studio itself.

B (15:53)

Got it. Yeah. This is awesome. And I think this looks so good also because of Gemini 3. Like, I don't think the previous versions could have done all these crazy styles like this, you know?

A (16:03)

Yeah, I agree with you. Yeah. Actually, Gemini 2.5 Pro was pretty good at copying UIs. That was actually still one of, like, prior to three, that was the most common use case that I would use AI Studio for was just, like, clone the ui. It just. It sort of. It definitely wasn't as good, but it was. It was still surprising how good it was in some cases.

B (16:23)

So. So, like, so I guess, you know, there's no point to Amar's job anymore. I'm just kidding.

A (16:29)

No, no. Amar is really. So here's. Here's the interesting thing is he is so talented that it makes. It puts how good Gemini 3 is into perspective.

B (16:39)

I see.

A (16:39)

And actually even more so, the other designers we have on our team are also super talented. So it is. I do very much appreciate it because they just think of things and there's, like, subtle details, like, I don't know. The limitation is I don't know how to instruct the model to even, like, do some of the things that they do from a design perspective. Like, I wouldn't have, like, thought about it that way. So the limitation is, like, if you have that knowledge, like, you can really wield these tools to do these things, but, like, I certainly don't have that knowledge. So.

B (17:07)

Yeah, that makes sense. They're promptly. In a different way and, like, yeah, it just looks like if you're an engineer, you can probably build better, you know, more robust apps.

A (17:13)

Right.

B (17:14)

So 100% makes sense. Should we try the floor plan one? Let's try to Floor plan?

A (17:19)

Yeah, you can do floor plan or I kicked off this other one. Before that you had suggested build an interactive map space chatbot that uses real time Google Maps data to answer location specific questions.

B (17:29)

Yeah, that's right.

A (17:30)

Okay, okay, Let me, let me get it ready. We'll see. So allow this app to request my geographic location. Sure. Let's riff. I think there's. Do you have a suggestion of like, how, how do we evolve this example in the world where it can't see my location? Do you have any suggestions? Cause we can just iterate on it.

B (17:46)

I think it's just, you know, like just finding like, you know, hikes or finding restaurants nearby or like, you know, finding the best, let's say find the best Asian restaurant in Richmond.

A (17:59)

Let me. I like it.

B (18:03)

And did you have to set up the Google Maps API yourself or it just works.

A (18:08)

That's the beauty. It just works.

B (18:10)

That's great, dude.

A (18:11)

Yeah, yeah. So we have. And part of the context of why it works is because we have this. So I don't know. I don't live in the Richmond district, but if somebody does and Fuwa dumplings or dub or Bao Bao dumplings are good, let us know in the comments. This is actually an inaccurate suggestion, but yeah, the reason that this works out of the box is because in the actual Gemini API we have this grounding with Google Search tool. Grounding with Google Search. We also have grounding with Google Maps and so you can pull in a bunch of Maps data without actually needing to like separately set up Maps or separately set up the Google Search API. It's all just like baked in as a, as a hosted tool, which is really cool.

B (18:50)

That's actually a big deal, man. I feel like a lot of new vibe coders, like just a set apart is like the huge, the biggest pain ass, you know?

A (18:58)

Yeah, yeah. And we're thinking about like, how do we do this for more Google services where there's interesting data that folks might want to build on top of.

B (19:04)

Yeah.

A (19:05)

So it's definitely a work in, a work in progress. But search and Maps works out of the box in AI Studio, which is really cool.

B (19:12)

That's awesome.

A (19:13)

Yeah. And then you get citations too. So if you want to go and like look at the menu and be like, do I really want to go to Baobao Dumplings? You can click on, on it and hopefully see like what the menu looks like and all that stuff.

B (19:22)

Got it.

A (19:23)

Yeah.

B (19:23)

Cool.

A (19:24)

We were looking at building something around this floor plan. So let's go and let's go kick that off. And what was the idea was taken. So I have the floor plan image that Nano Banana generated for me.

B (19:38)

The idea was to make it either. Either get them to generate a 3D image or maybe even have it like build like some sort of 3D. Like I don't know, using 3Js or something to build some sort of 3D. Yeah, I like it.

A (19:50)

Okay, okay, let me see. I'll yap to app and we'll try to make it happen. All right. I have this image of a floor plan. I want to build an app where I can upload a floor plan like the one that's in this image. And I want it to then take that image and turn it into an interactive floor plan. Maybe we have a couple of options for that. Like we can use for one, let's use 3js 3js to visualize that floor plan in 3D. But I'm also open to like an open Anna or like some other interactive way of sort of looking at the floor plan and remixing it or something like that. So come up with a few ideas of how I can bring this floor plan to life. Build a bunch of features around it and let me know when it's done. We'll see what this one comes up with and I'll make just a couple of other random comments like if you're, if you're a developer and actually, I think one of the use cases folks like the most right now is they start in AI Studio, they build something, they then go to Google's Anti Gravity and continue to develop whatever they build. Because it's a little bit richer of a development environment. We're working on a way to let people port over that stuff automatically. But you can download your code, you can save it to GitHub, whatever you want to do. And again, you can also share the app. And that's where you and me or you and a team could go in and iterate and fork and remix, which is a ton of fun.

B (21:10)

Yeah, that's awesome. Yeah.

A (21:13)

One of the other things and we should do this with this one is we built this annotate feature which folks have really resonated with. Maybe while this is building, I can show it on this example. So we'll do something like, I'll look at this example and I'll say I'll open up annotate mode and let's see. I hate this recently viewed section. So I'm going to say I would like to get rid of this. I'll just add My comment, I'll say, I really don't like this. Can we make it collapsible by default? And then I'll say, add to chat. It should pop over. You can just send the prompt, and it just says the prompt is apply the edit shown in the screenshot. And it literally just takes it. And hopefully we'll get an updated version of that. And our other app is cooking at the moment as well, so. Yeah.

B (22:03)

Oh, so it's just took the screenshot of the thing with your annotation on it, right?

A (22:08)

Exactly. Yeah.

B (22:09)

Got it.

A (22:10)

Yeah. And it works like a charm. This is like the multimodal. As we. As we sort of build AI Studio, one of the things that is, like, always top of mind is what is, like, where can we bet on the model, basically? And, like, this is a great example where, like, there's probably ways that you could build this that, like, don't bet on the model. But the multimodal understanding is so good with gem. Actually, we. I just posted some blog posts in the last couple of days that shows, like, the. The benchmark chart was just Gemini 3 Pro, State of the art on basically, like, every multimodal benchmark that exists, which is really co. Yeah. And like, you. You sort of feel that come to life in a bunch of these examples, which is. Which is really cool. So we're constantly pushing.

B (22:50)

Yeah. This is why I'm buying Google stock, man. Like, the first reason is because you and Josh work there. That's the first reason. But second reason. Yeah, I feel like the multimodal stuff for Gemini is just, like, unmatched. Like, it's like. And also, like, if you think about, like, social media, like, you know, Instagram is more popular than Twitter.

A (23:04)

Twitter.

B (23:04)

So, like, everything's going to be multimodal eventually. That's going to be the predominant format.

A (23:09)

Right? Yeah, it is interesting. I think there's. Everything's gonna be multimodal and everything's gonna be code, which I think is also fascinating. Like, I think the code as this sort of main scaffolding I think will be interesting. Okay, so this was a small feature, but I like this way more, to be honest. I'm gonna send this to our team after this because I hate every time I open up a studio, I hate seeing all of my, like, recently, it's just clutter. I'm like, I never click on any of them anyway, so I'd much rather have this collapse. So, Amar, if you're watching this, we need things collapsible by defau, so. All right. We got that. And let's go back and look at the floor plan. So we ran for 159 seconds. There was 42 errors. And thank goodness for Gemini. It thought for 38 seconds and hopefully fixed all 42 errors. So we have this. We have a little bit of an agent loop that sort of does some of this. But I'm going to put in this floor plan, and we'll see what it came up with. We kind of let Gemini take the reins on this one. So I don't know. Yeah, I don't know, but I like this. I like this name. Architectural intelligence. That is interesting.

B (24:10)

Dude, wouldn't it be great if, you know, because we're both PMs, wouldn't it be great if our job was just to just generate all these prototypes all day?

A (24:19)

You need to become a prompt engineer. That's what it is, Peter.

B (24:22)

That'd be nice. No more meetings, just prototyping.

A (24:25)

No more meanings. Exactly. All right, let's go back to this image. So we have a kitchen, primary bed, walk in closet, bathroom, garage, and another bedroom. Okay, let's look at the three schematic. And does this seem correct? Walk in closet. I, like, need them side by side. It kind of seems like it's. Oh, okay.

B (24:41)

Oh, okay. There we go. Okay. Yeah.

A (24:43)

Two car garage. Let me. Yeah, that seems directionally.

B (24:46)

Okay.

A (24:47)

Two car garage is there. Let me see. Two car garages there. Let me switch it. Wow, this is kind of fun.

B (24:54)

This is probably the 3js version.

A (24:56)

I'm guessing this is definitely the 3js version, but it looks roughly. Right. Garage. Kitchen stuff. Yeah, garage. I think it has it. The garage in this flipped. Or maybe I was looking at it flipped, but it seems reasonable. And then we have remix. So this will now take in that image that I said, and we can use nanobana Pro to actually change the image in some way. So we'll say maybe make it modern minimalist, and we'll see. We'll see what it does. This one, it looks like it's going to use Nano Banana by default and not Nano Banana Pro. So, yeah, it might not be as good. We might need pro to.

B (25:29)

No, that's pretty good. Yeah. Yeah.

A (25:31)

I just, like, animated it, which is interesting. Switch it to a blueprint schematic and see what it looks like. I think it's so the. The thing. Every time I do this, I'm like, I should just go and make businesses all day because I feel like this is like 50% of what people actually want, which is so funny. Like, you're you're actually like really close to actually working product, which is really interesting. It looks like it's actually doing some room composition. The primary suite seems interesting. 10 rooms. 1, 2, 3, 4, 5.

B (26:01)

Yeah, I totally agree. I think there's like so many businesses that can just build off Nano Banana Pro, like, because I don't think people want to prompt the stuff. They just want to like click a button and have it generate what they want. So yeah, there's so many businesses.

A (26:12)

Yeah, this one is awesome. I think there's like a lot of room design stuff that I think feels. Feels supernatural. That. Yeah, there, there's, there's. Yeah, there's so many cool companies to build.

B (26:24)

Do you have any, like, quick tips for prompting Nano Banana Pro? Because, like, sometimes it doesn't, you know, like, do you have any tips for that? Give it like an example or something or.

A (26:33)

Yeah, yeah, that's a good question. We actually have this Nano Banana prompting guide which we can put in the show notes potentially. Or folks, if you go online, just search up nanobanana Pro prompting guide. We, our team spend a bunch of time on this. We're also trying to like, part of one of the fun things about launching these models is like, you sort of learn a lot from customers and the creativity of what people are doing. So we also have this Nano Banana account on X where we're sort of like daily posting new examples of like, cool things and the prompts that get you to generate those cool things. Yeah, so we're sort of. We're figuring a lot of these like, new use cases out in real time. Like the weather one. I don't know if folks saw this, but like, there's a way that, that folks were taking in like weather data and building this like, really rich visualization of the city that you're in and the weather forecast and all this stuff. And it's like taking something that's really boring that like you regularly look at the weather and bringing it to life in a really exciting and fun way. I think is. Was super cool to see.

B (27:30)

Yeah. And this stuff can go like super viral and help grow Gemini and Nano Banana. So yeah, for sure. It's great.

A (27:37)

I mean, it really is already happening like that. I think there's, there's. Let's see if there's any other good examples. Yeah, the infographics I think is another one that folks are really interested in. This like, product mockup composition, I think is another one that we've seen a ton of use cases like people want to visualize product ideas and merge their brand with some physical thing. So it was really cool to see. Yeah, folks from our team and others built these, built some of these demos.

B (28:04)

I did. Well, I want to ask you some questions about just like working at Google now. So I guess my first question is like, you know, a couple of years ago, was it maybe like one or two years ago, Google had barred and people were making fun of Google, right. It's like, this stuff is not good. And then maybe it's because you joined the company. I don't know what happened. But now you guys are firing out all senders and what do you think? Has there been like a culture shift internally to actually prior to ship, shipping or what's your observation internally?

A (28:33)

Yeah, that's a good question. I think some of this is just like the part of the advantage that startups have is you're able to sort of like adapt to the trend quickly. And like, that is like a, that's a foundational advantage of being a startup. And I think if you look at like, what are the foundational advantages of being Google, I wouldn't describe one of them as like, you know, we can move on a dime as a company in order to like, adopt some new trend. And I think if I look back like two years ago, you know, putting the right pieces in place and getting the right shots on goal, I do think was important. But another thing is like, it just takes time. And this has been one of my realizations in the last, in the last six months is like, it really does take a long time to build a great product and to build great models. It takes a long time. And to build a team that sort of is rowing together in a sort of coherent direction takes time. So it is, I think people have high expectations for Google, which is why I think they hold, they hold us to a high standard. But I think some of it is it just takes time to do a lot of these things. I also think there's like, lots of structural things that have happened. One of them is bringing the DeepMind team and the Google Brain team and the folks from Google Research together into a single organization, Google DeepMind. And actually principally also having products in that portfolio as well, historically. DeepMind. And this is one of the things that gets me excited about the direction that we're going with AI Studio, the Gemini API, the Gemini app, et cetera, is that it's like actually inside of DeepMind. It's not like someone's throwing some model over the wall to us and we're building A product around it. It's like we're co building the model with the research teams to make it really great for all these use cases that our customers, developers, builders care about. And that's such a unique advantage that I think Google has with these products, which gets me excited.

B (30:30)

That's awesome. And you being kind of at the forefront of engaging customers on Twitter and other platforms and, and tell me about these like, feedback loops internally and externally. Internally, like you work with the researchers every day, you kind of feed them information and like, externally, you know, have you encouraged other PMs to talk to customers more or.

A (30:47)

Yeah, for sure. I think this is one of the. And so this, you know, I talk to internal teams about this sometimes as well. This is not without challenge. Like, it's. There are, there are very real downsides. We don't need to spend the time talking about the downsides. We'll talk about all the positives on this podcast. But I really do think as a, as a pm, it's like the most competitive advantage differentiator that I could possibly have. Like, I think normally you sort of have to fight to like, get feedback from users and in my job, like, and this is, I think a privilege of Google and you know, thankfully folks on the Internet care about what we're building. Like, I get so much good feedback and like, seeing what the ecosystem is saying and seeing what people, you know, think about AI Studio and Gemini and our models, like, very genuinely makes my job easier. It would be harder if I had to sort of figure out like, I wonder what people think about Gemini or I wonder what they think about AI Studio or I wonder what bugs they're running into. Like having to search for that data and that information would take a lot of cycles and time and I don't have that problem right now, which is really nice. So, yeah, that flywheel spins and I think more and more of our team and actually more and more folks across Google and you actually see like Demis and Sundar and others doing this as well. Like they're showing up for customers, they're doing it in public, they're fixing these problems, which I think is like a great trend line.

B (32:05)

Yeah, that's the way to do that, man. People don't want to hear from some corporate brand account. They want like the person and talk to the person.

A (32:11)

I 100 agree. Yeah, yeah. I think the brand accounts are actually really useful in telling the story. But like, my personal, my personal opinion and I've sort of, I've given this comment a lot to Folks is like, I don't think the brand account should ever respond to people on the Internet. I think it just feels kind of weird if you have real humans, like, have real humans talk to real humans. Like that's what I want as a user. I'm sure there's cases where that doesn't scale and doesn't make sense, but, but I do think it's sort of the golden path if you can do it.

B (32:38)

Okay, and another question internally, has Josh or other folks really empowered you guys to just ship?

A (32:45)

Right.

B (32:45)

Because I think in these big companies there's a lot of approvals and check all the boxes and do all the stuff. You got to do two year planning and blah, blah. But shipping is learning, man. Has the culture changed where you can just ship stuff?

A (33:01)

To a certain extent we do ship stuff. So I think the guardrails around shipping depends on what you're trying to ship. So like the like the core product velocity is exceptionally fast models. There's like a bunch of like legal and regulatory infrastructure around like shipping new models. So they're actually there. It is like slightly more heavyweight in that sense.

B (33:21)

Yeah.

A (33:22)

And we've actually done a lot to like make that streamlined and sort of, you know, do the, do the right diligence but also be able to get models out the door. And then there's very specific features that do require like you need to, you know, go and meet with security and privacy like that to understand. But by and large I think like the default is we ship stuff, which is really fun. I think it's not just like, I think Josh has done a great job of this across all of his teams. But I think this comes from like Demis and Sundar and this like at the, at the top level, this sense of urgency of like how we need to be showing up and moving quickly. And my personal worldview is like, I don't even know how you would do this if you weren't moving quickly. Like I don't think it doesn't feel like it's a, like I don't wake up being like, ah, we gotta ship fast today. It is just that there's one, there's one mode and the mode is we ship fast. There is no other option. There is no other alternative to what we're doing. Like it is literally the only path. And that's been my mindset since I joined Google. And I think you need the right pieces in place to actually be able to materialize that, that sort of mentality. But I think we have those pieces now. Which is really, really fun.

B (34:32)

That's great. Yeah, that's great. Yeah, it's definitely a very competitive environment. So you have to show up.

A (34:37)

Yeah, a hundred percent. A hundred percent. And it's like a, it's an environment where there's so much opportunity. Like, the models are so powerful and people are building so much interesting stuff that, like, I think the competition piece aside, like, it's exciting. Like, I want to ship quickly because our customers are like pushing the frontier of what's possible with AI and like, we want to be there along with them. So from that perspective, it's. It keeps me excited.

B (35:01)

Yeah. I feel like a lot of the stuff you ship can just be like, what are my most power users doing? And how do we make it easier for the, you know, beginners and people who are new to the product? But that's like one avenue I found that always works.

A (35:13)

So. Yeah. Yeah, I love that we're thinking a lot about this for the vibe coding experience right now because we have, you know, there's definitely like a barbell of who's vibe coding, like in a lot of kids. Like, I'm a. I was trained as a professional software engineer. Like, I was a software engineer. Yeah. But like, I don't actively, like write a bunch of code anymore. And then you have on the other spectrum, like people who actually are full time software engineers doing this still. And then you even have the folks who like, have never opened up code, don't know anything about coding. They just have an idea of what they want to build. Like, how do you build a product for all those people? Is like such a, such an interesting, challenging space. Yeah.

B (35:48)

Yeah. You got to figure out where to do different levels of exposure and simplify stuff. Okay, cool. So last question. So, you know, Gemini seems very exciting. Working for Josh seems very exciting. If people want to join your team, what kind of traits do you look for for hires? Yeah, like 10 years of fan experience or. No, what do you look for?

A (36:06)

Yeah, that's a great question. Yeah, there's definitely a couple of things like, and this is not an exhaustive or an exclusive list, but like, I love folks who are building in public. It actually makes it easier to just like assess work quality because part of what I enjoy and like developer relations is a good example of this. Like, it's actually, I know people talk about like, it's hard to hire devrel people and sometimes it's easy because you have this like exhaustive proof of work of like, what are the, like what are the outputs, the types of things they do, the quality at which they do it. So I know that there's some jobs in some companies where that type of stuff's not possible. So that's, that's really helpful. We, for everyone we interview and hire on the product side and also on Devrel and other functions, we do this exhaustive like friction logging exercise. And to me it's one of the most helpful things with like, what is your product taste and like how do you see problems that users have or you yourself have and like what is your principally not just like what is the thing that you identify as a problem, but like what is your, your sort of like suggestion and take on that. And like I almost, you know, we don't do this but like I almost, I almost in some sense wish that I could just use those things as like the interview decision making criteria because like it really is so helpful to see in a written format like what people think and you know, how they identify problems and all that stuff. So it's been incredibly helpful and it's something that I love to read.

B (37:36)

Interesting. Yeah, I think that's like Stripe does that too. Like just kind of detail, like what's your frustrations using AI Studio and like what you're going to do about it. Right, that kind of thing.

A (37:45)

Yeah, yeah. So if folks have these things, please send them to me. I would love to read through. It's super helpful to see people's product opinions.

B (37:52)

Yeah.

A (37:53)

And I think the last one is this like agency bit and I think it's hard to. That's one of the things that I found most difficult to like assess in a hiring process. But I want to work with people who are high agency and they're just like going and solving problems proactively like that. That is like my ideal outcome, even if some of the stuff is wrong, to be honest. Cause like I can, I can fix, I can, I can sort of fix that problem. It's hard to sort of instill agency in people. So.

B (38:19)

Yeah, yeah, people are so afraid of being ramen. Like PMs are still wanting to make the right choices each time, but if you move fast and like, you know, like, if you make like nine wrong decisions, but the ultimate one is right. Instead of someone like, you know, trying to plan for like, you know, three months to make the right decision, the first person is going to win it, man. So, so it's like it's totally fine to be wrong as long as you fix it.

A (38:41)

I love that there's a good, there's a good Sam Altman quote that I used to have up on my wall about sort of the speed of iteration and, like, the cost of being wrong. And I think it always ends up. It always ends up being true.

B (38:52)

Cool. All right, dude. Well, I think people know where to find you, so I think. Yeah, yeah.

A (38:57)

You're not X. Yeah, I'm on the Internet, so if you. If you need anything, if. If our team can be helpful, please reach out. Always happy to help folks building on Gemini or just have comments or feedback. And, Peter, thanks for having me on. Thanks for vibe, coding and building a bunch of stuff.

B (39:12)

Yeah, I love Gemini. Thanks for building a great product.

A (39:15)

Yeah, of course.

B (39:16)

Cheers.

A (39:16)

Thank you.