Latest / The Data Splash / The Fortune Teller of the Data World - with Yaser Najafi
Transcript
- Ido Bronstein: Welcome everyone to the Data Splash Podcast brought to you by Upriver If you're a data engineer , a data leader, or just someone trying to make sense of what the hell is going on with data and AI right now, you are in the right place. Here with me today is Yaser Najafi. Hey Yaser great to have you here. Yaser Najafi: Yeah, thanks for having me. Glad to be here. Ido Bronstein: Yes, very much. So to whom that does not know, Yaser leads technology alliance and investment and Snowflake he is one of the few people that decided which data startups is collaborating with Snowflake and Snowflake puts their money behind them. So I'm very happy that you're with me today. We had over the past month a lot of interesting conversations on like where data and AI is going in the future and happy to have this right now in the podcast. And so today we'll talk about of course, Snowflake and where it's heading and how the investment of Snowflake is affecting the future of Snowflake and the whole data space essentially. But before we are jumping in, we have a small tradition. It's called the 30 Second Splash. So Yaser, are you ready? Yaser Najafi: Let's do it. Ido Bronstein: So no time to think let's go, A buzzword you never want to hear in a pitch again Yaser Najafi: AI native. Ido Bronstein: most overrated feature in the modern data stack. Yaser Najafi: probably reverse ETL. Ido Bronstein: Yes, Name a category that is actually just a feature. Yaser Najafi: data catalogues. Ido Bronstein: dream vacation that has nothing to do with data. Yaser Najafi: probably anywhere where I don't have to take my family. Ido Bronstein: Okay, I understand it was a rough one. Okay, Thank you very much. It's very good that your family is not a data engineer in this space. Yaser Najafi: I'm taking them on a nice vacation next week, so it'll be okay. Ido Bronstein: Yeah, so it will be, I'm sure it will be super, fun. I think that first of all, it will be best for our audience to understand like who you are and how you eventually ended up in like a Snowflake investment group. Yaser Najafi: So yeah, I've been with Snowflake for a little over three years. you know, I came over from our alliances team. I managed our relationships with our major strategic ISVs. they did a lot around the Snowflake ecosystem. And so you know, Snowflake Ventures, we make strategic investments in startups, but mainly in partners, right? We want to ensure that our partners are Building and delivering the you know the next greatest technology that allow us to do a lot of great things. so it's really about finding the right startup and finding the right partner before the market does, and how Snowflake partnerships or investments can be the fuel around go to market to help these partners deliver faster. but I'm also a practitioner. I've got, you know, 13 plus years in technology sales. I love to build things on my own. My philosophy is that I like to put myself in the position where our partners and startups are actually trying to solve the problems. Cause if I can't feel the pain, it's very difficult for our 15,000 customers to understand what the value is when the integration actually happens. Ido Bronstein: Yeah, amazing. you said that you being hands-on built is what's helping you to evaluate a startup and understand the pain that they are solving in a sense. Yaser Najafi: Yeah, a hundred percent. It takes a lot because you know startups put founders put a lot of time, resources, sweat and energy into building the products and it's less about your logo on our website and more about "do I believe in what you're solving for? Can it solve a problem that I have?" Because if the answer is yes to both of those, it's like it's a very clear cut answer around how we want to partner or invest in. and so putting myself in the driver's seat has allowed me to have more of an opinion around what we can and can't do. Ido Bronstein: Amazing. So you said that the first thing that you all look on a startup is before the founders, before the technologies that's built, like which problem they're solving and if it resonates first of all with you as a builder and then as a proxy with the market. Yaser Najafi: Exactly. Yeah. Ido Bronstein: And when you're looking at where Snowflake is like investing today, like What do you see is the pattern? Because also Snowflake builds a lot of stuff, you know, and the platform involved. So how do you see the alignment between what Snowflake builds and what the startups that in the ecosystem brings to the table? Yaser Najafi: Yeah. this is a, you know, if I don't think there's a right answer for this. I'm gonna give you an answer and this answer is probably gonna change, you know, six months from now. But you know, we've made over 70 investments since our inception in 2019. We've made a lot of you know good investments, some okay investments. but in the last I would say the last twelve months, maybe last year, a lot of our focus has been around obviously AI as every other strategic or venture firm has been looking at. And it's mainly because, you know, obviously AI is the you know hottest trend right now, but it's also the what we're looking for around, hey, how is AI going to accelerate the use cases that Snowflake was really built for around data? So the companies that are successful are the ones that are solving the problems of yesterday with AI. plus the plus the infrastructure to back it up. So just it's less about throwing an LLM at a specific problem, but more about understanding what the workflow is, how customers did it before, right? In 2019, 2018. And then how Ido Bronstein: Yeah. Yaser Najafi: are they using data plus AI to help them make well-executed decisions? So that's where we're looking for. It's not a specific hey, point solution around, hey, we do this, so we're gonna likely make an investment. Yes, I think. If you look at all of our investments, there's probably a flavor of AI around it, but it's more around the actual problem that these companies are trying to solve for. Ido Bronstein: Interesting. So in that matter, it's a bit of a shift like what we, compared for what I see at least like couple of years ago, like in the beginning, Snowflake created an ecosystem around data. You invest in data catalog, semantic layer, observability, like pieces in the puzzle of the ecosystem. And now what I'm understanding that you are more looking on vertical solution in a sense, like how I'm using AI in order to change. a specific industry using Snowflake, the accelerator for that. Yaser Najafi: Yeah, we kinda lump them up into a few buckets. but the vertical industry solution is one of those buckets, mainly because data is still the defensible moat, right? creating the vast amount of data that customers have in Snowflake is very difficult and it's you cannot replace that. the value that customers want with a lot of these emerging startups solutions that are that are being you know that are coming out to market is how can I go create context of data that I collected five, 10 years ago and feed that to a business application that's gonna help me predict what's gonna happen tomorrow. that's the differentiation. But also around kind of the horizontal aspect of it. we're gonna see less investments, less focus around the application layer, right? Building software, building UIs are relatively easy these days. When you layer that with data, what can that do for me? Right. Ido Bronstein: Yeah, more focused on the workflow and how I can use the context in order to really enhance the workflow. Yaser Najafi: Exactly. Yeah. Ido Bronstein: Amazing. And I think that like, when you're looking at the data, something that like I see, like it's not the new trend but it's just getting bigger and bigger. Like is the ability to combine like unstructured data as part of your context layer. And Snowflake started originally from the very analytics structure data. How do you see this evolve and like. Snowflake as an investment group or as a platform looking on the context from end to end in that manner? Yaser Najafi: Yeah, I mean it's I'll give you my answer, but I'm sure you know someone from product will likely give you a different answer. And you know, last time we met and saw face to face at Snowflake Summit in June, I think context was the buzzword. you know, earlier when you when we went through our you know thirty second buzzword, I think context is probably number two. Context is very important because again, we have so much data, right? And this data means something to differently to different applications, different users, different individuals. and the challenge that we currently face is not all that data lives in Snowflake, right? And typically when someone uses something like Snowflake, they expect everything to be in Snowflake. Well, the reality of it is, AI agents don't care what the platform is. They just want to go answer the question that the whomever executed to the agent gave them. And so having the ability to not have not only have context of the data context of the data in Snowflake, but context of data outside of Snowflake is instrumental for how all these AI agents are going to be able to solve complex problems in the future. We believe that And I think we all agree that all the data will not will likely never land in Snowflake. While we want it to, right, there are going to be data sets, unstructured data sets on premise, likely in different data platforms. What we are trying to allow through partnerships and working with specific startups is the ability to have you know the unified context view. So that an agent always has the ability to access the right data sets in order to make the most accurate decision. And so it's extremely critical and it's very, hard to do. that's why it's that it's one of the things when we when we talk to startups and we talk to companies, it's hey, this is how we do it. And we say, okay, let's peel the onion back to really understand how you're doing this because it's a big problem and a lot of people are trying to solve it. Ido Bronstein: Yeah. And I think this is one of the shifts of our industry like past, like it's not about where the data is. It's not about the storage. It's about the layer that governance and control like all this and really translate that to the storage and the location to the agents from a government, from a semantic perspective. And yeah, so Yaser Najafi: Well you you you brought Ido Bronstein: will be interesting to see where it goes. Yaser Najafi: well you bring up another good point is around so context is one piece of it understanding the semantics of the data is one piece, but the governance and the security is another and a lot of times what we're finding is you know coding agents are great because they can go write code and they can write SQL and as long as you have access to the data, go do it. But do you have access to the data? Should you have access to the data? And unifying Active Directory, LDAP, that if it's on premise, you know, IM policies and our own role-based access controls, is there a way to bring that all together? Because at some point in time, you know, things are going to likely you're likely going to fall under risk, right? Or you're going to get inaccurate data because the LLM is going to hallucinate because it doesn't have access to the right data. So it's a layer cake. It's not just specifically about the context. And so it's more often than not, you know, the problem is bigger than we actually think it is. and that's why it's a good experience to have a lot of companies are doing it in security, and then there's a different way to look at how you want to solve the problem. Ido Bronstein: Yeah, It's a huge, problem. And yeah, we're in the industry still trying to figure out what is the best way to do that. And you can look on all the security AI company's out there in Israel in every second office, it's security AI company. So it's huge. Yaser Najafi: But that's and that's where the orchestration comes in, right? And that's that's Ido Bronstein: Exactly. Yaser Najafi: how you are able to orchestrate the right path in order for the agent to actually make the right decisions. Right. Ido Bronstein: Yeah, I want to talk a bit about CoCo. First of all, because it was one of your biggest announcement in the Snowflake Summit when we last talked. But I think because it it represent more about the change in the industry. Because in a sense it's becoming like agents, CoCo is an example, becoming the new interface of how companies managing your data, their data. And in a sense, like Snowflake all the time talked about simplicity. So it really changed what we believe simple is in that manner. So how do you see CoCo and how do you think it's affecting your customer? Yaser Najafi: Yeah. like for those of you don't know, CoCo is Cortex Code. Now we've rebranded as CoCo. it's the fastest growing Snowflake product feature we've ever delivered. and it's truly revolutionary. and look, I'm a user of all the coding agents out there. but for The value that Cortex Code or CoCo has is really like the context. You know, back in I can't remember 2022? I remember Devin was just released and the first thing I wanted to do was connect Devin to my Snowflake account. And I had to go and ensure that the right policies were set in place. I had to go make sure that the right context was made available so that Devin wasn't just gonna go and start deleting databases or deleting tables. and that was back then that was revolutionary. Like that was mind-blowing what it could do. And then Cursor really came out to market and it made things better. It made things easier, but it wasn't fully complete. What CoCo does is it completely removes the burden of entry of working with data in Snowflake. Right. It's really good at writing Python, writing SQL, writing JavaScript, as all other coding agents are. It understands your data, right? it knows the schemas, it knows the semantic models, it knows the RBAC policies, it knows where to go find things. And so building a UI is easy. Like that's table stakes. Anyone can do that. But finding the right data to feed into that UI so that people are actually gonna use it, that's where Code changes the game. I always I always you know tell and they may have told you this is you know Building code today is easier than it ever has been in the history of humanity. But shipping code and making it building it in way where our customers or users actually use it is a completely different world. And in the world around not only just Snowflake, but where data is outside in and outside of Snowflake, CoCo does give you the ability to move faster. it's still getting better. I don't think that it solves all the problems for you know software development lifecycle. But it does allow me to plug data into what I'm trying to build. a lot of our customers have petabytes of data, right? Scanning through petabytes of data takes a long time. It takes a team of data engineers building the right data pipelines, the right orchestration. CoCo allows you to do that in a much more efficient manner. Ido Bronstein: Yeah, Yaser Najafi: I know you guys are are advid, you know, users of CoCo. You know, it's Ido Bronstein: Yeah. Yaser Najafi: you tell me. How has it Ido Bronstein: So. Yaser Najafi: changed the way that you're working with you know your data? Ido Bronstein: So I Like CoCo is an amazing product. I think it really, allows you to like access the data easily. And I think this is the most important thing here. you have Snowflake, you have data and it really changed what you need to think about when you want to pull this data and use that. And you stop thinking about pipelines. You stop thinking about orchestration Data Wells says which size of data.. just think about what you want to achieve and CoCo will do the rest in that manner. So I think like for this Snowflake environment is like, it's changed who can interact with data. And I think this is the most important thing. Yes, velocity is important. Yes, now I don't like as an engineer, can build faster. But for me, what's really amazing in that it's the ability to change the users of who interacts with the data directly without any obstruction layer of like data engineers that cleanse the data before that and taking a lot of time to getting the reliable data because CoCo can help you do that as someone that is much more business oriented. I would say that like people still need to be technical in that manner. It's still hard to someone that is non-technical at all. to use those tools. So you need to still to understand what the concept of SQL and like stuff like that. Otherwise you are losing yourself, but it is in the direction that matters. Yaser Najafi: Yeah, I won't name any names, but you know, we've we have people that are on our team who before CoCo didn't know what Snowflake did. They didn't know the value of Snowflake, mainly because they're not technical. They're these are not non, you know, technical users. And after CoCo they realized, hey, this is wow. I know what Snowflake is now. I don't necessarily know how to build, you know, schemas and tables and Snowflake objects, but now I understand where things are going. And I think you know, you mentioned the technical aspect of it. You still have to understand kind of what it's doing. the dangers of code of all coding agents is relying too much on the coding agent. Ido Bronstein: Exactly. Yaser Najafi: and so, you know. It what it's doing is it's allowing a lot of the individuals who are not technical understand the workflow of what it actually takes in order to build software. and that software could be I want to build you know an application. The software could be, "Hey, summarize this PDF for me." There's a lot of steps that go into place and I think Ido Bronstein: There's Yaser Najafi: Yeah, it's empowering different workflows and I think what it's doing is it's removing a lot of the complexity, but it's also allowing us to think differently about how we want to solve problems. Ido Bronstein: Yeah. And like you mentioned earlier, that you are a builder and some things that I really remember that you shared with me a couple of months ago is that you had a chance to use CoCo before everybody. And, it's not only that, like the Snowflake CEO asked everyone to use CoCo in every situation that he has before they released this to public. So I think first of all, this is amazing. It's the best way to get feedback and build a product. But what was the experience to see like CoCo evolve and like be one of the first like having access to this technology? Yaser Najafi: Yeah. I mean look when it first came out it was kind of a okay, I was already heavily using Cursor for a lot of what I was doing. when it came out it was this is a great additional feature. the level of which it has accelerated to being on par with other coding agents and platforms is just is a testament to how committed we are. Of making this a first class citizen as it relates to Snowflake and what how you interact with Snowflake. the ability for us to add additional models, I think like that rapid development has been fantastic. and all the additional hooks around skill sharing, creating plugins, creating automations, and all the other components of it, I think have been very, I mean, it's been I think that was what six, eight months ago, maybe. Not even maybe even less. My context of time is different nowadays. But but it used to be a world where it didn't run as a standalone application. You had to log into Snowflake. Now it's when I wake up in the morning, I've got automations that are running that summarize my day, that summarize my emails. It go through and looks at, hey, you know, here's what we're doing with Upriver. This is what we need to follow up on. And so it's allowed me to do a lot more. Now, having said that, the sky's the limit. I think what we're doing with Snowflake CoWork and bringing that experience to market, I think is gonna revolutionize how again how we interact with Snowflake. But the reality of it is that I always try to push it. I like to try to push the boundaries. what else can I do that I'm not doing? Ido Bronstein: Exactly. So what you want to do that you cannot do, like what is next level after CoCo in that manner? Yaser Najafi: you know, for you know having there's a lot of things that I'm not sure trying to, you know, you know what we have announced, what we haven't announced, but the ability to have you know cloud agents, right? I think that's one component that is coming. the ability to take CoCo with me. whether I'm not bound to my laptop. It's where I can, you know, access CoCo on my phone. Those are the things that are going to I'm always driving and so I have ideas that come into my mind in my car. And I'm like, Ido Bronstein: You're Yaser Najafi: I can't wait to get home to run this agent to do this. And so having the ability to not be tied down to my laptop is going to be revolutionary. And I think that's we're gonna see that in the next few months. but also the collaboration piece. One of the biggest things right now is, you know, you're you you're a CoCo user, I'm a CoCo user, we're all building great things. How can we collaborate in a shared marketplace? where, you know, you can pick up skills, how can I monetize on the things that I'm creating? so that collaboration that Snowflake was grounded on, we're gonna see that kind of evolve into more of a not necessarily a CoCo framework, but more of a agentic framework that CoCo was the delivery mechanism. Ido Bronstein: Yeah. So the next phase in a sense, like how different agents can collaborate and like, can share my agent with you, stuff like that. Yeah. And I think that like, there's some things that the market. did not like figure out like, I don't see like even Claude or any other coding agents still very isolated in that manner. Interesting. So yeah, for, yeah, not an easy problem to solve. Yaser Najafi: Yeah, It's for good reason, right? Ido Bronstein: You don't want the agent to be crazy with what they can do with each other. We saw what happened when you give them a social network. Yaser Najafi: Yeah. Yeah, you wanna guard and I think that's the biggest piece is that the fact that your data, the model routing, is all happening in you know, the secured governance of Snowflake, I think that resonates very, well. And so I think there is there's a little bit of additional trust that you currently have with CoCo is that things are not gonna be exposed. outside of what you're currently doing. So and enterprises absolutely need that. there's a reason why they're not able to use Claude code internally, right? Or use GPT internally. but they've Ido Bronstein: Yeah. Yeah. Yaser Najafi: grounded on Snowflake. Ido Bronstein: Yeah, amazing. So one last question. And as an investor and someone that is in the frontier of the data technology, what do you see as the biggest opportunity today for someone starting a startup now? Yaser Najafi: There's a lot of great marketing content out there. If you go look at Pitchbook or CB Insights, there's a lot of you know massive addressable markets that you can go after. I think the biggest thing around kind of what we're looking for is really still it's still grounded on the data foundation for AI. and so most enterprises that who are Trying to incorporate AI into their work streams are failing because it's not the model, it's the data, right? How do you get the data from its raw form, not necessarily to a gold standard, but really to a position where AI can action on it? And getting access to both structured and unstructured, building kind of a unified data. vision for AI agents is that's the biggest problem that we're facing today. And that's where you know if you look at you know all these models have trained on the entire history of data. How do you create synthetic data that's going to fuel the future models? so if your a startup out there And you're trying to solve problems, if you can solve the data problem, which is still existing today, and Ido Bronstein: Yeah, still existing. Yaser Najafi: and solve for how people are trying to get access to data and get data to a clean state or get data to a usable state, if you can solve that efficiently and cost effectively, we're willing to talk to you. Ido Bronstein: This is huge. Yaser Najafi: yeah. Ido Bronstein: And I like how you told everything about that without using the word context. Yaser Najafi: Yeah, yeah. Ido Bronstein: But agree with you, this is the end. This is still the challenge. How you create data reliable and accessible, structured, unstructured, now to agent. It's still a huge challenge. Like from semantic, quality, governance, still needs to be solved. Yaser Najafi: Yeah. Yeah. and now you're getting data in from all the various sources now. So you have, you know, documents, you have structured tables, you've got video, audio, right? LLMs aren't gonna be able solve all that, right? Semantic layers aren't gonna be able to solve a lot by themselves. how do you build the pipeline efficiently? I'm using pipeline because it's probably the most easiest thing to visualize. But that's really the challenge. and so there's a lot of startups that are trying to solve for it. I think there's gonna be a lot of them that are gonna pop up here in the next few months to year that are gonna do that securely. I think security is gonna be a big piece of it. So trusting the LLM is also the other thing Ido Bronstein: All right. That is a wrap. Huge thanks to you, Yaser. Thank you very much for joining us. Yaser Najafi: Yeah. Thank you for having me. Yeah, this Ido Bronstein: I really enjoyed the conversation, truly. And yeah, look forward to see you next time that I'm in the Bay. Yaser Najafi: Well, if you're in the Bay Area, you know we've got the Silicon Valley AI hub, which is a shared working space for startups. and it's a great place to kind of sit down, work and you know, get a cup of coffee and always enjoy these conversations every time you're in the Bay Area. Yeah. Ido Bronstein: Yeah, you know that I love this place. So for our audience, if you enjoy this, follow the Data Splash Podcast. We have got a lot more coming your way. Until next time, keep your data flowing and we will see you in the next stream. Thank you very much, Yaser. Yaser Najafi: Thank you.