Confluent, Inc. (CFLT) Earnings Call Transcript & Summary

Thank you for joining our 2024 TIMT Conference. Just want to special thank you to the 100-plus companies that are attending over the next 2 days as well as the 700-plus investor clients that are here today and tomorrow. Before I introduce our guests, I also want to thank our partners. It takes a village to put these events on. I want to thank our partners in investment banking, sales as well as marketing and corporate access. We'll host close to 2,200 one-on-ones and small group meetings over the next 2 days. And without that partnership, that's not possible. So congratulations, and thank you for being here today. If we could have our first panel come up to the stage. I want to introduce Jay Kreps, the Co-founder and CEO of Confluent; Matt Hedberg, who is our Global Head of Technology Research; and Anurag Sehgal who runs Client Banking, Data Analytics for our tech organization. Without further ado, I'll hand it off to you. Thank you.

Thanks, Mark. All right. Let's do this, right?

We're doing it.

Thanks, everybody, for coming. This is -- I've been here. This is probably my 18th TIMT Conference at RBC, and this is going to be the best. It's going to be the best yet. We're excited to kick it off with Jay this morning and Anurag. Anurag and I have done -- we've co-teamed a lot of these calls in the past. So hopefully, an RBC IT element brings another interesting angle to this conversation. So Jay and Anurag, thanks for joining us. Mark, Benny, everybody else, thanks for coming here.

So all right, Jay, I think everybody should know Confluent, but it's still -- occasionally, I'll talk to investors and they're like, "I don't know what they do. I don't know what -- I don't know Confluent and Kafka." Maybe just a quick overview of the company, a bit of your background? And then maybe I thought it would be -- on your Q3 earnings call, you kind of talked about different phases of growth. Where are we in that trajectory of your company?

Yes. So probably the easiest way to understand Confluent is really by reference to other pieces of data infrastructure. So mostly when we think about data, we've thought about databases, kind of these big storage systems where you can put your data to the back end for some application. That's been probably the dominant force in data for a long time, biggest chunk of the spend. Our Confluent is a little bit different. If you think about that as data at rest, kind of where does data go to sit, Confluent is all about data in motion. And the observation is as softwares evolve, you don't just have one big database and one big application that kind of does everything. You have hundreds or in a larger organization, thousands of different data systems, data sets, use of application software, SaaS systems, data platforms, analytics tools. And it all has to interconnect, it has to all act as one company. And increasingly, the use of software has kind of moved from the edges, little productivity apps to something that's really driving big parts of the customer experience, that's driving big parts of how the goods and services are produced, kind of right in the real-time flow of the business. And so what Confluent does is kind of connect all these things up, allow the real-time flow of data and then allow applications to kind of react or respond continuously in real time to whatever is happening. And this data in motion or data streaming, now that's become a sizable component of architecture over time. And that's very much what we helped to kind of create and take out to also the original -- technology was actually a piece of open source software called Kafka. This is something I helped to create prior to Confluent when I was working at LinkedIn, and it was originally an internal piece of infrastructure there, went out and was released. This open source was adopted by a lot of the big tech companies at the time, Ubers and Netflixes of the world, and they were really moving to this architecture, where instead of shipping big files around at the end of the day, you would take data as it was generated as a continuous stream. And so what does that mean? You can imagine in an Uber wherever a car is driving, there's some stream of data, where is it, what's happening and whatever ride is requested. There's a very kind of real-time logistics component to that, that has to be continuously tracked and modeled, so you can do supply and demand and pricing and dispatch, everything kind of keys off this real-time model of the world. And you can imagine for them, being able to operate off of that, it's a very different type of data problem. And it turns out that although it might seem quite different, there's some similarity between that and the internals of the social network where you're bringing together all the data, what's happening for relevance or something like Netflix, where it's again a lot of personalization and recommendations and user behavior. And as this technology has been adopted, we found it doesn't stop there, right? It's useful in financial services, some of the largest banks in the world, many of their systems, very diverse, come together and are ultimately working off of, hey, a transaction occurred, we interacted with our customer. That triggers a lot of activity downstream that occurs. And so Kafka and Confluent have become a kind of key component really across virtually every industry you can imagine. So that's the 30-second story of the technology and a little bit of Confluent as well. So then how did we take that from an open source project to a company? Yes, it's happened in a couple of phases. So the first phase of the company was creating a commercial software, license software offering around this. And we started with that because there's a ton of people using the open source. This is the easiest way to get it out in the market. I would think of that as kind of [indiscernible]. That was -- the initial growth was built off that. But even from early on in the company, we knew and had planned to build a fully managed cloud service. And that was kind of the second wave of growth. So as we were going public in '21, we had a pretty significant investment in this cloud offering, but it was still relatively nascent, maybe it was about 17% of revenue, something like that, but we had a lot of confidence that that was going to be one of the drivers of the next kind of leg of growth as this came out and allowed us to capture a broader set of the open source usage. And indeed, that was the case. And now a lot of our focus has been on kind of broadening this platform. So if Kafka was this core stream of data that connects things, our vision is really to broaden that into a full platform. We're working with real-time data, how do I process data continuously in real time, how do I capture it in real time, how can I govern it across a company. We would call that a data streaming platform. So really taking everything people want to do with these real-time data streams and making it available in a coherent package. And that's received, I think, fantastic reception from our customers. A lot of the work is bringing all that functionality to life. Some of these processing capabilities, the governance capabilities. And that, to me, is very much our kind of third act as it was really going to a general purpose platform for real-time data, helping customers realize that across the thousands of use cases that is applicable for. And that's a significant focus for us now.

Excellent. Thanks for that. Anurag, maybe introduce yourself quickly and then maybe just describe how RBC uses Confluent today.

Sure, Matt. So I've been with RBC for about a little over 2 years now. Used to work at Credit Suisse before this, where I started using Confluent back in the day. And so if you think about Confluent is the enterprise-grade, enterprise standard for event streaming, what does that mean to us? 200-plus applications, streaming real-time events, ingesting events, distributing events and consuming events. There's about $3 billion plus events on a daily basis that are being streamed over Confluent Kafka. It is essentially growing exponentially as more and more real-time sources of data are consumed by various teams across RBC, but also distributed to clients. Use cases, I mean you think about any and all use cases in the current day and age, leveraging real-time data. I mean fraud detection is leveraging a significant portion of Kafka, you look at payments, real-time payment events in our retail bank, that's leveraging Kafka. Within capital markets, if you look at the recently launched U.S. cash management business, that is leveraging Confluent in both consuming, sharing data across regulatory events, payment events and as the sort of -- all of the cash balances events that are coming up with clients. We're also leveraging Confluent in being able to provide real-time customer alerts on certain events that happened. So again, like enterprise grade, well supported, right? I mean, we could easily use open source Kafka, but the reason why we choose to use Confluent is because it is enterprise supported. And so if you think about the day in the life of a developer, if they're using open source, the amount of time we end up spending in maintaining, the number of risks and vulnerabilities we see with open source, in managing that process ourselves versus leveraging a partner that is investing in the product but yet having the flexibility that it is open source and that it can be deployed across multiple clouds, on-prem or cloud, that makes our lives really much more simpler and allows us to scale much faster and allows us to achieve better time-to-market solutions, but being sustainable at the same time.

Excellent. I'm going to get into where we might be going with -- or maybe Jay, you might be interested in where we're going with Confluent in the future. But maybe just -- this is -- it's an opening keynote and a lot of us are interested in other things in software. I guess, Jay, from your perspective, you've been around tech for a long time. We've seen a lot of trends from the mainframe to the PCs to the Internet, generative AI and SaaS. What excites you the most? Like what is it about today that excites you the most about the future with all this incredible innovation out there from a tech perspective?

Yes. I think it's a pretty exciting time. I mean the set of things that are coming together, I think, is very interesting. And a lot of these waves kind of build on each other. The rise of AI, I think, is really interesting. And this is an area I've followed for a long time I've got into computer science by way of studying AI and have been involved.

Before it was GenAI.

Yes, yes. That was actually what brought me to LinkedIn was to work on some of the more analytical, data-driven, machine learning applications. That was some of the early use cases for Kafka is how do we actually get access to all this and be able to apply it back in some kind of customer experience. And so to see some of the harder use cases start to come to life, really a major leap forward. I think it's amazing. I think we're starting to see beyond just the language models, some of the things in the real world, some of the robotics applications are kind of on the verge of being something, I mean that takes time, but I think that's very exciting. So when you put all that together, it's certainly an exciting time for just technology if you put the pieces together over the last whatever decade you've gotten this incredible cloud and data systems to allow you to kind of harness some of the stuff you're getting these kind of AI capabilities that really allow you to start to close the loop in different ways. You've gotten kind of ubiquitous kind of Internet or IoT or the ability to get stuff out in the real world that can connect back into that system. And so you kind of put those pieces together the parts of the company or organization that can be kind of modeled or improved in software. And the digital realm is just like orders of magnitude, what it was. And so I think that's really interesting. I mean I think it's fascinating, I think we're going to see a lot of change come out of it in the next few years.

And a lot of disruption, too, right?

Yes, yes. It's interesting. People have always very bold predictions, but actually, when you have a lot changing like this, the distribution is flatter. You actually -- there's more uncertainty in how it all turns out. But I think it's -- certainly just clearly as a technologist, it's an exciting time. I'm excited to see where it goes.

Maybe because we'll kind of get the AI piece out of the way here early. I mean, Anurag, we've been leveraging AI for years. And all of a sudden now, we're starting to build our own generative AI applications. How do you think about sort of where we're at and what we're doing internally as a bank from an AI or a GenAI perspective?

Yes. I think, I mean, we're really leading the way at the moment, if I look at the work we're doing in generative AI with the launch of a product called Agent Assist and really think about Agent Assist like in 2 swim lanes, one being a generalist swim lane, which is essentially how do we enable the entire bank to leverage ChatGPT like capability in a way that's compliant, that's safe, that allows us to really experiment fast. But also behind it is a RAG architecture that allows individuals to upload their own information in a safe and secure manner. And the second swim lane is where we really start to see much more automation, which is the specialist swim lane where we're starting to look at the day in the life of a user persona and sort of how can we start to answer very sort of simple questions or provide straight through automation. A great example of that is within the research team where we've just launched QuickTakes, which really looks at real-time press releases, real-time company filings, SEC filings, real-time earnings call transcripts and allows us to generate out of the box QuickTakes that analyst can take and augment and publish to clients much faster time to market. And again, that's a really great use case where real-time streaming of data plays a really important role in that GenAI world. And so I think -- that's where I think the sort of Confluent, Kafka intersects with the GenAI world in a really nice way.

Interesting. The question that I always get is, it all sounds great, but like how are customers actually going to pay for this. Like to what amount is it table stakes? What is it? Where is it a monetizable feature? I mean how do you think about like from a -- I mean, it could be from a consumer perspective whether you're buying stuff for RBC or whether you -- what you guys are consuming externally from an AI perspective. How do you think about the customers' willingness to pay for this kind of technology?

Yes, yes, I have a few thoughts on that. I mean I've kind of followed that debate of like, okay, this much venture capital has been sunk to AI companies and only this much money has come out and one number is a lot bigger than the other number. Yes, it should, actually, right? Like the investment is either ahead of or much larger than the return we've seen so far. But if you look at what we've seen so far, it is kind of these very early things, it's a bunch of SaaS companies sticking to chatbots in whatever little tools you use. I think that that's not the end state here. I think it's unlikely to be end state that a company like Confluent is consuming AI as 115 little chatbots embedded in different SaaS tools. That's not the end state. Like even these chatbots are kind of a step, right? But a lot of what we would want is actually something more directly integrated into the work that we do, right? And I think that's true for a lot of our customers. I mean, even what we've seen in the usage of our platform. I think these chatbots are great. We show up in a lot of these different architectures, trying to get the data for RAG. But the kind of more interesting stuff is, I think, as you're saying, okay, what is the core business? How can we apply this? Something happens in the world, and we want the AI to do something. And so I think there's a great example Anurag gave, but there's a similar example as you look at like a customers in an insurance company. A claim is filed. There's a whole set of work that's done around processing. A fair amount of which is actually like not digital, it's humans, right? And so the ability to actually have the AI take a shot at some of that and give the human something that's like a pretty solid first draft, then they can edit it and you can kind of start to iterate and get that loop going up, maybe it doesn't have to be a first draft, maybe it could be a final draft. I think that's the type of use case that I think is where you're going to start to see the real payoff. That takes time because if you think about that company is kind of now really changing their business process in a meaningful way versus buying a module in a SaaS tool. But if you think about the payoff from that, I think it's much more substantial. And so I think that's not unusual for any technology that, that time to trickle out and actually have the impact. It's years, not months. And maybe that's surprising. Some people think it should not be. So I do think there is a question of like, okay, how rapidly can companies realize this. But I think the actual impact is pretty substantial. I mean the models continue to make progress. I mean, rumors to the contrary, right? Like they continue to progress. And already, just in what's there today, there's a fair amount of unrealized value. If companies took what was on the shelf today, when we put that to work, that would take some years to do, but you would already have a fair amount of value unlocked in that. So I'm pretty optimistic about where that's going. Nothing's overnight. But I do think that there's a wave of use cases and applications kind of building around that. I think that's just -- again, as a technology, it's exciting to see. I think a positive thing for Confluent and then that becomes a consumer of data and in many cases, something that actually directly operates on these streams. Something happens in the business, that's an event that they want to react to in process. And we've built functionality that helps with that. But even more broadly, I just think this is a -- it really closes the loop in software. Our paradigm in software has been very much like, let's give humans a UI that they can kind of click on in different ways and that's kind of the way software does work, right? And that's not really what we want, right? In the end, we want something that can take on more of that, do more of it. And I think you're starting to see little bits of that shine through here and there.

Anurag, one of the questions that I get all the time is, it seemed like AI really disrupted kind of run rate IT spending this year because I think a lot of people came into this year uncertain of how they're going to address AI and GenAI and I think it did cause a lot of disruption. Now that we're sort of a year into -- or how many years are we into the hype cycle? I don't know, 2 years, 3 years. Do you think there's some more predictability in how like a bank like RBC thinks about deploying assets, dollars for AI? And then it -- look, we're almost at like a new normal now. There's not as much of like, we got to cut some spending and divert it to AI. What's sort of your perspective on that?

I think somewhat there, but not quite, I'd say. When you look at generative AI, I mean, at the very early stages of thinking about how disruptive this can be, I think of it as a once-in-a-lifetime moment, honestly, because there's so much disruption that this can drive. The interface in how we operate and how we sort of work with clients and how we work across teams can really change with the generative AI. And so I think the prior question you asked, like when you think about any of the disruptive technologies in the past, it takes about 10 years before you start to realize, like an iPhone gets launched, well, you didn't know how many more the app ecosystem on day 1, no one could have realized that, or when the Internet came out, like you wouldn't realize like in the next 20 years how disruptive that could be, right? So I think it's the same with generative AI. We will, in the next 10 to 15 years, realize how disruptive this is going to be. So I think as you think about that, like think about the spend on AI and generative AI. I mean we're just setting up a GPU farm on-prem, just to be able to preserve capacity on GPUs, that's not a problem we would have like in the past, we could easily get GPU capacity on cloud. But at this moment, like everyone is going after GPUs. And so we're setting up our own sort of farm on-premise to be able to support that. I mean, I think our cloud spend in many of these cases will continue to go up. I mean, we weren't paying for open AI services like ChatGPT models or Cohere or any of these models. The compute that goes behind this on cloud will continue to grow as well. So a lot of that growth will depend on how many use cases we're tackling and how big these use cases end up being. I mean we know these models take large amount of compute, both on embedding or inference. So -- yes, I think I don't quite think that it's as predictive today. I think we'll get there in the next 2 to 3 years.

It's a good perspective. I think we all want like instant gratification from AI. But it's a good perspective that these things do take time, decades, but the excitement is certainly there. Jay, the other topical question that I'm getting, and we talked a little about a little bit beforehand, is now that we're post the election, at least there's less confusion around that. We've seen one Trump administration. I think there's a lot of questions about the second one. One of the questions I get all the time is like, with the department of government efficiency with things like potential tariffs or things like lower corporate tax rates, there's a lot to think about from a technology perspective. Do you have any perspective on what the next 4 years might look like from a tech perspective?

Yes. I mean I think this is again one where there's a lot of uncertainty. Certainly, we sell to the government, anybody who sells to the government, you do feel that there are some inefficiencies that can be improved. Like how this manifests, I don't know. Probably everybody in the room has a smarter opinion on what the impact of tariffs would be or whatever. I think that's fascinating, but I don't have any kind of concrete prediction. It certainly is the case that there is an opportunity in the government to apply technology and make things more efficient. Like so whenever I hear that, okay, there's 100,000 people working in the IRS. As a former computer programmer, I do feel like a fair amount of computing your tax return, it should be something that computers could do. I don't want to make it sound easy. And obviously, there's a lot of interfacing with people. The 100,000 employees is a lot of employees. And so I do feel like, okay, there must be some opportunity for the application of software in these areas to kind of drive...

The benefit, right?

Yes. Just catching up to where the rest of the world has been for the last 20 years or so. And I would -- I think that would be a great thing to see happen. I don't know that that's going to be what does happen. But if that's the case, I think that would be positive. Certainly, one of the interesting things, as you travel around and you can see how different countries have approached digital services, I think it is an opportunity for relatively low investment to drive a lot of efficiency. I think some of the stuff India has done is amazing. Like we talk -- we're involved in a lot of different payment systems. And when you look at what these payment systems do, they process some number of transactions. And they're very important systems. You go to India, the number of transactions is like, whatever, 100x, 1,000x larger than the next thing. And it's because there is a system that's universally used that you can buy anything that has kind of very low -- I mean, effectively no fee, no merchant fee, so you can go buy some fruit at a stand and pay with this. And this is something that they help standardize. It's kind of a private-public partnership. There's a whole thing set of things like that around banking, identity, et cetera. They actually just make it really easy to get services out of the government. If you think about what the U.S. government does, a shocking amount of it always comes back to figuring out who you are and what you have permission to do. It's kind of identity authorization, but all done with like some card and some PG&E bill or some -- I mean the most ridiculous way of proving who you are. And so it does seem like there's some like fundamental missing stuff there that would actually just make everything more efficient. I don't know if we'll get that. I don't -- maybe it's just a matter of whatever people have said, [ firing ] people with social security number and odd numbers or something. I don't know if that gets you somehow a better system or just maybe at least a smaller system. So I think at this point, nobody has any kind of deep insight, but I would certainly love to see it happen that we get kind of a more efficient government. I think a lot of what the government does is really important, and there's certainly some opportunity there that you see if you work with the government closely as a customer.

Well, it seems like software technology should be more of an enabler for increased efficiency if you've ever gotten…

Yes, that would certainly be the direction I would hope for. I think it's obviously, at this point, nobody knows.

Yes. That's great. I wanted to talk about open source. It's been one of the most disruptive trends that we've seen, and you guys were actively involved in that. I guess the high-level question is, when you think about the future, what are areas that you think in the tech stack that are potentially at risk from open source disrupting, sort of paid?

Yes. I mean there's risk and reward. Certainly, the areas that I think are most interesting and what I think open source does well is help create these kind of de facto standards that people want to use and build around. And so where I think that there's opportunities is where there's some missing standard that could actually change how some part of the industry works. One of the ones we've been most interested in and are kind of investing around ourselves is something called Apache Iceberg. And there's some investors who follow the data space may be aware of this, but what's happening is the kind of raw storage in the cloud, it's obviously something that's very open, it's a cloud service, right? And yet if you store data in most analytics tools, it's kind of locked up in that tool. In many ways, that's been kind of the foundation of how data warehousing, even in practice, a lot of these data lakes houses have worked. There's some silo that has the data. And anything you want to do with that data, you practice kind of have to do through their tool. So there's some kind of toll that gets paid for every usage of data. And you end up having many copies of the data in each different system and you end up paying that tool many times, even if that tool is not the best thing to use your data. And what this standard Apache Iceberg, it's an open source project, but I mean it's not even the world's most complicated piece of technology. It just makes tables -- like database tables of data available in object storage in a way that is across any of these different systems. And so you can imagine you're kind of going from a model where there's really only a single company that can sell you the analytics processing of your data to a model where effectively, it's an open market, right? Capitalism comes in and everybody can compete to be faster or more efficient or more cost effective or have better functionality for a particular use case. So instead of having kind of one global warehouse or lake house that solves all problems, you kind of end up with more of an ecosystem that allows you to harness data in different ways. And I think that's a very powerful thing when you get that kind of standardization unlocked. And so I think we're very excited about that. At Confluent, we've done something called Tableflow. So any of these streams that come in, it can just be opened up as Iceberg Tables then exposed out to this wider world of different analytics technologies. So you can query it in Snowflake or in cloud provider tools, things like Glue and Athena and AWS or other cloud providers, something that's very portable across. And this is really appealing to customers. You feel like, hey, in some of these cases, we've been little beholden to some of these vendors. We don't feel like we totally have control over the bill that we're getting in the end. And then beyond that, we often end up with tools that are not really the thing we want because that was what worked with them. And so I think that kind of openness is going to drive a lot of innovation, both by kind of using the existing players to do better as well as by inviting new players to come in and compete for different workloads. So we're excited about this idea of taking all these real-time streams, landing them in Iceberg in an up-to-date way and opening that up to the wider world. I think that's going to be a pretty exciting trend in the analytics space over the coming years. And we're starting to see adoption of that in a lot of the larger, more sophisticated customers already.

Anurag, Iceberg comes up in just about every sort of infrastructure type call, whether it's analytics or even observability. Are we leveraging Iceberg?

Absolutely. I mean, look, I think we look at our principles -- guiding principles on cloud, a lot of what we try to do is provide for open data formats and portability at its core, right? So today, if you're deep in one provider and you're sort of dealing with their formats, it becomes extremely hard, it's so sticky. And we've seen that time and time again, both on-prem and on cloud, where we've had providers increased pricing on us, and it is so hard to get away from them. And so open data format is at the core of what we are trying to do. Iceberg and Delta are the 2 leading formats for us. Certainly, Databricks talks open data all the time. When I started working with Databricks 8 years back, I mean open data was their driver, and that's the message that they drove home. And that really resonated with us back in the day and still does. They've since then acquired Tabular, which is really, again, big on Iceberg. And again, Iceberg and Delta, I mean, you could pick one, but Iceberg has more flex across more providers and more cloud providers and more analytics providers. So I think that there's no doubt there's value. There is still a journey to be had on how we move everyone else that lives in legacy world over to open formats, and that's going to be a hard path forward for many of us.

No, yes, it's -- so what I'm hearing from you, Jay, is because I think it could be disruptive to a lot of companies. You see it as additive to the Confluent story at this point.

Yes, yes. We're a little different. Obviously, some kind of data platforms have been about sort of locking up your data. We -- our business is actually to unlock it. We are trying to get it to many places. One of the exciting things for us is there hasn't really been a great way of landing data in real time for analytics. And so you have to do a lot of different kind of work around, but it's a fair amount of work for customers to hook that up. And so this is something that basically takes all the customer use cases we have today, and kind of opens up the analytics world to those. So yes, very much additive and very much something customers are excited about and excited about pursuing.

Yes. We have a couple of minutes left. I wanted to touch on just one thing really quickly because it strikes me interesting that there's been so many -- I'm thinking across tech changes in go-to-market approaches. You guys had one. Others have, I think -- honestly, I think, struggled with how to sell in a post-COVID world. It feels like, today, it's very much an ROI-driven sale, which feels like you guys are at the core of as well. Can you just reflect on like why we're seeing so much disruption in sales forces -- technology sales forces these days?

Yes. I mean I think the root thing for check is there's pressure on spend, which requires everybody to up their game, right? And so I don't -- I feel like that's one force that's happening. The other thing is, I think company -- as you up your game, what are you trying to do, I think companies are trying to get smarter about helping their customers. People say -- realize value, that sounds like a euphemism. It's actually not, right? Like what are you trying to do? You have some widget, gadget data thing, whatever the thing that you've built is, it's not inherently valuable on its own. Customers have to somehow put this in practice in some way. And I do think companies are getting much more sophisticated about how they help customers realize that value. And in many cases, the business models are aligning to that. So a company like us, it's kind of oriented around consumption for our cloud. It's very much the case that as customer uses more of our product, they're spending more. The driver for that is going to be kind of successful production use cases, which is usually the thing valuable to the customer. And so that's certainly a better model, I think, than kind of unused seats or whatever reallocated clusters are, whatever the precursor might have been for different companies. So the closer you can get to that actual unit of usage, and then that allows the providers to be smarter about ways that they can help. What are the things we can offer to customers to help them find use cases that are actually worth doing, that are exciting, they're going to have high ROI. And of course, the motivation for that is, yes, an environment where IT spend is a little tighter, you want to be smart, you want to be smart about it. And I think you're right that I don't think you switch from purely innovation to purely value selling or whatever else. I think these things are ultimately a mixture, right? Customers are looking for products, which unlock something genuinely new that's worth doing and save them money and have a really achievable path from point A to point B. And the things that get bought in tighter times are the ones that kind of check all the boxes.

Yes. Two minutes left. And you can't say generically GenAI. I want to know a big bold prediction. Do you have a bold prediction or something that we should be watching from a tech perspective that maybe you're most excited about, that maybe is not top of mind for folks? Both of you.

Yes. Yes. I think there's a number of interesting things happening. I won't say GenAI broadly, but I do think that watching these models get applied in different domains I think is one of the more interesting unlocks. So like I'm involved with this company, Anthropic. They just did this computer usage, right, which is like literally the model using the computer. That's one example of taking something where we tend to, in our imaginations, think based on what we've seen, right? So if we saw a chatbot, we think more chatbots, right? And I think that's an example of how this applies more broadly. I think you're starting to see some of these end-to-end models work with kind of vision and robotics as I was saying, the self-driving cars, which went through their own hype cycle, right? You know they're actually kind of working in San Francisco, you can get anywhere in a relatively complicated city. And I think that's one which is weirdly underrated in what a change it is. So I think the -- like this expansion of AI into these different kind of domains, I think it's actually a really big deal as you plug that in.

More vertical-based use cases, though?

Yes. Yes. Think of it as the I/O, right? Like to some extent, computers haven't changed, but having a phone that knows where you are and has these other capabilities makes it very different, right? I think it's the same thing for these AI models. We're going from chat to something else that you can interact with the world in different ways, actually really changes what's possible. And so I think we're in the early parts of kind of seeing that realized.

Anurag, what about you?

Well, I mean, I think again I'm not going to say generative AI, but the application of it in various verticals, as was mentioned, I mean, I look at entertainment and like how is this going to change making movies or music or art and like the unexplored. I think you can see the beginnings of that, that's, again, really interesting, or in health care where you're starting to see more and more of the sort of patents that are coming out and medicines for different diseases, different applications that are leveraging AI that's core or even when I think about our own data centers and intelligent operations, that's a more lower-hanging use case in my mind. Like that's probably in the next 3 years, like how we look at managing our data centers and capacity and everything else. When things go wrong, like predicting where things are going to go wrong and how to address them in an automated manner, I think bots are going to be doing a lot of that.

No, I appreciate that answer. As we wrap up here, one of -- just a fun last question. Is there any tool, a GenAI tool that use in your personal life that you find particularly helpful. I'll start and give you a second to think. If anybody -- if people have not used NotebookLM, it's a great tool. It's just -- it's a fun little tool on Google. So I would suggest checking out that. Are there any other fun little like personal tools that you guys have used from a GenAI perspective?

I mean I don't do a ton of programming anymore, but these programming interfaces, things like Cursor, I mean it's kind of amazing, the quality of the models for coding and just what this enables. So even if you've been away from it for a while and you're kind of rusty, you can actually suddenly do...

You're a coder again.

Yes, fairly meaningful things. And so I think we're early at actually seeing the impact of that. Like we kind of rolled some of the stuff out with our team. And like, yes, it's helpful, but it's not like the team is 50% more effective. But I think if you look at the progress of those models in the last year, just on like benchmarks, problem things, there's still work to do to get it fully integrated, but it's like amazing improvements. So I do think that's one that is going to have interesting implications where suddenly a much broader set of people are going to be able to do this.

Anurag, any last?

Well, I've used Notebook, it's really good. Obviously, ChatGPT is very regular use. And Meta's AI for sort of transforming your pictures.

What is it? Which one?

Meta. It uses language models in the back end, but how you see like your own picture visualized in a different, like it's pretty interesting.

All right. Helpful. Well, guys, we're out of time. Really appreciate the conversation from all of us at RBC. Thanks, thanks for your time.