EP 501: Google's Logan Kilpatrick: Gemini AI updates that create new possibilities live from Google Cloud Next
17:11 Share

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life.

There's so much new that was just announced at Google Cloud Next. I'm having a hard time wrapping my head around it. It seems like there was dozens of new AI updates. So I said, what better than to bring in one of their leaders to help us make sense of it all. So we're going to talk a little bit today about what's new inside Google Gemini, Google Gemini AI Studio, like

everything with Logan Kilpatrick, the Senior Product Manager at Google DeepMind. Logan, thank you for joining us a second time. Yeah, round two. This is going to be, I don't even remember what we were talking about for round one. It feels like it was super recently, but...

There's a ton of new stuff to talk about, so I'm happy to be back. Yeah, absolutely. So, you know, top to bottom, I mean, we saw new updates with Google Gemini 2.5 Pro being rolled out in other places, a new model in Google Gemini 2.5 Flash, right? So many things for developers, but where do you start? Or maybe like, what are you most excited about that was just announced here at Google Cloud Next?

Yeah, that's a great question. So I've been continually excited about 2.5 Pro. I think like we're seeing 2.5 Pro sort of rolling out across our developer products, our consumer products. It just landed in deep research yesterday, which like folks have been super jazzed about. I think if you're an advanced Gemini advanced user, you get like 20 deep research queries.

you know, our customers prefer the sort of Gemini Advanced with 2.5 Pro sort of two to one versus sort of other products in the market, which I think is just sort of a nice proxy of like,

actually this model unlocks new stuff from a deep research perspective, from a Canvas perspective, getting to see if folks haven't tried Canvas yet in the Gemini app, being able to sort of vibe code and agentically sort of write code for you without having to be a developer is like such a cool and special experience. So that's what I've historically been most excited about. I think today now at Cloud Next, we're, you know, tons of new stuff launched. Vio is available for developers.

which if folks haven't seen is our sort of state-of-the-art video generation model, which has been awesome. We just announced the live API, which I think a lot of folks, and actually like in parallel to that, the live mode is rolling out to some customers and Android, I think as well. So like there's everything happening. Like one of the things that I've been most happy about is it feels like more and more we're getting to the place where as these new capabilities come online, they end up sort of ubiquitously across the Google ecosystem, which is really cool because like, you know, some people are Gemini users,

you know, user. Some people are a Google AI studio user. We've got enterprise users. We've got people in search. And like, I think it's awesome to sort of get to a place where new thing launches available everywhere for, for the world to use. So I want to quickly dive into two of those things that you mentioned there. So the deep research, I,

I've been blown away, not just by, you know, I think there was a couple times you guys updated it first to 2.0, and now I think to 2.5 Pro. And yeah, you talked about some of the benchmarks that came out in terms of, you know, the preference, I'll say it, it's against open AIs, right? Which I thought was a great, you know, a great offering. But, you know, now seemingly your guys' is

way, way better. Uh, you know, what are you even using the deep research tools for? Like, I, I love asking the people that build it, like, what are you using it for? Cause I think people can learn from what you're using it for. Yeah, that's a great question. I think some of the stuff, um, like not, uh,

the two use cases that have been top of mind for me, one, I was looking up like what the general sentiment is about MCP. If folks haven't been following, there's this, we won't dive into MCP thread in this conversation, but if you haven't done MCP before, haven't looked into it, use deep research. It gives an actually like pretty robust answer and like gave me a bunch of supporting materials on like not just how people on Twitter are thinking about MCP, but like it's an agentic

way of interacting with tools. So that was one of the use cases because I was just very intrigued to know what happens when you do that. The other one is I've been doing a bunch of competitive analysis of just as we think about how we're showing up at the market, what do we look like comparatively against other providers? And this is a really interesting, I think for me, the deep research

that conduit has been really interesting because what deep research is actually able to capture is sort of like the information that's available on the internet. And I think it's like, you know, could I go and talk to customers and like get this perspective? Yes, that's actually a really interesting and useful perspective, but it's actually also interesting to capture like what is sort of the codified perspective on the internet of like how people think about, you know, the Gemini API or ASGD or stuff like that. So it's been really interesting just to like

have that experience. And like it, it actually like diverges from in some interesting ways, like what what people tell me in person about like how they think of the product is used and all this stuff. So really interesting if folks haven't done that exercise. If you have like a product that you build, or you have like a favorite thing, like just like ask

deep research functionality and the Gemini app to put together reports and see how that differs from your point of view or from your perspective. Yeah. And another thing you just talked about there is kind of like, you know, vibe coding in Canvas. So Canvas has been out what, like two weeks? Two weeks. Something like that, right? Like I use it so much already for a tool that's only been out two weeks, but maybe walk people through some of the, you know, practical applications. I think a lot of people are, you know, like, oh, like let's

you know, create a game and like, that's fun to get started. But, you know, in terms of, you know, business utility, what are you all seeing as, as some of the more impressive or useful applications for the new canvas mode inside Gemini? Yeah, I still think we're in the era of,

And I think like chatbots in general were in that place for a long time. And I think that sort of just as in the last like six to eight months, like gotten out of just like being sort of a novelty item. And I think like Canvas is sort of still in that realm where it's like, it can do interesting things to your point, like building games from scratch. Like it would take me a long time to program a game from scratch right now. You know, AI being able to do that is awesome. Yeah.

But like, where does the practical business value come in? I think for a lot of people, the practical business value comes in like when you're connecting this thing to your company's data and like all, and like, that's the kind of stuff that doesn't exist yet today, at least in the sort of canvas environment that we have. And that's what I'm most excited about because I think like ultimately for these tools to be useful, like you need to connect a bunch of your stuff to them and sort of let them, you know, have access to your email and then I can sort of build a tool around my email to do it. So I'm really excited about that. And I think,

From my, like, and I'm not a product manager on the Gemini app, but I'm a consumer of the Gemini app, and I love it, and I think it's a great product. One of the things that I'm most excited about is, like, this trend of the Gemini app sort of becoming this,

AI interface and this AI conduit to like all of the things that are happening inside of the Google ecosystem. And like the sort of pertinent example of this is the Gemini app also, in addition to all the other Canvas stuff and deep research and everything else going on, it has a personalization feature. And the personalization feature is actually built based on your Google search history. So you can opt in to be like, hey, you know, basically personalize the answers that the model's giving based on Google search. And like, that sounds very like

uninteresting at the surface level, but it starts to get to a world where like AI is this interface to like connect to this like vast set of data. And I think about this for myself in the work context, in the personal context, like I'm on YouTube all the time. I'm in Gmail all the time. I'm searching stuff all the time. I'm in docs all the time. So like, it's really wonderful to be able to sort of bring all that experience together. And I think Canvas is like

the first step of that with Docs specifically with Cone now. So I'm super excited. Yeah. And, you know, if you haven't had the time to use Canvas yet, I highly recommend it, right? It's literally being able to, you know, run and render code. You don't have to be able to even know coding. It's so easy. So another thing, Logan, there that you mentioned is

VO2 and some of the new capabilities in that that are available in Vertex as well, adjusting camera angles, right? What does this do for creatives, right? There are so many new things I wasn't even expecting that were announced today. The text to music, right? The updated chirp.

Like what is this going to do for creatives and how does this all unlock, you know, both in Vertex and AI Studio? Yeah, I think the general trend that gets me excited is like, and I was just having a conversation actually with the folks in the Vertex team about this and they sort of agree, which is like,

this general up leveling of people to be able to like go to the next level. Like I'm not a creative, like I'm also not a game designer. Like in the game design use case, I couldn't build a video game. I've tried before. It's horribly difficult. It's not, it's not fun to sort of bash your head against the wall trying to do that. I think there's a lot of cases where that's true in the video use case. Like you and me, like you, we were talking off camera, like,

Editing video is tough and there's a lot of great tools out there that help do it, but it's still kind of a pain in a lot of cases. And to be able to have all these AI tools start to take those steps and up-level the people who are really excited and take out the stuff that I'm not interested in doing, I'm super excited. I think Vio specifically has been the one that

folks have been like losing their mind for a long time. And this is actually like today is the first time that this, with the exception of YouTube, where it's like set up at a very specific product experience. It's the first time that like raw model is like generally available to the world to actually get their hands on, which feels like a crazy, uh, I don't think it's been as crazy of like a,

public moment yet as I think it actually is in reality. But like the world's best video generation model is like now available for people to actually use and start building. So I think we're going to see the technology start showing up in lots of new, interesting ways. Yeah. And it was impressive. And we'll share in the newsletter today the demo that they did. I'm sure that Google's going to be posting that online. Right. But being able to, you know, kind of do the live shots of Las Vegas and animate them and put them to music. Super impressive. Something another really

new update here. Firebase, is that what it is? Did Google just release an IDE out of nowhere? Tell us what Firebase is. How does it work? Yeah, this is a great question. So this is some slight developer context. So if you're not a developer, some of this stuff might not

It might not be relevant or it might not make too much sense. But so the original incarnation of that product, which today became Firebase Studio, was something called Project IDX, which we announced last year at Google I.O. And the intent of Project IDX was like, how can we build a next generation IDE, integrated development environment, for developers to actually use...

which I think was the unique, like today developers like download a local IDE onto their computer and they do their development locally. This was bringing the IDE to the developer, to the, to the,

to the browser. And sort of the next iteration of that product suite, and this was being created by the Firebase team, which is why it ended up as Firebase Studio. The next iteration of that product is how do you actually not just like do the basic developer environment, but how do you infuse AI into that? And how do you sort of help developers bootstrap actually going and creating apps and products and stuff like that? So I'm super excited for Firebase Studio. I think it's like the, for folks who aren't close to Firebase, like Firebase has a lot of like

like StreetCred as being like an incredibly like developer-centric team and product surface. So I think I haven't spent a bunch of time with, I spent a bunch of time with IDX. I haven't spent a bunch of time with Firebase Studio yet, but I have full conviction that that team is going to knock it out of the park. And hopefully we'll see like more of these tools that enable folks who aren't developers actually to like start coming in and working

building stuff like they couldn't before. Yeah. You know, one thing, just getting back to, you know, 2.5 Pro, I like, I think it's worth gushing about it a little bit. And I love that in the keynotes, you know, it was mentioned that the LM Arena and, you know, I think it came in with like a 39 point lead over the second models when it was released. How good is

is Gemini 2.5 Pro. It's mind-boggling to me. When I use it inside AI Studio, I feel like I'm stealing something because it's so good. It can handle so much data. And it's free. And it's free inside AI Studio. Talk about maybe some of the best use cases that you're seeing for Gemini 2.5 Pro. Yeah, that's a great example. And I actually think one of the interesting things, and I had this conversation with some of the folks on the DeepMind team, is like,

sometimes actually you see like a 40 point jump on some benchmark somewhere. And like, it actually doesn't even tell the story to the completeness of like just how much better it is. There's also this like other, and then I'll answer your question directly. There's also this other thread, which is like every time a new model comes, there's like an entire class of new companies that weren't possible before that, like just become possible. And like, it feels like that's true when you get this like,

massive jump in capability. I think 2.5 Pro is actually one of those models where like, there's a bunch of new companies now that are possible. I think there's a lot of coding stuff. It is interesting that it's one of the things that makes me most excited is that as you see these like general purpose frontier models, like take a step function change in capability, it's like,

across every use case. So like you, like, I think the one that's like, didn't work really well before that now works really well is coding. So like lots of people are like very excited about the model's ability to do code, but like, I've seen tons of creative writing examples. I've seen tons of people using 2.5 pro is like a harness to build agentic products, which is a little bit like in the weeds behind the scenes. Um,

Yeah. So, and I think this actually, we haven't even gotten to like a bunch of the yet, yet to be released, like a bunch of the multimodal stuff that I think we're seeing with 2.0 flash, which was another thing. I think that happens since the last time we caught up. There's too much stuff going on. It's hard. It is certainly hard to keep up.

with. All right. So it is hard to keep up. You know, Logan, I know you're a busy guy. You have to go speak to, you know, thousands of people. But, you know, as we wrap up today's expedited conversation, because, you know, maybe we'll have to get you on a third time. But, you know, what are some of the, you know, even speaking of kind of like a new class of companies, right, which is great with like a great way to think of it with Gemini 2.5 Pro. But, you know, what are you most excited about from this weekend and, you know, or

maybe for the average, you know, everyday business leader, what are you most excited for them to get their hands on? And how do you think, you know, kind of like there's, oh, a new class of companies now, is there going to be a new class or a new way that we do our everyday work because of what was announced here?

I think the live API is that, and we hadn't talked about it yet, but the live API is basically this, and I don't remember if we did demos. I don't think so. But if folks haven't tried this out, AISTudio.google.com slash live has this experience where you can come in and you can talk to the model, you can share your screen, the model can actually look at your camera if you give it permission. And it creates this really, what I think is this future of how people are going to work, which is the models can actually...

see the stuff that you see, which I think unlocks like it takes the drudgery out of having to use AI tools, which like my personal perspective is today,

the challenge with using AI is that you as the user of the AI product have to go and do a bunch of work to bring all the context to the model. And like, oftentimes, like for me as the person who wants to use AI, I'm like, the context is already there. I'm looking at it on the screen. Like, why is this so much work to like take that information and go bring it over to whatever AI product? And it's like a very simple thing that all of these new, back to the thread of like new classes of companies to be built, all of these new companies and products to be built

You just flip that switch and then all of a sudden, like, you know, whatever the random product is that you're using can see your screen and like help you reason through whatever the problem is that you're trying to solve. It can, you know, bring in real information in real time from Google search. It can execute code on your behalf, like all of this, like really, really interesting stuff that I don't, I don't actually think we've seen products built yet with this technology, which gets me excited because I think it's, it's going to be wicked.

All right. It's an exciting one. A fast but furious interview, just like what we've seen so far out of this conference. Fast and furious updates. So, Logan, thank you so much for taking time out of your day to join Everyday AI. We really appreciate it. Yeah. And for folks who aren't watching on video, Jordan has a sweet Everyday AI Nike shirt, which looks awesome. You're crushing it.

Now I'm going to be just drowning out with requests for it. All right. Well, hey, thanks again, Logan. And if you want more, we talked about a lot in a very short amount of time. It's all going to be in the newsletter. So if you haven't already, please go to youreverydayai.com. Sign up for the free daily newsletter. Let me know. Should we bring Logan on for the third time, the first person ever after 500 episodes to be on a third time? All right. Thanks for tuning in. We'll see you back tomorrow and every day for more Everyday AI. Thanks, y'all.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.