They Just Won $1M! See the AI Startups Rocking Snowflake Summit!

Hello, Snowflake Summit and Developer Day. We are back with another installment of the Snowflake Startup Challenge. Now in its sixth year and in partnership with the New York Stock Exchange, this competition continues to highlight and invest in the most innovative startups building on Snowflake. In a world that is quite literally changing right before our eyes, the increased adoption of AI in Gen AI have forever altered the business era and the companies who are able to properly monetize their data to drive meaningful business outcomes will be the ones to summit to new heights in the time of enterprise AI. That is what we are after. For the sixth snowflake startup challenge, which founder leads the team, has the vision, product roadmap, and gotom market strategy to build the next game-changing application on Snowflake? We've gone global to bring the AI data cloud ecosystem to light and have narrowed it down to our top three. Silicon Valley is waiting for the next big thing, and that is what today is all about. Picking a winner won't be easy, but our judges are ready. Benoad Dodgeville, a leading expert in parallel execution and self-tuning database systems, founded Snowflake [music] in 2012 and has served as president of products and scaled the AI data cloud to the powerhouse platform it is today. John Barat is a partner at Capital 1 Ventures where he leads the enterprise productivity practice. His work involves identifying, nurturing, and making strategic investments to drive industry innovation and accelerate Capital 1's partnerships with emerging companies. Denise Pearson, the architect of the Snowflake Startup Challenge and Snowflake CMO, has played a pivotal role in driving growth and scaling Snowflake's revenue from 1 million to billions with over 25 years of technology marketing experience at high- growth companies. And routing our judges out, we have John Herrick, chief product officer for the New York Stock Exchange, a part of Intercontinental Exchange, responsible [music] for product development and innovation across all areas of the exchanges business. John oversees the NYC's transactions, ETF listings, and market data businesses, driving competitive strategy and distribution. The time is now. Pitches have been refined. Talking points worked out. $1 million in investments from Snowflake Ventures is on the line. San Francisco, please make some noise. The Snowflake Startup Challenge grand finale starts right now. Please welcome to the stage your host, Snowflake news anchor, Ryan Green.

>> Hello, Snowflake developer day and welcome to the grand finale of THE SNOWFLAKE STARTUP CHALLENGE. HOW WE DOING? All right, I'm Ryan Green, Snowflakes news anchor, and I'll be your host for today's grand finale. And we are living in such a transformative moment in our history, where data and AI are no longer future concepts, but the driving force behind businesses innovate, operate, and compete every single day. That is what it's all about. What's once felt impossible is now happening in real time and organizations everywhere are racing to unlock the full potential of AI to transform customer experience, accelerate decision-m and redefine what's possible. That is what's on the line today. In the era of Agentic AI, the pace of innovation has never been faster. And the companies that lead with a data first strategy are truly the ones shaping the future. And that's exactly why we're all here today at Developer Day. Now, in its sixth year, the Snowflake Startup Challenge has grown into a global competition where visionary founders are building the next generation of AI and datadriven companies. And while the con competition continues to scale and evolve every year, one thing has remained consistent. Snowflake's commitment to finding bold ideas, exceptional teams in category defining innovation. And now, San Francisco, the moment has arrived. Our three incredible finalists are sitting right here. One stage, $1 million in investments from Snowflake is on the line in the opportunity to ring the iconic bell at the New York State New York Stock Exchange. So the question is this. Which team is ready to rise to the occasion and take home the grand prize? Mosone, do you think you can help us out with that? >> That is what I'm talking about. This

year's competition attracted groundbreaking startups from around the world. And after months of pitching and building and competing, it all comes down to these three finalists standing here today. So let's meet our final three. We have arrived. We have legend and we have Twine security. The stage is set. Our three finalists are ready and standing by. Now let's take a look at our judges who will be evaluating their decisions today. Now they will be looking on these key criteria. Overall business potential and market opportunity innovation and the uniqueness of what they built. How they leverage Snowflake's AI data cloud platform and the team as a whole. All right, Mosone. Clearly, I know our live audience is ready. And now it's time to meet our judges who I know are up to the challenge. First up, we have John Berg, partner at Capital 1 Ventures, San Francisco. Please make some noise. >> John, welcome on the stage. >> All right. Hello. >> To dive in from an investing standpoint, what's on your checklist here today? Well, you know, we're strategic investors, so we're looking for technologies that can help us innovate. Actually, you know, we were early investors in Snowflake and they helped us on our journey over uh 10 years ago to build, you know, the first all cloud financial institution. We knew we couldn't do it alone. There's always startups that can help us get there. And so looking for innovation, you know, we're looking at team, we're looking at the, you know, market opportunity. >> I love it, John. It's a pleasure to have you here. Okay, next up, joining all the way from Wall Street, back for his second year as a judge, we have the chief product officer of the New York Stock Exchange and a dear friend. Give it up for John Herrick. John, great to have you back. For over 200 years, the New York Stock Exchange capital markets have been the envy of the world. What's top of mind today? Well, first thanks for having me back and uh it's it's a pleasure uh the New York Stock Exchange gets to host this amazing event with Snowflake. Um I think uh as as you said in my introduction, looking after the transactions business and the the the breadth of of of technical sophistication that our markets represent today. I think I'm going to be looking for how do these innovators, how are they thinking about building products that can help industries that are fairly mature uh comply with governance requirements, with resiliency requirements while really helping uh you know bring things to a new new level. And then the second thing which is always important for me, I look for team chemistry. And I think a team that is greater than the sum of its parts um that really helps uh bring the product forward is is a real motivator to me. So uh I wish you all luck and thank you John. Welcome back. Thank you so much. And next we have the one and only CMO of Snowflake, Denise Pearson. Denise, welcome back. The competition summit has never been bigger. What excites you most about today? >> I mean, it's so exciting to be back here. This is truly the highlight of the week and I'm going to be looking at what is the size of the problem these companies are solving and do they have the passion and grit to go after it. >> Well said. Thank you, Denise. >> And last, but certainly not least, we have the ultimate visionary, a builder in Snowflake's own co-founder, Benois Dodge. >> Benois, welcome back. From a founders's perspective, what's top of mind >> for me? It's always, you know, innovation, how of course, you know, these startups are leveraging our platform and all the features we are building and the team. I would say these are the three aspects. >> I love it. Thank you, Benis. >> All right, San Francisco, you ready to meet our first contender? >> Yeah. >> All right. Enterprises sometimes

struggle to operationalize AI because they lack the necessary talent, infrastructure, and governance to build and manage autonomous reasoning models leading to fragmented and poorly overseen AI stacks. Well, go no further because Arrived provides Agentic OS, an enterprisegrade operating system that allows organizations to build, fine-tune, orchestrate, and govern autonomous AI agents across cyber security, IT, and business operations without needing specialized AI staff from Delin, California. We have arrived. >> Arrived is about destination reached. AI shouldn't be a science project. >> It should deliver outcomes. But today, enterprises are stuck spending millions on AI talent, running pilots that never scale and stitching point products that don't talk to each other. They don't have an AI problem. They have an execution problem. That's why we build. >> Hey, I am AJ. My name is Anor Got. >> Hi, my name is Chira Kacha. >> Hi, I'm Leoshi Karaja. >> I'm Nikki J. I am >> Hey, this is Ashwin Arch. A's vision is to democratize AI and make it accessible to every organization in the world. As a Snowflake customer, you already are on one of the most richest data foundations in the world. >> Our platform helps you consume that data and basically personalize language models, build systems that can think and act on that data. It's an open canvas on which your creativity and imagination can solve many business problem. >> Your data on the snowflake is ready. Your team is ready. The only thing missing is the AI infrastructure. And now it's not. This isn't another AI tool. This is how AI works in your enterprise. >> All right. And please welcome Arrive to the stage. >> Good morning, Snowflake. How's everyone doing? Thank you so much judges for this great opportunity. We are super excited. I know my wife is watching so I'll tell you a story. Every time I start my startup journey, I tell my wife, "Trust me, honey, it's going to be the last one." It actually worked six times, but this time when I told her, she looked at me and smiled. So, I looked at back at her and asked like, "What's up? What's going on?" So, she said, "You know what? I'm thinking about starting my own company." So, I was very excited. I thought maybe my entrepreneurship eventually rubbed on in a on her in a nice way. So I asked her what is your startup about thinking it's going to be agentic agentic but she said no it's going to be husband 2.0. So having said that I think it's going to be my last one and I'm going to put my full effort in it. So my name is Anoru. I'm the co-founder and CEO of arrived and along with me is AJ Singh co-founder and CTO. arrived. As you know, AI is out there and it's making everybody extremely productive. But it's not just about productivity. AI is actually transforming people to become creators, builders, and operators. Technologies like Suno, Sora, Midjourney, Firefly. We're translating intelligence into creation. And we believe that every organization will go through that transformation. But today businesses are struggling. We know every business decision will run autonomous. The question is when. But today organizations are trying to scout for AI talent and struggling. They're trying to leverage fragmented tools and technologies and trying to solve business problems and they are failing. Our vision is extremely simple. We want to make sure every business user is an AI expert. Just like how spreadsheets commoditized and democratized analytics or cloud democratized infrastructure arrived is democratizing AI. Our goal is every business user should be able to build, deploy and operate agents. Snowflake offers the richest technology, the most advanced AI data cloud. It offers technology and capabilities to us from core foundation to developer platform to intelligence using cortex. We sit on top of those technology techstack. We offer reasoning and an execution layer. Together snowflake and arrived transform businesses from data to intelligence to execution. We are a very young company but we are deployed globally in some of the largest organizations in the world ranging from insurance to restaurants to fintech companies to multiple banks and large telecom providers. But that's not all. We are building that operating system on which production grade highfidelity apps would run and customers will be able to create them as well. There are organizations, really large organizations that are managing their firewalls using our agents. The agents are looking at shadow rules, looking for compliance gaps and even offering the citizens to self-s serve the firewall rule management. They are organizations that are looking for entitlements and assessing the risk of each and every entitlement in their organization and then making a decision whether that entitlement should be maintained or revoked. We just went into production with one of the fintech companies and now just we are scaling it for a very large healthcare organization which has to manage 6 million entitlements to organizations. We powered the first sovereign cloud in a country which now is offering government and enterprise customers the ability to fine-tune models, build deep freezing systems and run agentic orchestrations with ease without the talent gap that they're struggling with to organizations where they want to discover shadow AI and they want to apply guardrails not only to every citizen but also to every developer who's actually using C claw code. to restaurants where the margins are super thin and they don't have an option other than to run their IT and security operations autonomous. We are building that operating system that is powering all these apps. Snowflake has created AI data cloud. We are on a journey. We are building something. We are building enterprise autonomous execution system. We believe enterprise software today is just not about presenting the information. It should able to decide, reason and act. And that's what the arrive platform is. Before I move on, Unra mentioned husband 2.0. So that reminds me, we are on a journey to builder 2.0. We did build 1.0 7 years back. The NLP product we build when we met first time in a startup the same founder same founding mindset builder mindset is today together and they are building something bigger something larger with a big vision thank you that's what arrived is thank you >> thank you >> thank you [applause] [applause]

>> all right judges any questions for arrived >> sure John lead us off >> happy to kick us off all right Um so uh the applications or the use cases you're going after are very broad across an entire enterprise which I think is important but as a a young company entering a big enterprise I'm thinking about at Capital One we have different buyers in cyber different buyers in IT different buyers in business. How do you how do you think about the strategy to land within an enterprise and expand or how do you think about those different kind of ICPS that you would encounter? So I'll I'll break this into two parts because everybody when we started this journey everybody wanted us to build a point product solutions and we had built seven point product solutions in a past journey. So we said we want want to go with a platform strategy. So I used a very simple analogy to explain to my team on how we going to accomplish it. We said think about a problem statement. Think about think it's like a toy and break that toy into small Lego pieces. So break that problem statement into foundational blocks that you can solve using first principle and create that technology where I can assemble those small Lego pieces into and create a completely new toy. So that's how we are able to manage firewalls to do entitlements to to shadow AI to do security operations. Now second question which is a really good question. Every account we go in as as soon as the CIO and the CISO lands onto our vision. They understand what our vision is. They immediately become very creative. They said you know what I think I can solve this problem and I think I can solve this problem and then it boils down to prove it to me that your technology works. So the beauty is that we start with multiple teams at the same time. The one multi team, the IT team, the sec ops team and so on and so forth and we win. The advantage is that now we become very sticky because now there are three different teams on our platform and our vision is to make them AI experts. So our stickiness just increases within that organization. >> Thank you. >> Thank you. >> Thank you. >> Thank you. So, uh, kind of following up on that thread, how would you describe, you know, you're pitching a a prospective customer and they ask you, okay, how are you going to differentiate what you can bring to the table in customized implementations versus other uh, larger scale enterprise solutions that also, let's say, could integrate with someone's data through Snowflake or what have you. Do you feel like that leans more into some of the firewall or entitlement use cases that a CISO or a CIO might might be really uh um connected to or do you find there are other areas that that >> Yes. So uh that happens pretty much all the time. Uh the typical response to us is we are an agent native company. We started our journey seven years ago building NLP text stack when transformers technology didn't exist. So we have some foundation that we are coming up with. There are a lot of technologies where you can overlay summarization technologies using GPTs and other models. We are not that. We are a foundational technology. We give you those Lego block creating capability which you can assemble together to consume apps or produce apps. So that's how we try and position ourselves. Thank you. So how does your business model um you have a land and expand strategy right? How does a business model support that? Is this a consumption based pricing model or how does this look like? >> We offer two business models to a customer. We offer pure consumptionbased model just like you have with cloud and your snowflake technology um and Gemini and so on and so forth. But I personally tell the customer do not go with consumptionbased model. You want to do a fixed enterprise licensing. Now, I'm taking a risk of reducing my margins because I can go negative. But there's a reason for that because I say if you go with a fixed price, you know what's your cost, but your creativity is not going to get limited by the consumption of tokens. I don't want you to have a heart attack like oh I'm burning through tokens left and right and now your creativity is stopping. So I'm as a startup taking that risk and saying all you can eat, no restrictions. Go as creative as you can as you can add any many as many teams as you can solve as many uses as you can at a fixed price but the size of the organization determines what the price point would be >> apart apart from AI what what are the the main feature of snowflakes that you leverage and and what is your secret source in in the technology where would you say these are IP >> and we are have a road map of incorporating your finetuning capabilities as well into our into our system. U so those are the core fundamentals. Now your question is where is the edge? Uh as you know the world is moving towards context graphs. That's something that we are doubling down on. We are showing the early versions to our customers and they are getting super excited and they want to buy it but we're still in beta so we're not trying to sell it and it's very expensive. I mean I tell them it's a very expensive skew but it will show you results that you cannot even believe. We have uh um technologies equivalent of u where we built a deeping system. It's a point and click. You don't have to go with complex technologies like langraph to do it. You come into our system, you say I want these agents which have these skills. I need these actions and all of a sudden you have a deep freezing system. Even our fine-tuning can actually learn from an agent and completely train a model which is a very unique concept that we offer. >> Okay. >> Okay. >> All right. Thank you so much for San Francisco. One more time. >> [applause] >> Okay, one down and two to go. So, let's

keep on moving. Earth data intelligence has relied on classical computer vision pixels in map tiles for two decades, leaving imagery trapped in vertically siloed legacy systems that are difficult for AI to parse. Legend is shifting the platform to geo embeddings by offering an open-source model agnostic foundation model called clay that makes various types of imagery searchable via natural language model with a dual headquarters in Berkeley and Philly. Please welcome Legend. Let me introduce you all to Legend. This is our overture. We believe you can't fix what you can't measure. I've never seen such a concentration of talent density. I can only imagine it's our shared propensity for working with people who blend kindness with intensity. This is the most incredible time to be a builder. All the inventions of the past have come together to make now a thriller of a moment full of potential. Watch us turn that energy into something kinetic and influential because being able to query the earth over space and time is essential to fixing a climate crisis that impacts all living beings and that brings up all sorts of feelings. The biggest of which is gratitude. for this team who bring their brilliance time and fortitude. See, back in the White House, I met my co-founders, Bruno and Dan. It's so awesome to work with people whose default is yes, [music] we can. And I want you all to know that I wrote this poem the same way Jeff reviews [music] our code. No LLM helped me impose this ode. We have brilliant minds like the feed [music] fan and Nate who make our embedding factories hum at exponential rates. And it's incredible. You can just now ask questions of the earth because Sophie, Sam, and Z can train a model at a clip like nobody's worth. A love bomb to Nome is high bar for excellence. And the butler's ability to organize everybody with benevolence. Sue spent a lifetime leading the government in geoan analysis. [music] While Bruno, Matt, and Constantine's AI research has sparked an industry catalyst. And when Kevin's not fly fishing, he's writing code that is miraculous. And we're all inspired by Dan in his NASA hat, he's miraculous. We make AI that understands [music] the earth to embed the world with resilience. But what's really going to blow your mind is when we make the earth understandable to artificial intelligence. And that's what gives us the opportunity to transcend beyond just investor dividends and take our place as legends. >> I absolutely love that video. Please welcome Legend to the stage. Hi everyone. So, you can tell uh how much I like words from that video, right? So, well, they say a picture is worth a thousand words. But what if that picture could tell you those thousand words? What if AI could look at a picture of the Earth the way a human does and explain what changed and then predict what might happen next like it's doing here from Masonei. That's the question that brought Legend to life. I'm Nat Manning, Legend CEO. And I'm Jeff Alrech, head of engineering. Now, you've heard of large language models. They don't do a great job of analyzing the Earth. Why? Because they're trained on language. It's right there in the words. At Legend, we build large Earth models trained on the 800 pabytes of pictures we've taken of this planet. Matt, I know you always wanted to give a TED talk, but this just isn't the right setting. This is Snowflake. These are engineers. Let me talk uh in their language. We didn't just build the model, right? We built the entire stack. Models, infrastructure, the agents, all powered by a deep integration into Snowflake across services like Snow Pipes, dynamic tables, and cortex search to name a few. And we built the application on top like discover here that you see on the screen. That full stack approach is our moat. We aren't just a wrapper. We built the whole thing. Here, for instance, you see an analysis of what's changed in Soma over the past few years. We make Earth understandable using AI in order to make Earth understandable to AI. Jeff, I I know the tech's cool. Uh, but here's the problem we're solving. Earth observation data is famously expensive, notoriously difficult to use, and almost entirely absent from the AI revolution we're currently living through. Don't believe me? Go ask your favorite LLM, where has there been deforestation in the Amazon this year? Or where should I build a data center in Texas? It'll give you an answer citing a report written by a human uh because it was trained on language, not on the earth. But the primary source data for all these questions already exists. It's sitting in billions of satellite images of Earth that are being captured every single day. We believe that satellite imagery has immense latent value and that legend unlocks that. We're building the geospatial co-pilot uh dramatically improving the ability uh to inform decision-m processes with geospatial data. Jeff, they they get it. They get how amazing the product is. But how about that TAM, right? Of the roughly 60 billion a year spent translating geospatial data into answers, twothirds of that goes to human services, not to data, not to software. That 40 billion is our target market and today it's growing at a 14% kagger and that doesn't even factor in what's going to happen with robotics, world models and agents. So let let's walk you through three real use cases because that's where our vision business and engineering all come together, right? Yeah, that's right. First, insurance and climate risk. So we're working with a major risk modeling company. They want to look at every wildfire in the United States, identify which properties burned and which didn't, and then answer why. Was it the roof material, a fire break? Well, Legend just lets them just ask ask those questions and get the answers, ultimately enabling them to better protect people from climate disasters. Second is intelligence. We're beginning to work with parts of the US government responsible for processing geospatial data. Um, these analysts are drowning in data. They're simply more imagery than they have time to process. Legend lets them query that long tail and analysts can ask questions like where are the military boats in the yellow and South China Seas? and Legend detects that change and then describes what it's seeing and writes up a quick report with monitoring alerts and notifications to follow. The third, imagine your travel AI agent filtering out hotels with active construction next door. Something that happened to me recently during my wife and my first trip away uh without our kids. It was a real bummer. Uh or imagine your AI real estate agent knowing you only want homes with walkable trails nearby and sidewalks out front. And when that analyst wants to track deforestation in the Amazon, they're working from the primary source data uh within their favorite AI tool, not like a secondhand report. And as such, we believe that in the not too distant future, our largest user bases won't be just humans. It'll be agents, robots, and other AI models. Because every AI that wants to understand the physical world is going to need what we're building. The AI revolution has already transformed how we understand text, video, and audio. And we think that Earth data is next. Legend's mission is to make the planet queryable across space and time the same way LLM did for text. And what we've learned is that people want answers, not pictures, not data, and not vector embeddings. Answers. If a picture is worth a thousand words, then each of those answers will be worth thousands of dollars. Thank you. Thank you. [applause]

>> All right, judges. A lot to unpack. What's top of mind? >> I can start. I mean, absolutely love the mission. I mean, you're solving a huge problem, great use cases. Where are you seeing some early traction? I don't know if you have any customers yet, but what what is the feedback where where you seeing those repeatable, you know, go to market uh motions that you can scale? Uh so like those use cases there uh we we separate them into commercial and then uh government uh can't talk about specifically some we're under NDAs but the there are customers uh in the insurance markets uh in asset monitoring so one's a railway that's that's all that it runs all around the United States and wants to be able to constantly monitor those assets uh some in early financial services intelligence that goes into the financial service sectors uh and then in government. I I think that's pretty straightforward. It's it's the intelligence has generally been the the largest market for Earth observation data and that's predominantly with the US government. >> Could you spend a little time talking about how images are sourced, the update frequency, like depending on the contract, how specific that might be to the per because I think that's something that uh you know is is going to impact the long-term viability of of what you're building. >> It's a it's a really good question. So, one of my uh you know that saying you can only have uh you can have something fast, cheap or good. You can only pick two. Uh in imagery, you can have it re high recent uh high resolution or free. You can only pick two. Uh so this is one of our moes is that you're able to we legend can work on any type of imagery. Uh and that's really important because everybody else out there only ever sells analytics on just their imagery and it doesn't mean that that picture was taken. So we're able to kind of intelligently tell oh do you want to use uh this source planet do you want to use this or can you use a free source like Sentinel 2. Uh so since there's free imagery that is taken of the whole world every seven days uh but it's not not very high resolution and then you have high resolution that's usually done about once a year and then other if you want something else you have to pay uh and then we work with all those partners and and answer the analytics if you want. We do a and one of the ways we solve this is we do a strategy that's called tip and Q. It's widely deployed in the geospatial industry and what that means is you can start with freely available imagery to sort of get an overview of what you're looking at. Um embeddings from large earth models are particularly good at this and then once you identify for free essentially uh what areas are interested in looking you can then go and task higher resolution commercial satellite imagery to sort of doubleclick on that location. Great example here is something like infrastructure monitoring around ports, ships coming and going. Um maybe from the free resolution data you see that there was a particular amount of activity in the straight of Hormuz, let's say a certain week and you want to then go and task specific imagery to go and collect that. >> You should tell them how clay works on the electromagnetic field. >> Yeah. So the way this works is uh one of the models that we work is called clay and it is extremely unique out of large earth models because it actually understands the electromagnetic spectrum. All of these satellites that are sort of rotating around the Earth, all they're doing is bouncing wavelengths either off the sun and collecting them or they're active sensors that are setting their own wavelengths and bouncing back. Um, what the sensor sees in space is a function of the physics and the biology of the of the material on the ground that that wavelength is interacting with. Um many of these large earth models are trained specifically on planet imagery or a specific sensor. But the model that we build and maintain um actually understands physics in the electromagnetic spectrum in a way that other models don't which lets it translate to other imagery providers. Um and that's a huge part of our moat. >> We're the only model that does that. The only model that can work on every imagery today. And and on on that note though, are there um you know, I would I would imagine kind of just different weather patterns, atmospheric conditions could >> yeah could distort or hallucinate the model. >> We um we work with radar sensors that can penetrate through the clouds for example. So if we're uh deploying our solution into some and operationally into something like post- disaster response um where maybe uh like this railroad use case, they want to know if a tree fell over their rail line um as soon as a hurricane hits. Um our model is able to understand radar data that penetrates through the clouds. Um and so this ability to work across different sensors and modalities gives us more uh more ability to be flexible at the operational level when we actually deploy our solution to the real world which if you look at the broad scope of geospatial that's typically where things fall apart. Um you can make some incredible PowerPoint decks that show how you can use satellite imagery to inform decision-m questions. They typically fall apart when you deploy them operationally at scale. Um and we're able to help solve that problem because of our multimodality approach. Yeah, one of our secret sauces is the old our own agents in house that's able to route to the right imagery source uh for the for the question that's being asked. All built on Cortex. >> Built on Cortex. >> That's good. That's good. [laughter] Very good. >> Yeah. And and and and specifically we we were one of I believe the first uh users of the bring your own embedding aspect of Cortex Search. And so we've worked very closely with uh the Cortex team and the product managers and that's been a great experience for us. Yeah, that's been a game changer as we scale. Well, gentlemen, thank you so much. Give it up one more time. [applause]

All right, Mosone, we are down to our final competitor. How we doing out there? No, don't stop now. Come on. All right. Cyber security teams and CISOs are overwhelmed by a continuously expanding attack surface. Too many tools and a severe shortage of skilled staff. Twine security addresses this resource shortage by building AI digital employees. Starting with their first product, Alex. Alex handles identity and access management tasks end to end. The leading attack service in cyber security. Futural digital employees will cover SOC, vulnerability management, apps, and network security. Please welcome Time Security all the way from Israel. We have twine >> mini [music] >> action. >> action. [laughter] >> As co-founder of clarity, I spent the last eight years giving CISOs work, providing visibility and prioritizing where they need to focus. Today, I want [music] to flip the equation. No more tools that show, only action. My new venture twine is building AI employee who execute complex cyber task from A to Z. We're really bringing the future building a digital employee that is able to take on task. People are always saying AI will do the work. This is we're building the AI to do the work. So it's super cool. >> Working on cutting edge technology, multi- aent system is something that is is quite still unique. It's pretty new. >> Our work processes are very much working together with each other. I never really feel like I'm working in silos. I feel like I'm really able to collaborate with everyone in the team which is very very important to me and also the culture that the people [music] we became friends >> building something that is being developed in the world at the moment. There's no best practices. We are building the practices ourselves. >> We have the puff room over there. I love that you're directing me right now. [laughter] >> And the people that we have here are just amazing and you can learn from them so much. Our our standards for hiring dogs are not as high as they are for hiring people, though. [music] >> All right, please welcome to the stage, Twine Security. >> Hello. Hello. Um, yeah, I'll just start by saying that our standards for hiring dogs are not as high for for people because it doesn't matter. when you're petting a dog and when you're playing with them, the levels of dopamine increase in your brain and you basically become way more relaxed and focused. So, um that's really important for us. Uh Snow, thank you very much for having us. My name is Ival Carmel. I'm TWW security VP R&D. Um and I'm here to present Twine to you guys. So, at Twine, we're building AI digital employees uh for cyber security teams. What it means is that basically we are helping them and empowering them making them 10x more efficient. So as you all saw this is our team sitting on the floor next to the building we work at in the middle of the day. Um but yeah basically it's just a bunch of uh amazing people all with amazing like in huge passion for building. Um and uh I think the most important part is that our human to dog ratio is like 7 to1 which is really important and we tend to keep it that way. Um cool but yeah let's focus on the problem that we're solving. Um if you're not familiar with cyber security teams um so they are basically swamped with information all the time. They have so many tools that are detecting issues um and gaps in their security of their organization and um they are pretty old school. Most of them are still working with active directory to manage their users. Some of them even have like multiple active directories to manage their users. It just so messy sometimes. So automation basically doesn't work there. It's really hard uh to automate something that is just all over the place. Um and uh each each and every one of these organizations they have their own policies and their own custom processes uh to solve their security gaps um and to make sure that their organization is as secure as they expect. So what we do at wine we use agentic and we bring agent AI to these teams to these old school teams um to build AI employees that complete the work end to end. So at wine we started with Alex who is our identity management expert but we are building a whole family of them and the main uh I think the most important part here is that snowflake is the backbone of these agents. We are using snowflake as the data lake the govern data lake for these agents. um all the organizational information is there, all the knowledge is there and every decision Alex is making and every action Alex is taking is backed by snowflake information and by data in snowflake. So the idea here is that Alex can act with evidence all the time and he understands the full picture of the organization. There are so many use cases that we are solving for these enterprises. Um starting with user access reviews um with overprivileged with policy drifts but I don't have time to talk about all of them. So I'll talk about a really interesting use case which is ticket resolution. So cyber security teams um they serve the whole organization. All right, like people from all around the organization are sending security related requests to this team and they're basically being swamped with with tickets all the time like this one right here. Um, so let's look at this example for instance. I'm getting a ticket with the with with the following uh request. Hey, can someone please give my Levi from finance access to Salesforce same as everybody else? as a human being getting this ticket. I'm quite frustrated because I don't know who my lia is. I'm have like 100k employees in this organization. Maybe there are several myivis and I don't know which kind of access she needs to sales force. What are the norms for the finance team? It's basically takes like something like two days to resolve something like that because I need to get some clarifying questions. I need to get some more answers and I need to maybe research some past tickets and understand how we used to do that in the past. Um maybe I need to I don't know um search like policy documents of the organization. It can take a while. So what we do at wine is that once these tickets are being opened we are funneling them directly into snowflake uh snowflake streams in where um Alex as a codec cortex agent not codex cortex agent um is pulling these tickets uh using identity similarity search to figure out who my is looking at the diff the rest of the team the finance team to understand what's their access to Salesforce running semantic search with Cortex search to understand how we used to solve these tickets in the past and builds a resolution plan. Later on, Alex is taking this resolution plan in our system because it's connected to all of the apps of the organization and basically runs this resolution plan end to end. So the result, the outcome is the tickets are being resolved 10x faster, even quicker than that. It's like 30 minutes if you compare it to two days. So the people just have way more time to pet their dogs and play with them. Thank you very much. [applause] >> Okay, over to you.

>> Cheers. >> Yeah. Oh, well, I was thinking about kind of the the trust aspect. You know, giving giving away the keys to the Ferrari is a a uh concerning thing, especially for large enterprises like Capital One. How do you build that trust and enable uh our compliance teams to have the you know very strict uh auditing information that they need to understand what changes were made and how these agents are working. >> So it's a great question. It's one of the biggest challenges trust uh and we're building it with um like a long relationship with the with the customers. We start by um creating this control membrane um that wraps Alex all around. It's like a deterministic layer that makes sure that there's human in the loop that approves every action that he does until there the trust is built over time and they can basically tell us okay I think that specific actions like I don't know adding users to groups giving some entitlements to users I I trust you to do this this stuff so the granularity is really fine we're making sure that we are um getting approvals for autonomous fully autonomous um actions um and um and Yeah, it's like it takes a little bit of time but over time it works. And of course snowflakes also helps the idea that all the data is secure in snowflake and that we have guardrails in snowflake that I'll be able to talk about them as well. Um it helps us as well to to know that um the customer will trust us way more. >> Uh what do you see applications beyond cyber like are you are you envisioning other areas that that you think uh this could tackle? So um actually yeah but we we are focused in cyber security because these are the teams that we are familiar with and we want to solve their issues like we have been doing it for a long time and we know exactly what are their needs. Um but yeah the system that we are building is actually um something that generates uh like takes data and like high level metrics and gaps and basically helps you to run the mitigation for these gaps. So maybe you can you can try to implicate it on other different uh aspects but yeah we are focused with cyber security because this is the area that we know really really well. >> So where are you on your journey now? Do you have customers and how does your go to market strategy looking like? Do you have are you expanding here into North America right now? >> So uh we have several customers. We like a dozen customers already. um really big ones in healthcare and manufacturing um and we are starting like we are working with a few banks now already as well uh so the like the idea is to basically expand to as many verticals as possible um our go to market strategy is currently mainly in the US but also we have some IMIA customers um like in Israel so it's it's helpful as well um and the idea is to grow as much as possible more and more enterprise solve their issues Thanks. Thanks. >> What what are the biggest technical challenges that you have? I mean what one that you have and where you see that maybe Snowflake should do better in some areas or did you eat some limitation with the technology I guess? Um >> um so uh like um like technology related challenges or >> Yeah. Yeah. I mean you know the the platform is it you know do you have everything that you need or do you need more and and what are the challenges? I mean I I thought that the first question was very good right because you know if you start to take you know action especially in cyber security and and and it has to be scary uh uh so but I assume that having user in the loop is not that bad either because it's it's less you know the investigation can still be done by the agents. So maybe that's not a big issue but is there like something that that you are working on or some >> Yes. I can give you one example that is really interesting for enterprise is um like banking are usually really sensitive with their information. So they want everything to be under their control. So they don't really like the fact that all the data is in my snowflake. They want to control this snowflake. So what we do there is that we give them the ability to control the encryption keys of the snowflake account like with KMS in in a AWS and then >> they have the sense of uh security and control that they need. They can basically say all right I don't trust you anymore. I revoke all the keys and you don't have any data anymore. So um we're using like the triage of snowflake for that and we also deploy our system like in their own environment when it's needed. So, >> is it a native app as actually that's >> So, it's not a native app yet, but we are talking about doing that as well. Yeah. >> Okay, you should do that. [laughter] >> All right, give it up San Francisco. Thank you. [applause] >> All right, everyone. Pictures are done,

which means one thing. We're about to go into deliberation time. But before we do, why don't we do initial temperature check and go down the line? Ben Wall, what's top of mind? >> I mean, it's difficult as as as always. I really like the the geographic because the the theme is uh understanding earth is such a ambitious goal and I like ambition. So but you know um you know security is also you know very concrete and uh and and a very you know important problem to solve. So so there there's there's a lot >> it's all going to happen on this stage. The real decisions coming here. This isn't predetermined. Denise over to you. No, as Ben was said, it's going to be a a hard uh decision for us here uh on no panic and uh again three very different you know companies with a lot of passion also a lot of experience you know in their in their field this is not first time startup founders uh you do a lot of mistakes the first first round they have learned a lot you know in in the past uh all three are really focused on this kind of execution problem I think that's kind of the new thing for this year and that's what we're hearing from from customers, you know, as well. So, um yeah, it's going to be a tough conversation here for us. >> Tough. Um I feel like we were in the same place last year, too. >> I don't envy. >> I know. Uh I think it's uh it'll it'll come down to scale, too, like how the different ideas can scale in different >> different areas or even within the vertical that they've chosen. But I think all fantastic companies and you should all be proud. Well, as the new guy here on this stage, I'm I'm a little disappointed I only had like, you know, half hour to dig into these companies. I could spend all day talking to them because they're they're so interesting. There's so much opportunity. I think you know go to market is I mean clearly very innovative in their own own right and go to market is a very tough thing to crack to figure out how do you find that product market fit which will enable you know scalability and and uh and expansion and um those are those are some of the things top of mind there. Yeah. >> Well the time has come for the deliberations with that back of house tech cut their mics. This is a a a confidential conversation. I'm gonna go jump into the crowd, get the crowd going. You guys ready to go into deliberation mode? All right, you guys can stay on stage if you guys just want to huddle up right here. I thought it

might be interesting to talk to an individual who's very much overseeing the show. Harsha, head of ventures. We just saw three great pitches, but what's top of mind for you for this year's competition? >> Turn this way. >> Yeah. You know, top of mind would be just kind of looking at the caliber of companies that have come out in this challenge. I think every year we've been up here and I've probably had an inkling of like who might rise to the top. I literally have no clue what the judges are going to come up with all three. I mean from the the satellite like understanding the earth to the agentic platform and workflow across the enterprise cyber security I mean I don't know if I want to say it's a resurgence but like in the agentic world I mean these are truly enterprise problems. um the the breth and scale is there, the traction is there with these companies already. Um you know I I think this is going to be a very tough competition. The other thing that's really interesting and a key takeaway for me is the switch from even last year, right, where there were still a lot of the companies that were building really interesting analytic solutions, right? >> To the flip and the transformation of like these are AI powered solutions on top of Snowflake and to, you know, I've been Snowflake 9 years. to see that transformation at Snowflake to live it through the startup challenge, not just the three finalists, but the top 10 that I get involved with. It's been really amazing to see how they're leveraging truly AI native uh technologies on top of Snowflake. >> You know, I'm so glad you mentioned kind of the duration of this competition. Obviously, we're here watching watching the finale, but your team has been working with all the contestants over over the past better part of of 10 months. How has that process been throughout to pick the finalist here today? >> It's been well, for me, it's been a little bit easier process this year. We we've kind of consolidated uh the startup challenge startup program and snowflake ventures under one umbrella. So I haven't been as involved to be honest. I've gotten more involved in the top 10. And so I would say this year like immediately I was like holy cow these 10 are amazing, right? And the the level of AI integration and usage on Snowflake was super powerful. But just going through like our own scale of mentorship and coaching and engaging with these companies like that whole process working with the team doing the reviews it's been unbelievable you know this this feeds perfectly into my next question you get the opportunity to spend each day with founders visionaries you know what advice would you give to the audience of who you know they want to be on next stage over over the next couple years you know what key takeaways or or advice you'd like to deliver >> I'm going to give the same advice I always give customers customers customers like relentless focus on customers should be on the top of mind of every founder. I think there's a trap right now in the in the AI evolution or the new world of AI which is oh we can solve all sorts of amazing things with AI but don't just solve it because you can don't just because it's cool like there's a lot of I've talked to a lot of founders who are like oh >> I can I can do what you know 20 engineers could do >> para what should we build I have no clue like we need founders who have that domain expertise if you don't have it find and partner with someone who does right talk to customers every day keep that relentless focus Focus on customers and make sure you are solving a problem that has staying power, right? The biggest question VCs are asking right now is will this scale? Will this be disrupted by coding agents, right? Like you have to have that data mode, that domain mode. And if you're thinking about that, you're engaged with customers. That should be pretty straightforward. >> Well, Harsha, I don't envy our judges right now making this decision. What's next for Snowflake Ventures and the startup program collectively? So I kind of alluded to a little bit earlier the way we're bringing these teams together. You know, uh in the past we were somewhat independent but all working together. I think now that we're under one umbrella. I think I have a a strong goal AC as we do across a team which is how can we bring more value to these startups. >> Well said. So the startups coming into the challenge, the startups in our programs, the ventures investment, like these companies are all pipeline for us, right? How can we give them even more value, more coaching, more mentorship, uh really turning our broader team into a go to market engine too for these early stage companies? I think John mentioned the go to market is key. We have a huge customer channel, right? How do we help these younger companies unlock that a little bit earlier? So these are kind of things that are top of mind for us. Yeah, it'll be things like credits and mentorship and resources, but pulling that all together, potential even investments into, you know, bigger investments in some of these companies. How do we start to scale that is is what I think will be the next step, you know, and that's why I love that we have the finale part of Dev Day. We're very much surrounded by by the next chapter of of visionaries and entrepreneurs. Any final remarks here today while we wait for their big deliberation? >> No, I mean, I think you and I are trying to kill a little bit of time here because this is not that easy, right? I'm looking up for eye contact. >> Yeah, we're like they're not giving us a signal. No, I mean >> I think I've said kind of it. I I do think that the key takeaway here and I'm super excited to see the caliber of companies here. I think uh it's a testament to and I'm going to do the plug like what we've done on the Snowflake platform. We meaning Beno the team >> as a collective we I used to be a product manager at Snowflake [laughter] so I had some little hand in it years ago. Uh but I think that transformation from data warehouse to data platform to data and AI platform and the true AI control plane that Shredar is talking about that's what's most exciting to me here as as and I think companies should be pushing the boundaries with Snowflake and that what'll get our attention. >> Well Harsha it's always a pleasure. Thank you so much. It's an exciting next chapter for the venture program for the startup community and dev day. I I did this poll at the beginning of Dev Day, but show of hands, who is this their first summit attending here today. Hands up. Ah, that's what it's all about.

Looks like we have an answer here. Are you guys ready? All right. Well, we're making our way to stage. Deliberation is over. We're making eye contact. This is big. All right. So, first off, just big shout out to all of you for being here. One more time, Moscone. [applause] Now, before we dive into the big reveal, I thought it might be great to resummarize exactly what our judges are looking for when making this big decision. Remember, overall business potential and market opportunity, innovation, and the uniqueness of what they built, how they leverage the Snowflake AI data cloud platform, and the team as a whole. Benwis, we're going to do a little bit of a reveal here. [laughter] Okay, Okay, I'm going to drop down here. Is there a specific one? >> I have to say it. >> No, no, no. You're gonna pick and I'm gonna reveal. >> Shuffle. >> Shuffle. >> All right. >> All right. It has been decided. San Francisco, make some noise for legend. [applause] Gentlemen, please make your way to the stage. Benois, we have Snowflake the Bear. We have the big award. Gentlemen, please come to stage. [applause] >> Nice job. Congrats. Nice job. Awesome. Congrats, man. Great job. >> Awesome. Congrats. >> This is great. >> Great job. Congratulations. >> Great. Great. Great. Great. >> Benwis, if you would like to present the award, Benois Denise, to our winners here at Legend. >> Congratulations. Jump in here. We got Snowflake the Bear. San Francisco. Come on. Yes. Yes. Yes. Yes. >> Get in the middle, gentlemen. >> Did you go? All right. Couple big pictures.

Now, Denise Benois, but before we go, I want to just pick your brains real quick. Why did you decide to go with Legend? Beno, why don't we start with you? I mean I I think there were many aspects you know the ones that I like the most is innovation and the ambition right it's a it's such a broad theme you know understanding earth as a whole so I love I love the slice too I mean I love the the yeah the ambition and and the innovation and yes >> and Denise from your seat >> yeah again I mean congratulations to all the finalists here and uh it really came back to to the mission of of the company the really important, you know, problems that they can solve for for Earth and humanity. So, that's the reason. >> I love that for all of us. Congratulations and thank you all so much for joining us for the 2026 Snowflake Startup Challenge grand finale. We'll see you here next year. Go start building. The 2027 challenge will officially kick off in the fall at the New York Stock Exchange, but there's more. Be sure to check out the Silicon Valley AI Hub for the latest. Thanks again, everybody.