 Welcome back to Snowflake Summit 2023. You're watching theCUBE's coverage. This is day one, wall-to-wall coverage. We go out to the events, we extract the signal from the noise. The man of the hour, Kristen Kleineman is here, Senior Vice President of Product at Snowflake. He had most of the keynote today. Kristen, good to see you again. Thanks for making some time for us. Thank you for having me. Excited to be here. Yeah, I mean, you still got a lot of energy. So it is only day one, but you've been prepping and you've been talking to customers. All data, all workloads. That's the high level message. And you know, George and I have been talking, leading up to this. You have the unified data platform. And we want to dig into that with you a little bit. You got a lot of ways to query. You keep expanding those. You got a lot of different data types that you support. And it's all integrated, is our understanding. Super important. Any ways to query, one engine. All integrated. So this is something that's a really important point that we wanted to talk to you about, because I think some of the times that gets lost in all the product fire hose. So explain that. For us, it was very important that the choice of programming language to query or program model, if you want to use a different API, it should be a choice of form, not a choice of capability or performance. The last thing you want to do is, oh, if you use Python, you get certain characteristics in terms of performance, but if you SQL use something else and Java is different. For us, it was a choice is choice, not trade-offs. So single engine, multiple ways to query the data. But, okay, but it's, I got to follow up on that. So it's not like a caravan where this, the whole caravan has to slow down for the slowest truck. Right? It's the opposite. You get the slowest truck and you figure out how to make it keep up with the caravan. Is that fair? So we started with some of the most demanding use cases. Like, if you look at SQL powering BI dashboards for large-scale data, that is a demanding workload. So we've been tuning this engine for 12, 14 years in those use cases. And I have no problem saying that by the time we started seeing some of the PySpark scenarios coming through Snowpark for Python, and it's like, why is this so much faster and so much better? It's because it was not, everything's slowed down. No, everything's sped up. And the same thing with Iceberg, you guys announced the sort of parody there. Our sense is, and just in discussions and just observing is, you want to be the best place to develop applications, fastest, most cost efficient, right? Most performant, et cetera. Correct. And very important is to articulate the motivation for that. If you believe the first four or five years of the company, and it still happens to his day, but the focus originally was, let's help our customers break down silos. The dream of having a single data fabric where you can ask questions about your customer holistically, your product line holistically. And what we saw was, okay, on one side you're unsiloing, and on the other side you are resiloing. And one of the biggest reasons for resiloing is applications, because every application realized with more data you do better. So our goal is keep customers unsiloed, bring applications, bring business logic to the data. Yeah, because George, I think a lot of people last year said, oh, Iceberg, that's just a competitive knockoff, as opposed to you want to be the best at. We want to be awesome at everything we do. And Iceberg is our way to meet customers where they are. If someone already invested in having tens or hundreds of petabytes of data, in parquet files, in cloud storage, I don't know that we need to force anyone down any one path, just Iceberg isn't a choice. And one of the big announcements today is these two modes of Iceberg Unified Tables. So if you just want to start querying data that you have in cloud storage, unmanaged tables are good for you. It's a read-only activity. But if you want to say Snowflake should take over the administration, management, consistency of that data, you can upgrade it to a managed table. So what we're giving customers is choice and stepping stones to decide where they are comfortable. So let's take that one step further. So some other vendors have managed to standardize on a table format and maybe even across vendors. But now, let's say taking this beyond that, where with the partnership with Blue Yonder, once upon a time an application vendor had to have many different data formats and engines for diagnostic, for predictive, for transactional. How can they standardize? Tell us how they're standardizing so that not only do they have one stack, but that stack can be shared, that data can be shared with other applications. What gets me excited about what we're doing is we're providing programming models that are familiar to folks. You know what you see, which is strictly a standard. Okay, that's one path. But data frame-based programming, I wouldn't call it a standard, but a de facto standard. And you can use that. So as long as you standardize or use quasi-standard access paths or access patterns, then how the data is represented matters a little bit less. But the benefit of supporting open file formats like Parquet, OpenTable formats like Iceberg, is even if you want to have a different way to access data, you can do so in an open way. So take that example of supply chain. So a company like Blue Yonder, which has a bunch of legacy manugistics workloads, could they take the container services, bring the manugistic workloads in there and then have it be a sort of first-party citizen in terms of their ability to use that data? From a realm of, is it possible? Absolutely. Once you have some business logic running in a container, you can export it into, say, a function. That function was callable from SQL or from Python. We put a lot of emphasis on the cleanliness of the architecture, so things are composable. And that's what you see there. It so happens that maybe Blue Yonder is not the greatest example, because they are in a journey to re-platformize directly on Snowflake. Well, they're basically re-architecting using Snowflake and relational AI, which is a heavy lift, but I love it, right? I mean, it's either that or go private equity, so good for you, Duncan. And many of the use cases on container service, especially for applications, is I am not ready to re-architect, but I want to bring that to run closer to the data. So that's one use case, but Blue Yonder, definitely very exciting what they're doing. Well, it's a very challenging problem. Frank has talked about this a lot. In fact, I think he even said one time, didn't he join a board of a supply chain company to really try to understand this better? This is a really, really hard problem. So let's talk about bringing data in, like the ingest pipeline and data engineering. For the longest time was a big data processing or a DBT, you know, and you're bringing raw data in, well, are you landing the raw data and modeling it, you know, from bronze to silver to gold? So now that you have more and more data types, and like Aplica for LLM, do you see all data pipelines becoming sort of multimodal, or multi-modal, I should say, where you're shredding the documents, you're refining them, you're enriching them with the analytic data, and so now your single source of truth is broad. So again, the purity of our design makes things composable and by virtue of that, usable in a variety of contexts. We show the user interface based off document AI, but there's also a function that you can call. That function can be included in a dynamic table or in a pipeline that you build yourself, you orchestrate it yourself. It could be something even orchestrated by Airflow or by DBT. So in no situation we're saying, oh, we got it all. No, that there's enough complexity in the world, but the key thing is if you have building blocks that can plug in and work well together, then you can do a lot of things. And the document AI is an example. You can also go and take language models, wrap them in a snow park function, even if you want to go out to the internet, and those functions all of them combine them in a single query. When it comes to gen AI, would you say your number one differentiator is your governance model? Yes, I would say the broad approach that we've been pursuing around bringing computation to the data is all about let's do rich programmability of data without compromising governance because we provide lots of capabilities to deliver governance. Gen AI in our mind is just one extra way to transform data into value. And frankly, what's out there is exciting, it's cool, but governance is not front and center. It was more capability and show of ability. But enterprises, they need to have both. Amazing capability, but make sure that my policies, my role-based access is in check and that's where many of the announcements of this week are exciting. So if we look at where application platforms are going, and I don't want to call you just a data platform, data cloud, but a foundation for building future applications, your power comes from all the data you're accumulating and curating in Snowflake because you use that not just to build apps, but to train the models that are part of the apps. That's your strength. There's two strengths, one is exactly what you said. You want the data to power these experiences, so that's great. But the other notion is once the logic is running close to the data, certifying or validating that that application is ready to run in my enterprise is a simpler process. Think of legal reviews and security reviews because I'm not copying data and that simplification of the process is appealing to both the app developer and the customer. So that validation or testing now involves the data and the algorithm? Correct. Okay. Imagine if I tell you, here's a big supply chain because we're there. Here's a solution that does supply chain. If I tell you, this app does everything you want, but you need to give me your data, how long does it take you to certify this app? But if we flip it, which is what we're doing, I say, here's the app, it's close to your data. We can vouch for this app is not copying your data out. We can vouch for this app is honoring your permissions. Will you be able to adopt that app faster? Our thesis and at this point is widely validated is yes, and that is attracted to you as the customer of the app, but also to the developer of the app. Can you explain the logic for the audience behind Neva? Yeah. We're very excited about Neva. I'll start saying that both the technology and the team, language models and JNAI in general, they demo extremely well. We all look at those things and it is magical. The closest we've got into magical experiences, but there's one part of JNAI, which is those models get limited context and then they start answering questions with high confidence based on their knowledge. And the implication of that is sometimes they make up stuff with very high confidence. It sounds like politicians. I'm not gonna come in there, I'm not gonna go there, but yes. And what Neva did and I think they were ahead of everyone was you don't turn search into an LLM problem. You just say, here's a question, give me the answer. What they learned how to do or they innovated is how do you take traditional search technologies, traditional information retrieval technology and augment it in the right places with language models? What they would do in their consumer search product is they could give you the results of a query that was conversational, but they could give you a citation. This statement we made is because of this link in the web. That in our mind is probably the single most important thing around JNAI, which is, you want to be able to trust it. You don't want to be able to ask questions and immediately have to go and fact check it and verify it. Imagine in the world of enterprise data, how it would work if you ask a question and you don't know if you can trust it. At least with public chat GPT things, you can at least go and ask Google search to fact check you, but on your enterprise it's harder. That is what they bring to the table, the ability to combine traditional information retrieval with LLM. What's the secret sauce behind their ability to do that or now your ability to do that? It's the fact that you're not just giving the entire problem to a language model. It's the fact that it's primarily a traditional search indexing ranking technology, but it knows how to better understand language, how to better produce language, how to better produce summarization. So it's the combination of all the new technologies in what we think is better results. And so let me just drill down on the capability a little bit. So I imagine in the ingest pipeline that might have used document AI to structure, better structure your documents and complex structure stuff, you might create embeddings, store that in a vector database and then Neva could do the semantic search. They would find me similar stuff, retrieve it, imprompt it invisibly and then summarize it. Yes, a lot of the expertise they did was knowing where to use embeddings, where to augment context, where to provide additional context. So it's that combination of traditional and new technologies that gets you to a new world and that's why we're very excited because we're bringing that to the enterprise. So zoom out and zoom in on announcements. So the big three, Frank set them up this morning, Iceberg OpenTables, native app framework, I would call the App Store for the enterprise and then Snowpark Container Services. And then you got into it. I mean, you had new open source Snowflake command line interface, you had new logging and tracing APIs, you had AutoSync with Git repositories. I mean, it's like, so help us understand the strategy and the announcements and how it all fits together. So, obviously the core of the announcements are bring more computation to Snowflake. That's where Snowpark Container Services dramatically expands on what we've done. We want to be able to provide distribution and monetization to this application, so this business logic. That's where native apps fits in. And one of the examples that we're showing was a native app backed by a Snowpark Container Service. So all of this efforts are synergistic. But some of the items that you mentioned, Dave, come from the fact that, okay, you cannot have an app development framework if you don't have a platform that developers like. Because developers, as we all know, they're opinionated. And everything you just mentioned, the Git integration, which is super cool, logging, tracing, command lines, Python APIs, all of that is to say, you cannot have platform without tools. And the message hopefully came out loud and clear. We're investing in the platform, but also in the tools to build on that platform. Yeah, making it easier for the developers is, I mean, if you're going to be the app store for the enterprise, it's got to be the place that they all want to go. Correct, baby. Best APIs is another one. They think of the app stores and the phones. It's not just the ability to run those, it's how do you build them? And it comes with that entire developer experience, that's where we're delivering in addition to the platform. When you look out at your opportunities, I mean, there's no shortage of TAM, as we like to say. Where do you see the limits of your architecture? That is a great question. I talk a lot about this with Benoit, that some of these extensibility bets, Snowpark and Snowpark Container Services in general, are going to test us and are going to push us into scenarios that we cannot envision. In the same way that, I don't know that Steve Jobs ever thought of having an app that controls the toothbrush. Like, is this use cases that are mind boggling? So it's going to push us, at this point we don't know exactly what those limits are, but the goal of the end of the keynote was to show, as we have today the technology, it already enables a very vast set of capabilities. So there will be limits, I don't know who they are, but we're excited about what's possible right now. And you're in control of some of those limits, as you said, it's really the ecosystem that is actually going to extend your reach pretty dramatically. The ecosystem here has grown quite a bit since last year. Yeah, and I'll mention some early partners already are pushing us beyond what we had in the prior preview. Smaller things that were in the plan, but by the time this is all generally available, we're very confident and very excited about what's possible. At which point some of these items need additional support from us, I don't know. So customer feedback today, I mean you've talked to a lot of customers, you get stopped in the hallways, I'm sure. You're actually just probably walking over from the other venue. What's been the feedback? What are they like? What are they pushing you on? What do they want to see more of? I think snow park container services overwhelmed the sentiment. I was just talking to someone literally, as you said on the way here, say, oh, I had a long list of things I wanted to try and my list just got way longer. But the one thing that I think captured everyone's imagination is the container services. I heard expressions like mind blown, overwhelmed, can't stop thinking. So, and that was a little bit of what we aspired to convey. Not because we want to do it, it's because that's how we feel about it. Yeah, well, you keep evolving. I think, you know, it's been described, oh, we started out here as the data warehouse and then we get the data cloud. Now we're more than the data cloud and can't wait to see what you come up with in the next couple of years, Christian. Same here, I'll say our partners in the handful of weeks that we gave them access to container services surprised us already. So it's going to be incredibly exciting from here on. It's great, well, thanks for coming on and you're going to come on next. We're really excited to bring on NVIDIA. Manu Radhas is coming on and we're going to dig into that big announcement that Snowflake and NVIDIA made last night. Dave Vellante for George Gilbert, Lisa Martins in the house. Christian Kleineman will be right back right after this short break.