Jon Kelly in conversation with Philipp Schindler

Yeah, thanks for having me.

There is no doubt these are big numbers. I can't really talk about the other players in the industry. I can talk about our numbers quite a bit. I'm very positive we can return on that invested capital for many different reasons. So first of all, as you all, we talked about this. We're taking this full stack approach. And that is really pretty unique in the industry. You look at everything from the models to the chips to the safety and security layers to so many others. We really operate full stack, and that's a very, very good position to be in, because all those elements are intertwined in a very interesting way, and you can make each of them better by also controlling the other parts of that stack. So that's a really, really nice position to be in. That's number one. The other thing to think about is we run this stack as a horizontal across the whole company. And then on top of this horizontal you have all those different verticals. And the verticals are on one hand, of course, our Cloud business, which I think we publicly said is roughly 50% of that investment. And a lot of this benefits, frankly, all the companies in the world that we're trying to help the transition into the AI world, many here in the room, because obviously they can create a positive ROI back on this, and we're actually kind of proud that we're taking our best technology. And in many cases, frankly, we're handing some of our best technology to other companies, frankly, before we can even deploy it at scale in our own consumer business for many different reasons. So in the Cloud business, you always get the cutting edge of our technology. So that's one way to return some of those capital. And then on the other hand, we have this amazing consumer business where we have, and you heard some of the numbers. Again, you can infuse AI in so many interesting ways across those products that it truly generates value from a consumer perspective. And then, of course, as we talked about all day today, you have the ability to then take this full stack approach and frankly, integrate it into our ad stack across every single piece of the marketing value chain and make it so much better. And I'm not going to repeat everything you heard throughout the day, but this is really one of these areas where this whole horizontal investment then truly, directly pays back in combination with a better consumer products across the entire advertising ecosystem. And I think this is a really, really positive cycle that we can create there. So I feel we're actually in a really good place. How it will play out across the industry in the end, it's very hard to say. My prediction is it's going to be competitive for a long time. In that sense, competition is a good thing because frankly, consumers benefit from it. Other businesses benefit from it. We all ourselves benefit from it because it keeps us on our toes. It keeps development speed high. It keeps everybody like hustling and trying to build better products. And so I feel, yeah, look, we're probably one of the most intense, competitive times of our lives. And that's probably how it's supposed to be. It's a good thing.

Yeah look, it's a great question. In a weird way, we're probably in the same boat as so many of you. We have to basically be on customer 0, and we have to apply the technology that we're developing. I literally want to know what you're doing. We have to truly apply it to ourselves. And look, we're 185,000 or something employees. We're not the smallest of all companies anymore. So that's quite a transition process we have to run through ourselves. And I'm obviously heavily involved. Maybe a few things I've learned along the way. The first one might seem a little counterintuitive, but I actually think it really helps to aggressively apply it in your personal life. This is an area I don't know. For some reason, I felt, I feel when you push the boundaries in your personal life, it's sometimes a safer space than to do it in the business world, but it really helps you in the transfer learning later. It's actually not that complicated. And I pushed the boundaries. We talked about a little bit backstage. You would not believe. I just came off a long 2 and 1/2 week around the world trip. I did multiple times back and forth around the world, really hard to manage from a jet lag perspective. And I figured out that actually the best jet lag advisor out there is Gemini. I vetted my travel schedule. I vetted all my meetings. I said to Gemini, tell me when to sleep, when to eat, when to work out. I said, here's the sleeping pills I like to take, but I really don't make me take them if possible. And it was completely magical. I said, please try to keep me in California time zone because I don't have time to adjust anyway, and it managed to hold process for 2 and 1/2 weeks for me. It told me literally every single step of the way what to do. At one point I said at o'clock AM in China, China time, I should go to the gym. That was a bit rough. I remember that one, but it knew that this was my biological time. I could do it right. So that's a very aggressive one. I use it on the border between my personal life, my business life. So for example, we do rehearsal for the DML keynote here and we record a test video and it's like, whatever, 15, 20 minute test video. And I feed the whole video into Gemini. And I literally say to Gemini, hey, you're the world's best speechwriter. Please analyze the video and tell me what I could be doing better. And then it gives me like five pages of the most honest feedback you've ever received in your life. And let me tell you, I could barely read it because I was weeping all over it like it was soaked in my tears because nobody had ever dared to tell me, right. Hey, Gemini, do it more politely than that. So look, it's the magic of our models. The large context window in combination with the multimodality. It literally says, Philip, you're standing on stage in the first two minutes and you're always doing this. This looks weird. Stop doing this right. That's the level of detail it can now do and it went way beyond that. So truly impressive. And once you understand that, wow. It can analyze a 15 minute video like this. You can see how it can help you analyze a video based process or maybe a process you want to capture via video in your business environment. So I think this is a really interesting one. But truly on the business transformation, there's a couple of things you really need to get right. The first one is you absolutely need to lead by example. That is so important. So you now have to take the learnings from your personal life into your business life. I'm going as extreme as opening some of my employees here around. Notice I'm sometimes opening the videos. They have a presentation for me. I say, wait for a second. I feed the presentation into Gemini. I wait a few seconds. I have the full presentation analyzed. I say, let's discuss first what the AI is saying and let's see where Gemini is right. Then we don't need to spend our own mental tokens on it. And then let's ask ourselves where's actually Gemini may be wrong or where we can add value. Just to make the point. And it's actually working surprisingly, surprisingly well. So you have to be that customer 0 here. The next thing you have to do very aggressively, in my view, is you have to skate towards the future a little bit because transformation processes and companies take time. And so if somebody says to you, well, AI isn't at this point and can't do it yet, it does not matter because it will take you half a year to a year to maybe a year and a half anyway for any transformation process. So you have to be very aggressive and say, look, I think this problem will be solved in a year or in a year and a half. You still have a little hallucination problem. You have whatever context window problem, anything. It doesn't matter because most likely it will be solved. So skate towards where the future is. I remember I came out of our labs in I think it was in 2018, and I saw early versions of LLMs, transformer paper was written in 2017, 2018. I saw early stage chatbots that we could interact with, and I was so blown away that I went. I run as part of a few other things, I run all of our customer services. So I went to our customer service team and I said to them, this is unbelievable. The future of customer service is going to be chatbot AI based interactions in 2018, and I reorganized all of our customer service teams in 28 teams to be ready for it because you need to have a slightly different, more process oriented structure in order to do this. Well, were we a little bit early. Yes, we were a little bit early because I didn't quite see at the time that there were still some issues with hallucination rates and error rates and some toxicity and so on, which took a little bit longer to figure out. And then others launched products in the market. And then we came later when we thought we were in a better place. But did it really hurt the business that we were early in this transformation. Frankly, not really. And now we're like the second the models and the breakthrough for us came, frankly, with the Flash models, the precision of the early Flash models, I think was mostly 2.5 Flash, and now we have 3.5 Flash, which is amazing. Amazing, right. Or I can already see the teams. I know they love it already because we tested it. So you really can see the breakthroughs and whether you're a bit early or not, it's usually not a problem in this area. So that's a really interesting development. You have to pay attention to a few challenges here though. You need to be very keenly aware. So the first one is I talked about leading from the top, but you still have the diffusion problem into your organization. And look, the reality here is I mean, pick a number 20% of your people, maybe 15, maybe 25. They naturally gravitate towards this. And they adopt this at an incredible rate at an incredible speed. So that's really not your challenge. There's another percentage. Let's just call it x percent. They could be ready and they probably with the right leadership can be taken along on the journey. But then there is a percentage. Let's call it y percent. And that y percent will be hard to transition whatever you do. The real question is, how big is x and how big is actually y. And if y is too big, then we have a challenge. This goes back to what we discussed before about what's the net effect. And then we need to have, let's be Frank, a societal discussion about what to do with this and how to help those folks, because there will be a y percent and they will have a challenge in this transition, and they will probably be struggling in where the world is heading. That is just the reality. That's a big one. Other challenges we see when we're trying to apply it ourselves is AI doesn't really respect company structures. A lot of the things that AI is really great to optimize are actually horizontal processes. A lot of companies, as we yes, you have some horizontal teams that come along, but you have a lot of vertical structures. But that's really not the ideal playing ground. Think about it more as a process optimization. But you don't have the leadership teams that usually can control all these processes. So you have to cross cut across multiple different areas, let's say HR and finance. If you want to do a full deep exploration of whatever the life of an employee in a company. Imagine how many you want to do this with AI. It probably should be done with AI in the future. Imagine how many different teams you would need to touch and who would actually lead and run this transition process. So that's a pretty hard thing to figure out. There's a third element, which sounds really a little nerdy and boring, but you have to pay a lot of attention to get right. And it's actually mostly on us as a company to help you get it right, which is all the primitives, because you can have the smartest AI agents and harnesses and everything in the world. But if you log in your authentication, your audit trails, your access rights, your safety and security setups are not working the way you expect it. You can just let the agents go and work for yourself. So you run into roadblocks and it will just slow you down. And this is where frankly, I think a company of Google based on Workspace with a Gemini for enterprise and all the other offerings we're having now with anti-gravity and so on, can really help in figuring out how to truly manage those primitives to a point that they actually work. And then in the end, the last big remaining challenge in this is, and I spend a bit more time on it because it's such a deep question, how to run the company transformation is really the cultural piece. And how do you bring the culture along and how do you manage this. And it makes total sense. You lead from the top, you drive it hard, but then there is a resistance level on a cultural level. People are asking, hey, why should I invest so much time and energy into something that could actually take my job. And what I've realized is that this distinction, and I'm not quite sure who mentioned it the first time, it might have been Jensen from NVIDIA, right. This distinction between task and purpose is actually really, really helpful. AI is excellent at simplifying and automating the task, but this does not mean it goes really to the purpose of your job. I mean, look at some of the stuff we said and we presented today. Will it make the task, the complexity, all the workloads of what you have to do as a CMO easier. Absolutely. But is this really your job if you're a CMO, as an example. No, your job is to drive profitable company growth, to expand markets, to look, ideally even create a flywheel where some of the things you learn feed back into the next generation of product developments and so on and so on. So I think the separation between task and purpose is a really helpful way to think about where AI can play a huge role, and where humans can actually free up time to then spend so much more on what their true purpose in their job is. And I'm actually very confident that understanding this correctly can truly lead to an expansionary moment here.

Yeah, this is a very good point. Look, can I just address the elephant in the room. And this is something I've been thinking about frankly, for years when I saw the first early stage generative developments. What we don't want to see is, and excuse me, I have a German background. And Germans love those really big, complex, compound words, right. So I'd love to sometimes use them in English as well. And what we really don't want to see is a race to the bottom lowest common denominator price sniping type solution. That's it probably sounds horrible in English, but I just love it. So we really do not want to see this. There's always a small percentage of customers, by the way, who like this, but even from a customer experience perspective, this is not a great experience. Customers shopping journeys, for example, is one part and it could include a lot of other products, not just traditional retail products are so much more complex. And there's a certain part, and we did a lot of analysis on this. There's a certain part that customers actually don't like, and it's fantastic to automate and to simplify for AI. And then there's a whole set of those that customers actually really love that they really enjoy. And you probably don't want to take this away. You want to enhance this and make this better. And one of the reasons why we launched so many different products is because we had over the last few days, whether it's the universal commerce protocol, enhancements that we talked about, whether it's some of the other agentic points on the payment side and so on, is because we truly realize that this is a more complex journey, but we need to have help people along the way. And you just take a look at the agent payment protocol. I mean, look, what could possibly go wrong giving an agent right access to your credit card. So it probably it makes sense. Well, I mean, I don't know if you have me, a 13-year-old who likes to play video games. How it feels to have a semi-autonomous agent, like using your credit card without any checks and balances. So, so. But what could possibly go wrong. So you want to have probably some framework around it, some harness and guidance. So that's what we're creating some of those products. And I think they're going to be amazing. But overall like we have no interest. We want to really be respectful to our customers and partners. They have so much more than just the price offering. They have a service offering. They have really interesting like loyalty programs. There's so much more. We need to be totally tuned into the inventory situation. They have a brand that they represent. And that they want to represent, and we want to make sure that wherever we go in the future, we actually do it together with the ecosystem and be really respectful of all this work on the other side where it's just like the press, press, press, press, press, right. This cannot be the solution, in my view. And all of our product teams understand this and that's the direction we're taking.

I mean, look, there is a reason why we call it the expansionary moment for search. You remember the days of 10 blue links. I think those have been over for a while. And you've seen we talked about it a lot. I talk about it on the earnings calls a lot. The AI Mode and IO and the longer queries and the higher complexity queries and so much more. You can do now and then you have all the different modalities kicking in. You have something like remember the circle to search and you have the lens where suddenly think about it, look, wonderful blue sweater. Oh I like it. Yeah we're kind of color matching here, right. I mean, could I have described the sweater or you're wearing a really nice jacket there, right. Could I have described this jacket. Would have you even tried to describe this jacket a few years ago in search, like. Oh, this is what a blue sweater. Find me a blue sweater. No way. Oh, this is a jacket with something. No right now I think I'll give you the link afterwards. Yeah Well, so. So now I just take a picture. And have a super high probability of finding exactly this product. We have billions of billions of queries a month on Google Lens already. It's absolutely fascinating. And then you can see the expansion, of course, into video. And again, with the multimodal models, our ability to understand what's going on there. And so you have the video expansion. And of course, you can imagine and we talked about it at I/O as well. The new form factors like glasses. I mean truly smart glasses. Not some of the stuff you see out there, right. Think about true smart glasses with a full multimodal Gemini app integration. I mean, that is next generation. And so I'm very excited about it. And I'm not even talking about all the work we can do on translation, which I mentioned before, SMB expansion and so on. And then you look at what you can do on the let's take our other hero product on the YouTube side, right. I mean, what is going to happen to the world if you give every YouTube creator basically access to a full Hollywood Studio production capability and beyond at close to no marginal extra cost. I mean, imagine what it does to the creativity of the world. Imagine what this little kid at one. Whatever take this little kid in Indonesia that today, right. Cannot express himself or herself the way he or she wants. Imagine what they can do in the future with those tools and what you will see in terms of creativity and where a platform like YouTube will go just with those tools. And we started the heavy integration already a year ago. And you saw again the launches yesterday. So I think this is a really exciting development. So yeah, I think we're in it for quite a ride on those products.

I mean, look on the input side. We launched a few products today alone, ask advisor as a Studio and so on and so on. Yes, input will become a lot easier. It will take a lot of the workload off your plate. It's already doing it. So input. Yes, it will make a big difference. The magic will happen on the output side. There's just no doubt about it. If you take every element of the marketing value chain and you enhance it with AI and you make it so much better, it doesn't matter what it is. From the insights generation to the bidding to all the measurement pieces, you had Gaurav talk about it and so on. This is where the breakthrough will happen. And of course, the vision that at one point and I don't think we're too far away from this. You can actually take all the performance data, everything you have from your own insights generation process. And you can basically have the models. Think about the creative generation. Of course, you can feed in your brand guidelines. And whatever approval process you want, but having a complete like performance to creation loop that is working for you at super scale, potentially then customized over time to segments or maybe depending on how far you want to take it to every individual user is a vision that is not completely unimaginable anymore. We're coming actually pretty close to it. You can then combine it with large scale simulations where you run a lot of this first in simulations before you even take it to the market. So you can truly test whether those things are working right. So we can all imagine where this is going. And I think this is where the true difference kicks in. I personally don't think about ads really as advertising in the traditional sense anymore. I think about it as super value ad amazing commercial experiences that we will create for customers, for our partners out there. And yeah, I think those this is an exciting time for the whole industry.

I mean, look, in terms of final thoughts, I would say I've never seen anything like it. And I've been in tech for so long. That's how I started the thing. And I'm trying to bring this sentiment across a little bit. How wonderful, how exhilarating this moment is. We have to be cautious. We have to pay a lot of attention to things. But it is an absolutely unique moment in time. It is on an exponential. I often talk to my teams and say, yeah, if something is on an exponential and you're moving really fast, you could still be off the exponential pretty quickly. Again, that's the pretty simple math. You think you're moving fast. Well, right. The gap could widen pretty quickly. So this is an important one. I'm super excited where we can take our products. On the consumer side, I'm super excited obviously about our Cloud business. I extremely grateful for everybody here in the room and so many others watching for the ability to work closely with you because this is where true magic is being created. Frankly, it's not being created only in Google technology land. It is created on this magical intersection between your understanding of your business, your open and honest feedback, and our ability to build some pretty cool cutting edge technology and where. Actually, this is really where I think true innovation happens. And so thank you for your partnership on all of this. We've worked with you through many big and difficult moments. And again, there's difficult moments in this as well. I talked about the difficulty of some of the transitions into AI. If you remember, we had the digital transformation. At one point. We had the mobile revolution that we managed through. We had the connected TV, the rise of connected TV, we had the privacy revolution. We had the brand safety revolution. I'm not sure revolution is the right word for this one. We had the fact that we had to completely reinvent reliable, agnostic measurement over the years. I think we've done this very, very successfully. So now we're staring at the AI world. And again, I'm very confident we can do this. I think it will be. I'm actually not only think this I'm convinced it will be an expansionary moment. And as I said at the beginning, the progress and the ability for us to potentially take scientific discovery to the next level with all of this. And I have to thank you because frankly, via your partnership with us, I'm not sure you're so aware of this, but you're actually funding some of the amazing stuff we're doing on the scientific side. When you put it that way, that is just the reality. So if you ever wonder, right, wonder whether you should be spending with us or not. The reality is right. We do some of the most groundbreaking, world changing things in the world of Science, Medicine, sustainability and so on. And we're very proud of this. So thank you for your partnership. Thank you for being who you are. Thank you for being here with us. And yeah, very, very proud to be able to work with a group like this one.