Snowflake Summit 2026 Platform Keynote

What happens when ideas meet intelligence? Progress is accelerated. Creativity is amplified. Inspiration becomes impact. This is what collaboration sounds like.

Good morning, Snowflake Summit. That's better. Thank you for being with us today. Super, super excited to see you all. Many of you know me. It is true: Snowflake Summit is my favorite week of the year. I was running up a little bit earlier. It is my favorite time of the year. We get to reconnect. We get to share with you the innovations that we've been making, and we get to learn from one another. We get to see the cool things all of you are building with Snowflake. And I want to start with a quote today. This quote is older than I am, and that's a lot: 'Any sufficiently advanced technology is indistinguishable from magic.' And I've been saying this for years because when you see a great product, it's magical. And we all live in a time where this seems to be true almost every week in some new way. It's exciting times. AI is changing what is possible for all of us and is expanding the opportunities for all of us to innovate. And us at Snowflake are surely innovating. Hopefully all of you see the pace at which we're going, because we love building technology to help all of you be more productive and help your organizations be more successful. We do hundreds of launches every quarter, and we keep picking up speed. We innovate with AI. We're leveraging AI ourselves, but we're also innovating to help all of you get the benefit of AI. We have a number of Snowflake members, engineering team, product team here. Many of them are watching online. I want all of you to give it up for the Snowflake product development team. Thank you to everyone. And among all that innovation, two products have stood out in the last 6 to 12 months. Shrier talked about it. Ben mentioned Snowflake Intelligence and Snowflake Cortex Code. The instantiation of that control plane that helps all of you and everyone in your organization be more productive. When we introduced Cortex Code, very quickly many of you started saying 'Coco,' and you hear us even saying 'Coco.' So Denise, who is here, said, 'Ah, we should just be done with Cortex Code. How about we just call it Coco?' What do you all think? Good. So from here on, no more Cortex Code. It is officially Snowflake Coco. And throughout the rest of the conference, it's Coco. And then Snowflake Intelligence seemed to be standing out. So we're like, okay, the scope of Snowflake Intelligence is so much broader than when we started. It's changing how we work. So we're also renaming it to Snowflake Co-Work. I'm going to cover a ton of innovation, and I left a ton out. That's how much there is going on at the conference. So I want you to at least remember one thing when we're done: we want you to rethink what is possible. The art of the possible is changing. And I want you to know that Snowflake will help you empower everyone in your organization, everyone, to leverage AI to be more productive with the context of your company, with a peace of mind on governance and security. And our goal for the next hour or so is to showcase, share with you some of the latest innovations that we have, and give you evidence that we help you leverage AI again with security and compliance in mind. And we're going to do this in four acts. So let's get started.

So, it's no secret we all live in a very complex world, and complexity in any dimension you look at it, there's a lot going on. Data technologies, new systems, new models, AI keeps evolving. And what Benoit and Thierry set out to do, and we continue to do to this day, is focus on making the complex easy. When we talk about Snowflake ease of use, we truly mean it. We want all of you to focus on adding value to your organization, not wrestling with the technology. And in the last six-plus months, we've seen a massive inflection on ease of use. Anyone knows what has changed ease of use for Snowflake and for data management? Okay, someone said Coco. Maybe unsolicited advice: when you're not sure, the answer is Coco, and then you think about the question in any context. So yeah, it's true. Coco is truly changing how we think about Snowflake, the surface area, and how you're more productive and more agile at getting things done. And it's crazy that we all talk about Coco, which has been around for just over six months. And here you see a timeline of the evolution and the pace of improvement of Coco. Started with CLI and a Snowsight experience. Had knowledge about Snowflake, and we expanded it to Airflow, DBT, Spark, other concepts, MCP, ACP. We did an SDK, agent teams, and of course at Snowflake Summit, we will continue to show you improvements in Coco. So today at the conference, we're announcing Cloud Agents, almost in general availability. And what this lets you do is for Snowsight, you have a sandbox in the back that lets you run commands. So a lot of the power that you see in the CLI is now available in Snowsight. Similarly, we're introducing a sandbox for the local development environment for CLI. We're introducing automations and the ability to have autonomous agents through scheduled operations, through async APIs, and we're introducing a skill catalog that lets you share skills and plugins, discover them, and reuse. And that also works with Snowwork. And I want to acknowledge a number of partners that are leveraging Coco to help many of your organizations achieve results faster. If your company is here on the screen, thank you. If your company's not on this screen, let's get going. Coco is going to help you. And Coco has transformed the entire lifecycle of what we do. You've seen this schematic. I've shown it before. It's simplified data sources, processing, and consumption. And I'm going to walk you quickly left to right through some of the innovations that we're producing, and many of them come with just Coco at the forefront. Let's start with what we're doing on sources. A year ago here at Snowflake Summit, we introduced Snowflake OpenFlow, a way to manage service to do data integration, structured and unstructured data. We're launching at Summit an APIs and programmatic way, object model to program OpenFlow. Why do you think we did that? Coco. I told you, Coco is the answer. We're also, many of you told us you want private connectivity. We're introducing a data connectivity proxy to have private connectivity to OpenFlow, and we keep adding additional connectors. The Oracle connectors, GA, Viva, Shopify. Many of these are now part of OpenFlow. But sometimes you don't want to ingest data that already landed somewhere else. Sometimes you want to capture it up front. Many of you, I know from talking to many of you, have complemented Snowflake with a streaming solution, most commonly Kafka. And probably some of the names that are on the screen you say, 'Oh yeah, I deal with all of that.' So we said, you know what, we want to help you capture events upstream but not deal with all of this complexity. And that's what today I am very excited to introduce to all of you: Snowflake Data Stream. What is Snowflake Data Stream? A fully managed streaming service built directly into Snowflake. In true Snowflake fashion, it has a separation of storage and compute. It does zero-copy streaming, which lets you stream data to and from Snowflake with sub-second latency. Most important, it's Kafka wire compatible. So all your clients and applications can stream into Data Stream directly. And of course, we have unified streaming analytics where you can instantiate a topic into a Snowflake table. This will be in private preview shortly. But also, AI has changed how we think about migrations. We've now unified all of our migration efforts under this term: AI-powered migrations. And if you still have a number of legacy database platforms and you want to move on to Snowflake, Coco and the tools that we have help you go faster. If you have Spark workloads, you want to move them, Coco and AI help you move it faster. I just heard a story of someone took some RDD code, moved it to Snowflake, and it was like five times faster. And for those of you still waiting on when to move off of a Teradata system, we've introduced virtualization where you can move your workload, still be Teradata SQL and BTEQ compliant, move it, get the benefits of Snowflake, and then later on you convert whenever you want. But we also keep looking at how do we improve the processing phase of this lifecycle. And often times when we talk about unstructured data, the first answer that we think of is AI functions. This is probably one of the most common ways where many of you are leveraging AI in the context of data. We're also adding additional capabilities. For example, AI Complete now takes audio and video as inputs and lets you reason and think through all of those. The number of use cases, sentiment, classification, all of that keeps compounding. And today we're introducing the public preview of something called Cortex Function Studio that lets you create your own AI functions where you can specialize it. You can create a function, evaluate, and control what AI operations users of your platform do. One of the key things out of AI functions is AI Complete, which is how do you interface with models. And one of the first parameters, actually the first parameter, is a model. And our commitment to all of you, as I said yesterday, is to always have the latest and greatest models available in Snowflake. We want to make sure that you have choice when it comes to models. If tomorrow something else is better, our commitment is to bring it. And that's why I am really excited that we're bringing SpaceX's AI models available into Cortex. I don't know if you're following all the different leaderboards, benchmarks, evals out there, but these models are making really good progress and they're quite compelling both from a price and a performance perspective. We're very excited about the collaboration between Snowflake and SpaceX. And this went into private preview as of yesterday. We're also announcing a public preview of something we call Agentic Search. And I think of it as something super cool, which is the best of unstructured world and structured world. What this lets you do is ask questions via Snow Intelligence, Co-Work, or Cortex Agents, but ask questions that require precise analytical answers. Imagine saying, 'How many contracts are dated in 2025?' You see three examples also on the screen. And what Agentic Search lets you do is instead of doing RAG, which just gives you a top-k type of result, it will leverage AI functions, extract the information from the unstructured data, put it in structured form, run an analytical query, and give you precise analytical results from unstructured data. This is in public preview, very excited. Any of you ever had a Python file that you developed elsewhere and you want to bring it onto Snowflake and you want to just run it in Snowflake? If you've done it, I'll guarantee you you've encountered friction. You had to wrap it, copy-paste, put in a sort of procedure, permissions. So today we're also introducing Code Bundles, a simple concept: take code, Python or Java, and deploy it and run it in Snowflake directly from the file that you have. No wrappers, no copy-paste. You can just execute SQL directly or execute the code directly or schedule the operation. And as I mentioned, this is now in public preview. Some of you may say someone liked it. Good. You deserve a t-shirt. I don't have anything to give, but you deserve a t-shirt. Some of you may be thinking, 'Oh yeah, Snowflake never got into this ML thing and machine learning.' The reality is we have a full stack, offline, online, whatever you want to do with machine learning. Snowflake is there for you, and the results are amazing. Our training APIs are two times faster, three times cheaper than, let's call it, other platforms. And guess who has skills that makes all machine learning easier? Now you know, right? Coco. But we're also announcing new innovations for machine learning. Today we're introducing Cortex Training, which lets you customize and train foundation models and do fully managed experience for fine-tuning or reinforcement learning. We're also introducing extensions to VS Code and Cursor if that happens to be your preferred development tool. We're also announcing the GA of streaming features so you can do serving, feature serving, feature vectors in real time, as low latency as you need it, tens of milliseconds. And last but not least, we're helping you safely evaluate new models through the introduction of online A/B testing. We can do control experiments, diverse traffic, eval, and if you like it, deploy it. Now, in terms of transformation, how many of you have been wanting Snowflake to add a visual pipeline editor? No one. Okay, JB, we built it for JB and for someone that's a 'woo' back there. We're also introducing a new project type in Snowsight which we call it Snowsight Pipeline Builder, and what you see is exactly what you get. Yeah. Yeah. I'm sure you want it even though you didn't understand my words. Visual representation and editing of pipelines. You can make changes. You can bootstrap it from a notebook, from an ML package, and you can see errors, make changes. And this is in private preview. We'll roll it out soon as soon as we get some validation. And last piece on the consumption part of the lifecycle. Of course, the marquee way to consume data is Co-Work. And we'll talk more about it in a second, but for now, Streamlit is a framework that makes you so much more powerful relative to your organization. You can build beautiful data experiences. We have 1.7 million monthly active developers on Streamlit and gaining momentum, and we're announcing the general availability of Streamlit hosted in Snowflake through a new integration. It gives you workspaces integration, Git integration, runs in containers, faster, cheaper, and you can just build amazing, amazing experiences on data. But if you want a little bit more control, someone liked it. If you want more control, some of you said, 'Oh, Streamlit is good, but I want just my own code, my own React app.' So we are introducing a Snowflake App Runtime. It's in public preview at Summit. It lets you run Node.js, very soon Python, which means you can run a full React application. And once you've built it, the easiest way to deploy it is yes, Coco can help you. We also have a one-line command as part of the Snowflake CLI. You can just do 'snow app deploy' and we take care of every single detail. We're also introducing something we call Snap and Ask, which is a way to visually give context to Coco and ask questions about it. I'll show it to you in a second. All in all, we're trying to look at the entire lifecycle of data and with the help of AI, helping all of you be more productive. But some of you said, 'I also use AI in other form factors and I want to make sure I can get the value of Coco.' So we're introducing a number of new form factors for this: a new Coco plugin for Excel, an extension for VS Code, and in the marketplace of Cloud Code. And last but not least, what if you could have the power of Coco in the CLI with the usability and excitement of Coco in Snowsight? And that's why today we're introducing Coco for Desktop, generally available today. And I'm not going to say anything else. I want you all to see it. You want to see a demo?

So, okay. So, Dash is here behind me. He's going to show you some of the technology we just talked about. Take it away, Dash.

Awesome. Thank you, Christian. Thank you so much for being here. Who's excited to see some live demos? Come on, let's go. Awesome. So our demos today will focus on a fictional company called Snow Music. Now they operate live tour operations and also high traffic applications. Now my first demo is going to focus on data engineers and app developers. Who is excited even more now? Let's go. Come on. Awesome. So let's get started. So what you see here is Coco Desktop application. Here I can click on this button to build an app, create a skill, or I can also actually type a custom prompt for it to build out a plan for my application. Now in the interest of time, I have already built the application. So let's go ahead and look at the deployed application built by Coco. Here we go. So you will see there's four cards at the very top. This is a live streaming application. You will see there is some data coming in, and I want to have Coco fix this data pipeline for us. Okay. So I'm going to open this panel on the right-hand side, my right-hand side. I'm going to paste the prompt. All I'm asking it to do is basically see what's wrong and help me fix it using natural language. This is amazing. You will see the data is flowing in. Coco is trying to figure out what exactly is the problem here. You will look at existing tasks. Are they suspended? Are there any columns missing? If you look at my prompt, it's very generic. Now, while Coco is cooking, yes, I just made that up. While Coco is cooking, let's look at where the data is coming from. So data is being streamed live using Data Stream and also inserted into an Iceberg table. Now that data is joined with Salesforce using Salesforce zero-copy data connector. All of that happening right in front of us. And you'll see on the right-hand side it's figured out that there is device type missing in one of the tasks. It's already altered the task and it's trying to verify what else can it do to fix the pipeline for us. Now let's go ahead and also look at a couple of other things down below. You will see that there is a selector, if you will, for selecting different models. What I've chosen is auto, but depending on your use cases, you can select one from the dropdown. Now let's go ahead and see what Coco is cooking. It's asking me to run some of these commands. It's doing it so that destructive commands are not executed automatically. It requires user permission before it can move forward. So let's go ahead and see what else is going to run. While Coco is cooking, I'm going to show you where the data is coming from. This is a direct share that's streaming the data from Salesforce zero-copy connector. These are the tables that are being joined live: fan profiles and ticket history. These are coming from Salesforce zero-copy data connector. Now the task has executed successfully. I believe it's diagnosed and fixed the problem. We will see that the cards will actually flip green here in just a second. If not, I can always verify the data pipeline. Go ahead and click that and let's see what Coco does in addition to what we already asked it to do. Now, Christian mentioned that there's a cool feature called Snap and Ask. I'm going to show you next. But let's wait while Coco is cooking, and I will show you on the desktop app to see what the plan it's built out. Here you will see every single step is laid out as part of the plan using Coco Desktop. Now I can either go ahead and edit the plan or also click on build where it will start building the application very similar to what we just saw in Snowsight. Let's go back and identify lowest row count, no data loss. Let me go ahead and just refresh here really quick and see what it has done now. So there are two things in common between AI and Dash Decai: my last name has AI, and Dash and AI can both make mistakes. Okay, now I want to show you something really cool. Now I'm going to go ahead and say, 'Please fix the cards.' Now while Coco is cooking, I want to wrap up what we just saw. Okay, app developers and data engineers can use Coco from either Snowsight or also from Coco Desktop application. We saw live streaming events being populated along with Salesforce data, and we also saw how Coco does not run any destructive commands without user giving permission. I think that's all the time I have right now. Back to you, Christian.

I hope you think it's cool. Okay, so recap what we've seen so far. We are in the friction elimination business. If in anything you do, you encounter friction and you want us to help you, find any of us. We truly see all of that as opportunity to help all of you. Let's move on.

Daniela and Sridhar spoke yesterday about trust, and trust is probably one of the most important things in what we do. Your companies trust you, you trust Snowflake, and trust is something that takes some time to build. It expects consistency on a number of fronts: security, of course, consistency of performance, observability, cost governance, business continuity. So let's see what we're doing to continue to further that trust. And for us, trust and governance comes together through the Horizon Catalog. It's a built-in universal governance solution. And you're going to get tired of this, but there's a lot of skills in Coco that help you manage governance dramatically easier than you've been doing in the past. Today we're also introducing the concept of intent-driven governance, which is again AI and Coco driven. But you can say, 'Take all the PII data in my database and make sure that it is protected,' and this will trigger classification, find out what's PII data, create the right policies, and make sure that it continues to be well governed all the time. So you express the intent, we take care of the details. It's not just governance, it's also security. And in the era of agents, we want to make sure all of you can secure your agents and you have multiple levels of protection and of course built-in security so that you can make sure that you can sleep well at night. We introduced earlier in the year Horizon AI Guardrails, which is protections built into both Coco and Co-Work to prevent high-risk threats, jailbreaking or prompt injection type of attacks. And this is built-in, detecting zero-day attacks. It gives you some policy, some control, but we want to make sure that you all are protected. Today we're also introducing the concept of agent identity. So you can tell when some piece of code or some activity in Snowflake is happening under an agent context, and we give you a context function, what you see here. So for example, in a masking policy or in a row policy, you can say if it is an agentic context, maybe you give it less visibility or maybe you give it more, whatever you want to do. But we want to give you that control over what agents are doing. We're also introducing data movement policies. So you can say this data that has this tag shall not move to a stage, internal or external, or shall not be downloaded in the Snowsight UI. You are in control. Policies are to help you be in control of what you want AI and agents to do. We also have a detection package in Trust Center, in public preview now, that helps you monitor unusual data transfers because we all want to know: is data moving outside of my secure perimeter? A few months ago, we introduced the concept of backups as a way to create an immutable point-in-time snapshot of, I don't know, an object, a schema, a database. And we introduced the concept of a retention lock which makes it impossible to delete. Not even an account admin can delete it. Why is that useful? Part of a cyber resiliency strategy, part of an anti-ransomware strategy. Today we're also introducing multi-party approvals where for certain highly sensitive operations, you decide which ones can require a mandatory second person verification. So even a rogue insider admin or an agent trying to change something cannot make highly sensitive operations like disable MFA from everyone. It cannot be done without two administrators in the system agreeing on this. This is in private preview now. And we're adding a number of AI security checks to Trust Center to enable all of you to check for what we consider best practices in both the configuration of agents and the configuration of AI. But you may have heard from us and from others that what really makes AI work, especially in the enterprise context, is exactly that: understanding your data, understanding context. And that's why today we're introducing Horizon Context. Intelligence alone is not enough. And oftentimes what you really miss is that context. And what Horizon Context does is part of Horizon, it's not a different technology. It's just a part of Horizon Catalog that helps collect signals, enrich those signals, and make them available to Coco, Co-Work, or Cortex Agents for you to get more context and more semantic information. A lot of what we've been doing in semantic views continues to move forward and it's part of Horizon Context. We keep innovating. I think the team has done dozens of launches on improving the expressive power of semantic views. And we're also introducing a number of metadata connectors that become part of Horizon Context where you can get context from BI tools, data transformation tools, or other databases. All in all, Horizon is the place where we innovate to help you govern your data and govern what AI does. But I can talk about governance and trust a lot. What I think is way more visceral is if you hear it directly from one of our customers, from Thomson Reuters, probably one of the companies that has some of the most stringent regulatory requirements. Please join me in welcoming Kaitlyn Hufherty. Kaitlyn, welcome.

Thank you for being here. Welcome to Snowflake Summit.

Thank you for having me. Great to see everyone.

Okay, so correct and trust is a big deal for you. I think you're trying to build this fiduciary grade standard. Tell us more about it.

Absolutely. So at Thomson Reuters, we serve our customers: lawyers, tax, accounting, audit professionals. So as you can imagine, their name and reputation is on the line. They can't be wrong. So they carry professional liability in every decision that we make. And so we build our products and services to hit that bar of trust. There's an acronym I like to use. You're welcome to take it back with you: WIN. What's important now? And for us, I think about it as keep our data and our products safe, invest heavily in innovation and M&A activity, and just really make TR a fantastic and great place to work. Maybe like many of you, we're builders. So as you said, Christian, we build to a fiduciary standard. And by that, we mean there's the content, there's a commitment to data privacy and security, there's enlisting the help of all of our subject matter expertise, leveraging our expertise, and then there's making sure that our output is transparent and we can validate that piece of work. So Co-Counsel is our flagship AI capability, and we have more than a million professionals using it in their day-to-day work every day.

So some people think of governance as a constraint, something that holds you back, and you told me something cool that governance is an enabler. Tell us more.

That's right. That's right. Governance for us has enabled us to de-risk and accelerate our AI transformation both internally and externally. So this fiduciary grade standard that we deliver must be built on a trusted data foundation, and we've really leveraged our partnership with Snowflake and are very appreciative of that. Things like our centralized access controls, our curated content and data sets really enable us to build with speed and scale. One initiative that we've been focused on is our semantic intelligence. So for anyone else who's building out semantic, I'd love to talk. It's been an incredible opportunity to really take business intelligence across all of our data for the firm. We have more than 1,500 internal users across our finance and business communities, and they're using it for all our most critical business and financial decisions every day. And that's been an incredible initiative for us and a reflection of our partnership.

How does, tell us, how does Snowflake help you rolling out and leveraging AI throughout your organization?

Yes. So Snowflake's really given us what we've needed, which is the governed, consistent, curated data. The foundation on which we're building both our internal data and AI capabilities and products, and the way in which we ship our AI capabilities to market. So we've been able to pull across many different data sources. For us, it's Reuters news, business, product, marketing, financials, HR data. Really pull that together inclusive of structured and unstructured data with the right governance and access controls in place. And it's really enabled us to move forward with speed.

You said you're a builder and you're building some apps. Can you say more about those applications?

Yeah, absolutely. So 2026 is the year of customer obsession for us, and we're fortunate in that we have some longstanding, really loyal customers that have been with us for a long time. We also know it's an incredibly competitive environment out there. There's competitive threats at every turn, and so we have to do better to continue to deliver exceptional customer experience. One example is we've pulled together our product usage data along with customer sentiment, billing, collection, renewals. So we have this real holistic customer 360 experience. We're able to do alerting, campaigns, grow a digital business, and really activate that enriched customer experience. So it's been an incredible example of what we've been able to deliver for the business. And super briefly, Sridhar mentioned we want to help people put AI in production. Your AI is not a demo or a trial. It's production. Right.

Right. And maybe that's what's different than some is we've moved beyond pilot to production. This is finance-validated metrics embedded in our key workflows. And one of the ways in which we ensure that we meet this fiduciary grade standard is we have responsible AI. So every AI capability before it integrates into market, into product, and ships to market runs through this process. So it's another way in which we ensure responsible AI is threaded throughout our work.

Kaitlyn, thank you so much for being at Summit. Give it up for Kaitlyn.

Thank you. Thank you.

Last year at Snowflake Summit, we introduced Adaptive Compute as the next generation compute paradigm for Snowflake, where we figure out what is the right amount of resources and it varies depending on workloads. Since last year, two things have happened. One, we put a ton of performance innovation and enhancements into Adaptive Compute. You see on the screen some of the benefits you can expect. I've learned to always caveat that mileage varies, but roughly a mental model is Adaptive Compute is roughly twice as fast as the original warehouse, generation one warehouses. And the second thing is we're super excited about bringing Adaptive Compute now into general availability.

We also introduced not long ago Snowflake Postgres, the most popular open source database out there. It went general availability in February and we've been busy innovating. We added private link support, customer managed keys, really secure, because we want to make sure that a lot of the safety and capabilities that you know Snowflake for are available in Snowflake Postgres.

Earlier this year we introduced PG Lake, or we open sourced an extension that helps synchronize data from Postgres into an open, interoperable lake. We've now taken that extension, put it as part of the service, so there's a managed version of PGA, and this will be going GA later this year.

Today we're also introducing Postgres data mirroring. You can say Postgres is more powerful, you can move things with a lot more flexibility, but if you just want to mirror a table from Postgres into Snowflake, you just flip a switch, you turn on mirroring, you get Snowflake to do all the heavy lifting, the change data capture, the synchronizing on the other side, very low latency, and this is going to public preview.

We also introduced the concept of interactive workloads, which come to life with interactive tables running on interactive warehouses, and again, the innovation is not stopping here. We're changing the size of the cluster and key, which brings a massive performance boost for most use cases. We're doing pre-caching. And in case it's not clear that everything these days gets simpler and easier via Koko, we have a number of skills that helps you with optimization, clustering, key selection, and so on. But that was not enough.

Aided by AI, today we're very, very excited to share with you the introduction of a new interactive compiler, which is a new query compiler for Snowflake. If you are familiar with how we run queries, we first spend some time compiling, then spend some time executing. It's more memory efficient. And again, with disclaimers on absolute performance, but early workloads that we did with one of our largest customers shows a 40 times, 40x faster compile time, which roughly accelerates this customer's workload by 3 to 4x. The engineer building this told me with this interactive compiler, hopefully you'll never have to think about compile time in Snowflake again. Does that sound cool?

And because we want to share with you all these performance enhancements, Unistore is not left behind. We're introducing a massive new engine optimization that improves latency and throughput by roughly 8x. And you see a schematic on if you tried hybrid tables in the past, you should retry because it's gotten materially better. This is also in public preview now, and if you have not, this may be a good chance for you to go and try it. Yeah.

Also earlier this year, we completed the acquisition of Observe, a platform that combines logs, application performance monitoring, and infrastructure monitoring all into a Snowflake native solution. And it has a very competitive cost structure. And of course, we keep innovating, we keep making it better. Observe has now introduced a CLI. Anyone can guess why, even though the penguin is giving it away.

Koko. Okay, the front line knows Koko. So yes, we're introducing a much simpler experience for Observe for you to configure, for triage, investigate alerts, anything you need to do in observability, all through a Koko interface or a command line interface. And also we've done the work to support Observe on Snowflake iceberg tables.

Now, let me switch to something that creates or erodes trust and is important to all of you, and it's cost governance. I've said it many times. We do not want any of you spending money with Snowflake in any use case if you're not getting more value in return. And we're committed to giving you full visibility, controls, and optimizations to make sure that your spend is all efficient. AI cost controls. Many of you told us, make sure that everything you've done with budgets works and your wishes are commanded. AI works with budgets as you would expect. We're introducing per-user quotas. We're introducing the capability to do cost governance on a shared warehouse if you have different departments or different users. And we're introducing budget custom action so you can invoke a stored procedure or perform some activity when some threshold is reached. And last but not least in this section of trust, I want to talk about business continuity. I'll challenge anyone to say we have the best solution in the industry for business continuity. When there was an outage from the cloud providers last year, over 300 workloads failed over and there was nothing to see here. Business kept going. And today we're introducing the next generation of account replication. It uses logs and you see the numbers, it's roughly 20 times faster. It gives the opportunity to offer you an SLA backed RPO assurance, which is we stand behind the latency and the data gap on a failover because of the performance in hand that we've seen. Is that cool?

And this is a good time to get Dash to show you the next demo. Give it up for Dash.

Thank you, Christian. Okay, so before I move on to the second demo, I want to show you what Koko was cooking while I was gone. So, let's go ahead and see what happened to our pipeline that we asked it to fix before. So, let's go ahead and look at the screen again. You'll see that the cards have been flipped green. What do you guys think?

Pretty cool. And look at the root cause. It actually figured out that there was a device type, one of the columns that was missing, one of the tables, and it diagnosed and actually fixed it. Now, I want to show you one of the coolest things you're going to see today that even Christian mentioned earlier. Snap and ask. Okay. So, I'm going to go ahead and refresh this page and we'll see that there's going to be a chart hopefully here. Engagement chart. What I can do here is drag this section and click on explain. This is going to give Koko context into what I'm asking it and will give us deep insights into what exactly is causing this drop in engagement. Pretty amazing. Yes, it's pretty awesome.

One of the coolest things you're going to see today. Now, let's move on to demo two and let's have Koko cook for now. So, for that, I'm going to switch my tabs. This is cowork. Christian just introduced this. And what I'm going to show you here is how we can leverage data movement policy, masking policies, and agent identity right at the agent layer. So, as a tour ops employee, what I'm going to do is ask cowok to export or get me a list of VIPs with their contact information. Now, I'm not allowed to access this information. So even though the agent has the ability to look at the table, it's not going to actually be able to give me the values that I'm looking for. For example, contact information for each of the VIPs. We'll see when the data comes back, either it's going to be masked or it's going to say I can't provide that information. That is pretty amazing.

Now, how do you actually set up DMPS for example, data movement policy and agent identity? These are first class Snowflake objects just like as you would create a database or a table in your account. So here you'll see that I have some create statements and I can also actually look at all the DMP violations right here, or also as a security officer you can look at trust center where you have a holistic view of everything that's happening within this account through Horizon catalog and trust center. Here you'll see violations, manage scanners. Now these are constantly looking for things that it can help you either fix or remediate. Data security and AI security. This is where as a security officer you have a holistic view of who created which agents, what kind of security scanners have been enabled. And AI guardrails is one of the top things that security officer would look for. Basically, it's going to in runtime protect against prompt injection.

Thank you. Let's go back to cowok and see what it did. So here it said that I have all these columns in the data in the table, but it's not able to actually give me actual values because of the policies put in place. Now let me also ask to export the data. So as a tour ops employee I'm asking it to export the data to an external stage. Now again data movement policy will stop this data exploitation and we'll see that here in just a second. Even if I expand this, you look at the actual thinking behind this and then we'll see that it will run into a data movement policy down below that many records are in here. And there's a failure. Thank you. This is a real failure, not a user error or not an AI error. And here you'll see that it's being blocked by data movement policy. Pretty sweet.

Thank you. Okay, so let me wrap up. What we just saw is pretty amazing things that a security officer is able to do using either trust center and also apply these data movement policies, agent identity at the agent layer and that trickles down all the way to where your data lives in Snowflake. With that, I'm going to hand it back to Christian. Thank you.

Thank you, Dash. Okay, TLDR of this section. We work very hard to earn and maintain your trust and we will continue to innovate to make sure you get the best of AI but with security and governance at heart. Let's keep going.

If you think from the very beginning of Snowflake when we were like we want to tear down silos, a lot of silos have been created just because technology made us do it, not because anyone woke up and said, 'I feel like creating a new silo.' Right? So, we have a number of vectors of innovation to make sure that you get access to the data that you need, people in your organization get access to the data you need, and AI and agents get access to the data that they need. First in the topic of interoperability, we are as committed as anyone can be in making sure that nobody feels locked in with Snowflake and we are at the forefront of investing for and implementing the Apache Iceberg spec. We have the broadest implementation of the V3 spec and we're working to shape the V4 specification.

We've also taken the Iceberg REST catalog interfaces from Apache Polaris, folded them into the Horizon catalog, and I think we are the only vendor that has full bidirectional support and interop between catalogs and engines, allowing reads and writes to any data whether it's managed by the Snowflake Horizon catalog or by an external catalog.

And the other piece, we've also introduced Snowflake managed storage for Iceberg tables. It is at Summit generally available for both AWS and Azure and is coming shortly for GCP.

And we are also committed to the interoperability of semantics. That's why we led the creation of the open semantic interchange group. The number of companies that are signing up to say I want interoperable semantics keeps growing, and I want to make sure everyone knows we're committed to making sure that Snowflake is open and interoperable.

Now data sharing we introduced it in the market in 2018, a long time back. Half of you in the room roughly are using data sharing on a regular basis whether it's internally to your organization or across organizations. And we keep investing in what you can share and collaborate with Snowflake. Whether it's semantic views or agents or models, we want to make sure that collaboration is essential. There's no silos at the data, no silos on the business logic.

And I think we've addressed every single request that you've had for us over the years at Summit this year. First one, you may have had data listings and you said, 'But I want to make sure that they plug in easily to AI to Cortex agents' and your wishes are command. At this point, I'm not going to ask you again, but you know who is making it super easy to create a semantic view, create an agent, and update the listing or the data share to say my data is AI ready.

Yeah, someone liked it. When we first introduced sharing, I think a week later, many of you tried to do this, which is you took something that someone had shared to you and you try to reshare it. And those triangles with this little bubble saying, 'Ah, you cannot do that' happens. It's an interesting problem, but we're incredibly excited to say now and generally available you can freely reshare data with other people.

And the other thing that all of you were very clear with us, I want to share with someone that doesn't yet see the light on Snowflake, and we are introducing what we're calling open sharing. And we're leveraging Iceberg and the Iceberg REST catalog to share with those public platforms. What this lets you do is effectively take data, make it available to non-Snowflake consumers and through Iceberg and the Iceberg REST catalog APIs, they become consumers of that data. It's a way where you see two of the technology we have, sharing and interoperability, coming together and delivering a great story. This is in public preview now.

Okay. And then the other thing you all told us, 'Oh, but sharing is only between two parties and it's unidirectional.' So today we're happy to introduce multi-party collaboration. As the name implies, multiple parties can now collaborate in a single secure environment. What does this mean? You can have different roles in a single environment and you can say I may be someone that contributes data or I may be someone that just does analytics. This starts with our clean room technology but it has been built on the foundation of broader global collaboration. Anyone here from Netflix? I know that's a long shot. Anyone here watches Netflix? So I mentioned it because they are at the forefront of adopting this collaboration technology. They're building clean rooms, collaborating with a number of partners, it's super cool, and all of this is now generally available.

The other thing we've been doing and Freda mentioned it yesterday is zero copy partnerships. There are many application platforms that have important data for you but you want to unsilo it. We started with Salesforce. We've now announced a partnership with Workday with the data cloud. We talked, we're introducing at Snowflake IBM, what's on X data, so you can get mainframe and DB2 data to zero copy with Snowflake, and the same thing with Aviva Connect if you're using them for industrial data. One of the most requested partnerships we ever got from you was SAP, and we're obviously incredibly aligned and excited to say that our integration with them is generally available.

And last but not least, sometimes you just have data in other systems and you want to be able to query across from those systems. The key use case that we're focusing on enabling is getting Snowflake Co-work to give you all the power of Snowflake and Snowflake AI on data that may sit in Redshift or Postgres and other sources. And we're enabling the query across.

So, I want you to hear from another one of our customers, Under Armour. They're doing amazing things with Snowflake. And for this, I want you to watch the following video. Let's roll the video.

At Under Armour, we use the power of sports to expand every playing field. We are obsessed with athletes who strive for more in sports, in life, and in the world. And our mission is to inspire athletes with innovative performance and design solutions they can't live without. I am Patrick D. Roso. I am the chief data and AI officer at ONAM. My responsibility is to really deliver and drive an AI strategy to enable outcomes for the organization globally across the enterprise. The biggest challenge that always showed up was the fragmented data that had always existed. It wasn't until that we actually aligned as a strategy to bring that into one consolidated place so that we had a uniform view, a trusted view way of looking at data.

The biggest challenges that we faced were the data was unstructured and the attributing wasn't as consistent as it is now. In order to find those insights, you really had to do a lot of manual work with the data in order to find them. So we ended up spending more time pulling the data and less time available to actually use the data for the insights. You really kind of miss the opportunity when a leader or someone was trying to make a decision, their attention and insight is at that very point in time. So if you're missing that opportunity to have the right information, you really lose that momentum. We have a concept that we consider critical to partnership, what we call a monument. What Snowflake was fulfilling for us is this idea that I'm not locked in from a technology perspective, but I am able to grow and scale as technology evolves and grows. After Snowflake, we were able to bring data into our platform much easier. We had a lot of capabilities to enable traditional BI, advanced analytics, and also share data within our ecosystem at a fraction of the time. So once we got rid of the bottlenecks of really just finding the insights within the data, we were able to move faster with decision-making. We work in a business that's super fast-paced, especially within our direct to consumer business where things are happening sometimes every hour. What conversational AI has really unlocked for our leadership team is they're able to ask that first initial question and then be able to surface like the deeper insight we really should be researching to understand how to drive a decision or an action. Our ability to really innovate at the speed that the technology is moving. The next chapter is important.

So, I love what Under Armour is doing, and you saw Patrick in the video, but he also happens to be here at Snowflake Summit. So, why don't we all just welcome Patrick on stage. Patrick, come on.

Okay. Awesome to have you here.

Thank you for having me.

You and I had lunch at Summit a year ago, and you told me you were doing these crazy things with AI. So tell us a little about the initiatives that you're doing and you trust Snowflake for AI, right?

Definitely trust Snowflake with AI. We started, we've been part of the Snowflake family for seven years. I would say probably five, six years ago is when it transitioned from a transactional relationship to a true partnership and that's what made the difference. That partnership is what enabled us to go beyond the great technology that you all deliver but ultimately meeting us where we are as a full business. So that's what transitions to our partners, our functional business partners, and making sure that they trust what we're doing because we trust what you're doing. As we relate to what we've been doing with that, we've actually been enabling, you saw in the video, all the conversational AI with our functional partners. And there's a concept we call the signal strength. And that signal strength is when we've delivered data, it's gone through a rigorous process that is fully trusted. And if it's trusted, all the things associated with that have been taken care of. And that really starts the foundation of the data.

Yeah. So we are very committed to open and interoperability because we want data to be accessible. How does that architecture and those design choices that we make help you and play out for your teams at Under Armour?

The key thing with the interoperability is now that they trust and we've built a platform around our data platform and our data architecture. Our business partners have applications that they want to bring and engage with that data at an enterprise level and we don't ultimately own that ecosystem. So the fact that we're able to operate with them, leverage things like hybrid tables, dynamic tables, ensuring that we have TT TSS enabled, running at the speeds, the things that you were talking about earlier to make sure performance meets the business needs. Ultimately, that's the outcomes that we're all aiming for and shooting for.

And what does that enable you to create? What do you end up with?

So if you could imagine, we've built assistance with... So you saw the conversational assistant that was done yesterday with Seni. We have a version of that internally for ourselves where our teammates are able to chat and engage with our agent that we refer to as ADA to enable and answer questions that they haven't been able to answer quickly on their own by themselves.

So, one of the things that struck me last year was you told me, you know what, there's all these SaaS apps that you actually I can build something more personalized, more streamlined. How's that going? What's happening?