AI Pulse: April 2026

Santiago Giraldo: Hello, and welcome, everybody, to Snowflake's First AI Pulse. This is our monthly webinar where we share the latest and greatest in everything Snowflake AI. Just to start off, I want to start off with the safe harbor disclaimer. Everything that you see here is subject to change except for historical facts. I won't go into too much detail there. But as this world is constantly evolving, we need to make sure that that we're continuing to evolve with it. To start off, I wanted to just introduce our speakers. My name is Santiago Giraldo. I'm the senior director of product marketing for artificial intelligence. I'm joined by my colleague, William Allen, who is the head of product for Snowflake Intelligence and Agents, and Umesh Unnikrishnan, the head of product for Cortex Code. To level set on what we're gonna be talking about today, we're gonna be going through a couple deep dives across a certain subset of products, showcasing the latest innovations that we've implemented into both Snowflake Intelligence and Cortex Code. I'll start off with level setting a bit about what these products are, explaining a bit of what's new at a very high level, and then I'll hand it off to my colleagues who'll take us through a bit more of a deep dive into what's what's on the track right now, what's been released, and what's shortly coming. Following that, we'll do a live q and a. And and from there, we'll be able to answer any questions about the latest innovations and get you tuned in to the latest assets as well as we close out the session. So what's new? These are the coolest features hot off the press. And I wanted to first preface this with talking a little bit about our philosophy across Snowflake AI products. What we really focus on is enabling the entire lifecycle of your data to deliver AI as quickly as possible. And this means enabling your data teams, your technical teams with the latest tools possible to get the work done as effectively and as quickly as possible to better serve the organization. And then on the other side, it's about enabling business teams, business users, and line of business teams to be able to use that information, to be able to use that analysis, that machine learning, and those agents to quickly deliver results, make predictions, and solve problems for the business. So what are we showcasing today? First and foremost, I wanted to talk about the differences across Cortex Code and Snowflake Intelligence. Now, Cortex Code is one of our latest innovations. It's designed for developers, data teams, and analysts to really deliver solutions as quickly as possible in a prompt driven environment. What's unique about this is that it retains all of the security and governance, all of the context of your Snowflake infrastructure, and lets you use natural language to really deliver full fledged prototypes, applications, agents, and everything that can be used down the line, whether it be in agentic applications or Snowflake Intelligence. It also enables an interoperable data ecosystem, which we'll dive into today. This ecosystem, as we continue to expand it, is really designed to meet our customers where they are today, making it as easy as possible for them to leverage the best inference, the best model building, the best agent building, all from one place, no matter where their data resides. This all leads to what we've proven to see 10x faster data and developer workloads at scale, enabling workflows that used to take weeks or months to be accomplished in hours or days or sometimes even minutes. On the Snowflake Intelligence side, this is really designed for those business users and line of business teams that really need to turn that insight into decisions, into predictions, into understanding why certain things happen and what the next steps are that they should take. They can they essentially receive answers that they can trust. They're grounded and verified in the data with all the context and semantics that backs up that backs the Snowflake system so that they know that the decisions that they're making are well informed, and they don't have to worry about things like drift or hallucinations. In addition to that, because it's governed and extensible by design, teams can scale with confidence, bringing in different connectors from the marketplace, bringing in different agents to solve bespoke problems across business teams. So what are we launching today? We have a whole slew of features across Snowflake Intelligence and Cortex Code to talk about. First and foremost, in Snowflake Intelligence, we'll be showcasing artifacts, which really are advanced visualizations that enable you to do sharing, that enable you to do saving, and enable you to collaborate across these teams very effectively. We've also enhanced our deep research capability, which enables you to really dig deep into the data and really rationalize with that thought partner, that work agent, to solve critical problems and make decisions as quickly as possible. Expanding on our ecosystem, our MCP connectors are going GA soon, and this will broaden the landscape of how you connect with your data and how you interoperate across these systems. In addition to that, we're bringing innovations from things like Snow Work and Cortex Code directly into Snowflake Intelligence. This includes things like administrator skills and user skills that make repeatable tasks very easy to execute, and then things like agentic Text to SQL, which can generate very high accuracy results for your teams basically by leveraging all the capabilities underneath of the Cortex AI product. Last but not least, we'll be talking about the Cortex Code or sorry, the code sandbox inside of Snowflake Intelligence. And then also, we'll be teasing our soon coming mobile applications. On the Cortex Code side, we have a lot of new innovations that are coming into the fold. One is we're going to be talking about additional data system supports. This includes support for things like Databricks and AWS Glue and several others. We're also implementing MCP server as well as ACP. What this basically enables you to do is to have an agent protocol that makes it much easier for you to be able to build those agents and trust those agents as they hit your business teams. Beyond that, we'll be launching Cloud Agents, which enables a sandbox environment directly inside of the UI of Snowsight. And then last but not least, our agent SDK, which will make building and delivering agents much simpler than ever before. So without further ado, I'm going to hand it off to my colleague, William, to take us through the first part of Snowflake Intelligence. William?

Will Allen: Amazing. Santiago, thank you so much. What a great introduction. Will Allen here, lead product for Snowflake Intelligence and agents. Pumped to be talking to you all. We've got a lot of stuff to cover, so I'm gonna dive right in. Look. This is the reality that we're all living investing very heavily, very deeply into AI, but that last mile is still difficult, and it's difficult for a lot of reasons, some of which, we are trying to solve for you with some of these new products that are out there. Let's just imagine for a second what your typical day is like. And I have experienced this, and you all have experienced this as well. For a business user, when you don't have AI, you wake up first thing in the morning, I need to get data from Salesforce, from some ERP, from three other systems. Okay. So how do I go out and sort of get this right data from these different systems? Wait a second. It's ten thirty already. I've spent two hours on this thing, and none of the data matches. Okay. I need to find my data team. Who is my data team? Where is my data team? I file a ticket. I bug them on Slack. Nothing happens yet. Okay. Great. So I'm gonna give up on that. I'm gonna go to a dashboard. I look at the dashboard. It's showing me some stuff, but maybe it's a little bit out of date, and and it doesn't give me the why that I'm looking for. Okay. Fine. I'm gonna go out of the dashboard. I'm gonna build this thing manually. I go into Google Slides. I go into PowerPoint. I go back to Excel. I start building this thing. I finally have a time to present to my my leadership, my my CEO, my CTO, my CIO, and they say, not quite. This data doesn't look great. Can we see it cut by x, y, and z things? And I'm back to square one. The whole day ended. I had no decisions were made, and the insights were still pending. This was the unfortunate reality for so many folks across the board that we are trying to, trying to solve for. When you use Snowflake Intelligence, we cut down on all of this. We can help you go from data to decisions to action all in a single interface. It's built upon your data. It's built upon the same governance you've come to sort of know and trust from from Snowflake, and you can move immediately into the action part. So what used to take days or weeks can now be done almost instantly, and this can be done at the fingertips of everyone across the organization. That ability for anyone across the company to be able to access the important data that they have access to is a really critical part about Snowflake Intelligence. This is not theoretical. We've got incredible collaborators and companies like Fanatics that are out there running this in production today. They have what they call their FanGraph. This is a great view of what I think is the greatest ballpark in America for the San Francisco Giants. Fanatics is there thinking about sort of how to make an incredible fan experience. They're using Snowflake Intelligence to deeply power that work for all of their employees. How do they sort of understand the key insights and new monetization opportunities across their business? Thanks to Snowflake Intelligence. So how do we think about this in general? We wanna go with sort of data drives decision and action across the board. Deep analysis that can drive real action. So it's one place to get work done with AI. This is built for actually how work happens. You wanna automate things. You wanna visualize them. You wanna share those decisions, and all of this happens inside of your enterprise perimeter. The key things that we all look for that our IT admins and our CTO and CSOs really care about are all there. All of this sort of key RBAC and governance is baked in from the ground floor. Let's talk about a couple of the features that we're launching here. Two, on this first part that I'm particularly excited about, Deep Research and MCP connectors. So you're inside of Snowflake Intelligence on the web. And don't worry, I'm gonna go to a a live demo here in just a second. Sometimes you ask a question of your of your data, and it actually is is pretty deep. You need to sort of look across multiple, you know, tables, different parts of the data, and really have the AI go deeper across the board. We just launched Deep Research. It's coming to the public very soon, and this allows you to run multiple sub agents to get deep into the the data that you already have access to and to answer those questions that don't have an immediate, easy answer. You can use it to answer complex cross data questions, save lots and lots of time in manual analysis, and at the end, you get this beautiful sort of explainable report that you can then share with your colleagues across the board. When I show the demo of this in a minute, you'll really get to see the power of deep research. It's one of my favorite new features that we've launched. Next up, MCP. We've all heard of, the model context protocol. What this means for you inside of Snowflake Intelligence is that now you can bring all of your data from the different systems into a singular interface. So I can be inside of Snowflake Intelligence. I can query all of my governed data that sits inside of Snowflake, but then I need to cross reference it to what I have in Jira, what I have in Salesforce, conversations I'm having in Slack, one of my upcoming calendar meetings. We do make this incredibly simple for the end user, that business user, but, again, done in a way that the, you know, the governance and RBAC you've come to expect. It's got the governed RBAC. It's got sort of the, this, roles and grants that are are critical to it and full auditability logs. So you as the end user, the business user, can bring all of this context into your decision making almost instantly and take immediate action no matter where the data lives. A couple of the things we're launching, which I'm I'm also very pumped about, skills, artifacts, and our iOS app. Let's talk about those. Skills. These are things we we've sort of come to know and expect in a a couple different systems. We're bringing these to Snowflake Intelligence. These are workflows that we sort of do, again and again. For example, maybe I wake up every morning, and I have this, skill we call the good morning skill, and it sort of searches across whole host of different data and sort of helps me understand the key insights and actions that I need to take based upon what I am doing in my particular workflow. Maybe it's in sales. Maybe it's in marketing. Maybe it's in web analytics. Maybe it's in pharmaceutical development. I get to sort of build and customize the skills to match my particular workflow. These are then shareable across multiple teams, and it's a single prompt that can sort of trigger this workflow. It's a great new user experience inside of Snowflake Intelligence. The other thing, and this is a great example of I'm gonna show it to you and not just tell you about it. We're launching artifacts. You know, one great visualization is worth a thousand prompts. How can we see this beautiful sort of visualizations of your data in addition to the deeper context that you can get in combination with deep research and the conversations that are inside of there? You can visualize what's going on. You can save these. You can share these artifacts across your team and then continue to sort of build on and ask further questions about what's happening across your business. Most importantly, this doesn't isn't limited to your laptop or your desktop when you're sitting at your desk. We have a very soon to come iOS app that will be in public preview in, in the coming days. I'm incredibly pumped about this one. I have found it to be indispensable to my personal workflow. When I walk into meetings, I don't have my laptop open. I can immediately open up my phone, ask a couple key questions about the insights that I wanted to prep for for that next meeting, and get the answers right there on the go. It's a beautiful iOS app, built natively, very fast. You get all of the artifacts. And, again, it's sort of baked on sort of the same core governance layer that we have across the board. And lastly, the things that, again, you've come to sort of know and expect from Snowflake, we are designed for enterprise scale. A few of the things that that happen anytime you access your data, it's governed by the same role based access, the role level policies if your enterprise has them. We have really phenomenal budget controls, so you can see particular resources or teams or individuals, how they're using their tokens. It's really helpful for the token maximizers among us. Of course, bring your own identity provider, full auditability so you can see the sort of the full logs, and evals, with those logs to be able to understand what's happening across the interface. And if you wanna launch this just to, users so they can only have access to Snowflake Intelligence, we have SI only users as well. It's really important that across the board, there's no separate layer where data is being copied or moved or exposed. It's all baked into those same key controls, from there. We've got over nine thousand customers that are using Snowflake AI every single week. Some of my favorite companies. We talked about Fanatics before. We've got Toyota. We've got Whoop. We've got Decile, United Rentals. These are folks that are powering key parts of their business, to really think about how do you sort of operate, how do you bring sort of the key insights, of your company, and make, sort of bring those into action almost immediately. So I'm gonna switch, gears for one second and show my screen here and show you what this looks like inside of Snowflake Intelligence. So I have this is, you're seeing some real data here from a website that I run. It's called wills guitars dot com. I get to see these sort of beautiful artifacts, right, of the key things that I care about, the HTTP response codes for how my website's doing, what are the most popular user agents that are that are sort of coming to access my content? What's the cash distribution? What's the time to first bite? These are things that I would sort of use on a daily basis to understand how my website traffic is doing. This is geared and tuned for me in my personal world that can fit for you and your business. If you're in sales or marketing or finance, whatever it is, you can customize and curate these artifacts so they work really well for you. Let's take a quick look here to see what some of these skills might look like. So here's one that I've created. It's a weekly report. I can type in what happened this past week. And even you know, let's see if it still works even though I misspell things when I'm in there. It's gonna pull in this immediate, skill that I wanna use and run the key queries that I particularly care about. While this is running, you can actually see and sort of dive into here all of the important information that it's going through. It's actually running through. It's grabbing the SQL. It's running the queries. It's aggregating these things together, planning the next steps, all this done from a single prompt that I even, misspelled, the keywords on there. You can see I ran this earlier while this sort of, sort of comes through. You can see what this then, results in, this beautiful sort of quick ability that I can run the same skill any day of the week, and get this beautiful, sort of report on what happened. I see the key visitors over the past week. I get this nice sort of, again, artifact that shows what happens here, the geographic breakdown, the time to first bite, a really important thing for website metrics. How many of these visits are are actually humans or bots that are coming through? And these are things that I care about. I can then sort of go further. Even after I sort of run this particular skill and workflow, I can then sort of keep diving into this data and understand what's going on. So that's one aspect. Skills, are these repeatable workflows that I can sort of bring inside of Snowflake Intelligence anytime I need to. The other thing we mentioned was deep research. This is when you need to go you've got a particular thing that you're worried about, the thing you're you're trying to understand about your business, you wanna go a layer deeper. What I love about deep research is that it goes it spins off these sub agents. You can see where I I ran this before. I had selected this deep research toggle from inside of here. I asked this question, and it spun off these sub agents to go really deep and broad across my full data set to understand exactly what's happening with the business. And so what was the cash performance on my website? Pretty simple question. Look at this beautiful report I get. I get the key metrics right across the board. And not only that, I get this really helpful executive summary. What appeared to be at first sort of a critically low number actually comes with context and a summarization of what actually happened here that helps me again take sort of this insight and move it immediately into action. I get sort of the cache distribution. I can see this nice sort of ability to see, about the cache. I'd already saved this as an artifact. If I actually love this view and I wanna share with my colleagues, I copy it right here. I now have a link that I can share with anyone inside of my organization. When I share that artifact, it still gets the same RBAC, that's attached to it, you don't have to worry about data leakage. It's all inside of the Snowflake perimeter, And I get sort of more details here. I found this to be an incredibly helpful part of my workflow to, again, have this core data underlying all of these, various systems in one easy spot to be able to understand what's going on and take action on them. So I'm going to pause there. I'm going to stop sharing. We've got a lot more things to come. Like I said, the iOS app, we've got skills, MCP connectors, many fun things that that are going on there. Excited to talk about those in the q and a. But with that, I'm going to pass it back over to my colleague, Umesh, to talk about Cortex Code. Awesome.

Umesh Unnikrishnan: Thank you, Will. Alright. I am super excited to talk to you about some of the, innovations that we have launched and announced today in Cortex Code. Now, if you remember why we launched Cortex Code, it's because the agentic coding workflow usually doesn't have a lot of context about the data that is in your systems. Right? So what this means is that even though everyone wants to use AI for coding, there's a fragmented tool chain. There are lots of different tools that people are using. Iteration is slow, and then there's a lot of friction around governance. So, that's why we built and launched Cortex Code back in February as the solution for AI based development on Snowflake. Now, we've taken that to the next level now. And what we've seen over the last couple of months, the momentum that Cortex Code has right now has really blown our minds. Even we did not expect this level of adoption and this level of excitement from customers, users, developers over the past few months. We now have over one hundred and fifty thousand active users, end users for Cortex Code and more than half of Snowflake's customers' accounts are actually using Cortex Code. And they're not just using it, they're actually building really cool things with it. And we've gotten some quotes from some of our customers, whether they're developers, individual developers, large corporations, consultants. They've all found it super useful. They find it to be giving them like five to 10x the productivity gains that they would have without using it. So, if you've not used Cortex Code yet, this is the time to go try it out. You can enable it on your account. If you're an independent developer, you can go download our standalone trial and try it out as well. Now, with that, let me talk a little bit about how I think about Cortex Code and what it means to your data stack. The way I think about Cortex Code is that we don't want it to just be the best coding agent for Snowflake. We actually want it to be at the center of all of the data engineering that you're trying to do. And when we launched Cortex Code two months ago, it was really great at working with Snowflake. You could create your agents, you could create semantic views, you could create your pipelines, and all of those kinds of things worked really well. And the first steps we took towards that vision was making sure that it worked really well with some of the more popular data management tools, right? So dbt, if you have a dbt pipeline, have six hundred or so dbt pipelines internally and our data science team is using it to manage those pipelines and optimize them and monitor them daily. We've got that capability to all of our users by launching support for dbt two months ago. Similarly, Apache Airflow, again, super popular and we wanted to make sure we had support for that and you're now able to create Airflow DAGs with Cortex Code. Now we're taking the next step in that evolution and we're bringing in support for a few more popular data products. What this means is that if you are building Spark pipelines or if you're building dbt pipelines that also interact with another system like Databricks, for example, all you have a Postgres database that has all of your operational data in it, right? Now you can manage all of those systems also through Cortex Code. And no lakehouse federation, no data movement, no having to re explain your data model to some other coding agent, the same kind of zero intelligence zero setup, rather. Zero setup intelligence that Cortex Code brings to your Snowflake workflows now extends throughout your data stack. I'm super excited about this, and both AWS Glue and Databricks are now available in Cortex Code as a plug in. I think you need type in slash plug ins and enable them. And Postgres, there's an existing Postgres skill that you can just use out of the box. And let's do a quick demo of this in action. I think now the video will start playing. Alright. So, yeah, let's look at the AWS Glue skill in action. So I just told CoCo what I wanted. I had a couple of bucket files that are in an s three bucket, and I want to convert them into Iceberg tables that are managed by Snowflake. So I'm not running AWS commands here, no terminal switching, no AWS console, and just from this prompt, CoCo is able to identify the intent, route to the right skill, and then starts planning. And so you can see that you can query these Iceberg tables in Snowflake via a catalog linked database. So, I'll skip that step for now and just focus on getting those Iceberg tables into the Glue catalog. So this is where it gets interesting. What CoCo is doing now is that it actually does this and then laid out a five phase execution plan. And it knows that it needs to do the IAM roles, set all the policies, discover the crawler, and then set up all of the crawler configuration, all of those kinds of things. It knows that all of these things need to be done in AWS and it's just able to go and do those things. That's succeeded. It's verifying that the Iceberg table is now registered in AWS. And there, it's registered. And let's see if we can switch over to AWS and see if it's there. And there you see. All those tables are now in AWS Glue. Cool. Hopefully, that gave you an overview of what's possible with AWS. Similar cool things you can do with Databricks and Postgres as well. So the next big step for Cortex Code is that we wanted to evolve from just being a tool that you use at the command line to being more of an agent platform. And even over the last couple of months, we've heard of a lot of customers who've said, I really like that I can prompt Cortex Code and make it do things, but I just want to run it in the background. Or I have an existing application that my business users are using or my customers are using, and it does some Snowflake operations. And can I build or take all of CoCo and just bundle it as a tool within that application? So, for this, we've enabled two things starting today. First of all, a Cortex Code SDK is now available, so it is available as a programmable service. We've exposed the SDK in Python and TypeScript, And with that, you can embed Cortex Code into your existing workflows. So, if you have an application, if you have a workflow, you can now wrap it around Cortex Code and deploy it in a container somewhere. The next thing is we've always supported MCP clients, so you could always connect to Jira and GitHub and all of those other kinds of tools. But now, we're also enabling Cortex Code to be an MCP server. What this means is that it's an STD IO server and you can start sending it commands. Again, if you don't want to use the full fledged SDK, you can just use the MCP server and send it commands and build your workflows around Cortex Code. So that's on the agent platform side. The other big piece of feedback that we've gotten recently is that people love working in the terminal or they really like working in Snowsight where Cortex Code is already available. But a lot of people use Visual Studio. I used to work on Visual Studio a long time ago. It's a great editor. So what we're doing now is that the Snowflake Visual Studio extension will now have Cortex Code built into it. So, whether using Visual Studio or Cursor or any other VS Code based IDE, Cortex Code is now embedded on the right hand side and you can use it directly from there without using it in the terminal. You can, of course, also continue to use it in the terminal. We also support agent client protocol, which is a popular protocol for IDEs to talk to agents, coding agents. So, if you're using an editor, Neovim, some of the JetBrains IDEs, Emacs, any of those things that support ACP, you can now call Cortex Code from within that. And then finally, a lot of people use Claude Code and they're like, I really enjoy working with Claude Code. I use it for all of my coding tasks and I just wanted to delegate to Cortex Code when there's a Snowflake specific task or a data specific task. Can I do that? And yes, now that's possible. We are launching Claude Code plugin and that will be available in the next day or so in the Anthropic marketplace. And you can go install the Cortex Code plugin for Claude Code and it does two things. One, whenever you give it a Snowflake specific request, it automatically routes to Cortex Code. And secondly, you can explicitly invoke Cortex Code from within Claude Code so that you can delegate Snowflake or data specific tasks to that. That is on expanded availability and here's a quick demo that shows the VS Code extension. Let's see it in action. Here it is. It looks just like the Cortex Code that you know and love within Snowsight. So, you can just open it up, shows up on the right hand side or left hand side, whatever you prefer, and you can ask it to start doing your data tasks. And then you did a quick query and you want to now see those changes implemented according to plan. And of course, it opens the code in Visual Studio. So, now you can inspect it. And this is actually in a private preview right now. Think reach out to your Snowflake rep to get included in the private preview and test it out. Send us feedback and we'll continue to make it better. Alright. With that, I wanted to talk a little bit about some of the things that we are doing for the Cortex Code in Snowsight. So, Cortex Code has been in Snowsight for about a month now. It's been in preview for a little longer than that. And while it is extremely powerful while you're writing code within Snowsight, it suffers from some of the limitations that you would expect from a web based environment. It cannot call MCPs. It cannot install custom binaries, any of those things, right? So, we are announcing in preview right now cloud agents. What this means is that there's actually a sandbox runtime that's built into the Cortex Code in Snowsight that actually runs the Cortex Code CLI but within the same secure environment that Snowflake provides. So, what this means is that you can actually run fairly complex operations, background jobs, long running operations, all those kinds of things within the Cortex Code in Snowsight. And you can do some really powerful things that way. I'll give you a very simple demo here that just gives you a sense of that power. This is in an early preview. We think we'll have it ready to launch over the next couple of months, but I wanted to get this out to anyone who wants to try it out in a preview. So, here I'm asking it to create a Streamlit app that retrieves stock prices for any stock and show its trajectory. So what you'll see here is that you might have missed that, but it said it's provisioning a sandbox right now and it's actually doing a PIP install of the Yahoo Finance library within that sandbox environment. So this is essentially a full fledged container that's available within this environment so that you can now run pretty much anything you can do in the CLI. Again, simple demo that shows the concept of what's possible. You can see that it's actually running shell commands within the UI. It's pretty cool. And it's actually running a bash command to run the Streamlit app, and then you can connect to it. So we are super excited about this. It's both highly innovative and it opens up a lot of possibilities for you. I'm really excited to see what people do. A couple of other things, somewhat smaller changes that we're making in the Cortex Code in Snowsight. First of all, one of the biggest piece of feedback we've gotten over the last few months is that people love that Cortex Code in Snowsight does things right away, but they really wanted a plan mode, so we added that. Plan mode is now accessible to just a toggle. You can do that and you can get a plan and then ask Cortex Code to go run it. For those of you who are using Cortex Code for things beyond just writing code, you obviously know that you can also use it to manage your account, you can ask it to do cost management and things like that. Now, you have the ability to look at any screen in Snowsight. An example here is the cost management screen and it can read the chart that's on the left hand side and you can ask it questions about it. So, in this example, we're asking why are there some spikes and it's able to go in and look at the data that's underlying this chart and give you information based on that. And then finally, we are enabling the latest models from the foundation model providers and today we announced support for Claude Opus four point seven and for OpenAI GPT five point four. GPT five point two was already available, Opus four point six was already available and now both of these are available. They will not be the default to start with, especially with OPUS. We know there are some concerns around token costs and things like that. So we're making both of these available in preview and you can opt into that. And with that, again, super excited to announce all these things. I'm really hoping you'll go out and try these features, send us your feedback.

Santiago Giraldo: And I wanted to open it up to Q and A. Awesome. Thank you, Umesh. Thank you, Will. Lots of great innovations. I'm really excited to see how these products continue to evolve. Before we dive into Q and A, just a quick reminder that we do run these sessions every third Tuesday of every month. So please tune into these. This is where you're gonna find the latest and greatest across the Snowflake ecosystem, everything that we're doing with AI, and we're moving very, very quickly to make these products the best they can possibly be. And this this is gonna be one of your main sources of information moving forward. So make sure you mark your calendars. With that, I'm going to dive into a couple of questions here from the crowd. First question, how does Snowflake Intelligence handle cost governance? And I think that we can also apply this to Cortex Code. What visibility do I have into usage? And what controls can I put in place to manage spend? And Will, you want to kick us off with that one?

Will Allen: Yeah, it's a great question. And we have been rolling out a lot of features here. So the TLDR is you can look at it at multiple levels. So you can sort of think of it as a as a resource level, a workflow level, and then an individual level. And in each of those, you can look at sort of the, inside of Snowflake Intelligence, the the token spend in particular. So you get really sort of solid governance and auditability across each of those. So I could say, great, Santiago. You have a a budget of x, y, and z. And, Umesh, you have a a budget of a, b, c. And then start to build alerts on top of that from there. So resource level is in public preview now or will be in the next couple of days. And then individual level is coming up pretty soon. And if you have other feedback or things we should build in terms of that cost governance, let us know.

Santiago Giraldo: Awesome. Thank you, Rahul. Umesh, anything to add on the Cortex Code side?

Umesh Unnikrishnan: Yeah. I I think Cortex Code, you know, everything that Will said about AI governance, cost governance in general, also applies to Cortex Code. We support all of the same budgeting capabilities. The two additional things that we've added is one, there is now the ability to set limits on a per user basis. And these are hard cutoffs, these are not budgets, but you want to make sure that someone is not running a job that spikes your credit usage so you can actually run that. You can set a parameter at an account level so that you want to make sure no one is using more than ten credits a day, for example. And you can also build an exception so that there you have service accounts or you have some teams that are heavy users, you can set different limits for that. We just launched that last week, I think, and looking for feedback on how to make that better. The other piece is for Cortex Code. We usually default to the best available model, which means most of the times that's using an Opus four point six or something like that. That's a more expensive model. We are looking at making that more configurable, but at a user level and at an account level, you can always switch people to use a cheaper model. So you can use the model allow list to only allow Sonnet, for example, which is significantly cheaper.

Santiago Giraldo: Got it. Thank you, Umesh. Okay. Next question here I have is, can you explain deep research and how it's different from, the extended thinking and Instant Answers capabilities? When should we use each one?

Will Allen: Yeah. And we think of this as complementary, we actually have even more things coming there pretty soon. One of things we wanna do is to make it so you don't have to worry about choosing, which one of these. And so there's some fun work that we are doing to make it more automatic for you where we would suggest deep research or extended thinking. Deep research really thinks about sort of spreading out, sort of across data sources that you have, and extended thinking is higher accuracy, or sort of looping through to sort of verify across a narrower narrower set, of questions. So we think sort of almost like a a t shape. So deep research going wide, extended thinking sort of going deep. And but, today, you can choose these toggles sort of inside the the plus for the deep research and then hit the extended thinking inside of Snowflake Intelligence. We're doing a lot of work to make that user experience even better, so we would suggest it automatically for you.

Santiago Giraldo: Awesome. Looks like we've got a handful of other questions, but I wanted to dive into you mentioned the iOS app. Obviously, we mentioned it throughout the call. You provide a bit more context onto the capabilities of the iOS mobile app, when it's going to be available, and how folks on the call can get access?

Will Allen: Yeah. It'll be in public preview in days. So I would love for you all to check it out when it's in there. It's in the iOS store. I will show you a little bit of what it looks like really quickly. This is a video that I took earlier on my phone, and it's all the things you expect inside of Snowflake Intelligence. Right? It's I can ask questions. I can see these artifacts and sort of get the full experience and watch the full planning and have it on the go. So in the App Store, public preview in in the coming days, and you log in with your same sort of login that you'd log into Snowflake Intelligence sort of with your your IDP, whoever that might be. If you're not sure you've you have access, talk to your your account rep, your AE, and they can help you understand sort of next steps there by public preview in the coming days.

Santiago Giraldo: Awesome. Thank you all. Okay. A couple more questions here. This one right here, Umesh, is for you. How does a Cloud Agent govern how does Cloud Agent governance work? How is Snowflake governing the libraries, binaries that are installed in the Cortex Code Snowsight container?

Umesh Unnikrishnan: Yeah. It's a great question. This is not a persistent cloud agent. It is something that is very it is dynamic, and it can just go away after a while. We're still working through this. You know, that's why it's it's a this feature is not quite launched yet. We are working to figure out how we impose some governance on it and how we make sure that you control who opens it, who can use it and who cannot, and what can be installed. So this is still a work in progress. We'll have more to share when it's ready.

Santiago Giraldo: Awesome. Awesome. Thank you, Umesh. Okay. Let's see what other questions we have here. This one's a follow-up for you, Will. When spreading out across your data, does that include sources outside of Snowflake or particularly other database management systems?

Will Allen: Yeah. It's a good question. You can do that through the MCP connectors inside of there. So if you you have all of your you know, some core data inside of Snowflake, and then you say, great. Well, I also wanna connect my Slack or my Jira or my Gmail. You can connect those at sort of the agent level via MCP to query those as well. So then I could say, you know, who's my top customer? What products are they using? And do I have any meetings with them in the next thirty days? And if so, send me like, create a draft email to send to them about, I don't know, summarizing whatever we're gonna we're gonna talk about. So the key like, one way is through the MCP connectors. There are other sort of connections you can do at the agent level when you're building that agent across the board through sort of OpenFlow and which is another sort of great, you know, primitive across Snowflake. But a lot of what we're seeing is sort of interacting with, in particular, your your sort of data that you have to run your day to day life via MCP.

Santiago Giraldo: Awesome. Great answer. Okay. This question has come up a couple of times. What are the sharing capabilities of agents? Are they going to be integrated with MCP for things like Slack, MS Teams, other things like that?

Will Allen: Yeah. So for Cortex agents, you can access them, inside of Snowflake Intelligence, which is what we we sort of we saw here. Each of those things that I I accessed there, you saw probably a little thing that said Will's Guitars. That is, the agent that I built. That agent is governed. It's got monitoring. I can choose the model that it uses. I can choose the tools that it connects to. I can give it skills. I can connect it via MCP servers. So it really becomes an atomic unit that can answer, you know, particular questions, and then that gets exposed via Snowflake Intelligence or through the the surface that you choose. And so that can it's all accessible via the API or an MCP server connecting to that. So, again, sort of think of agents that can reach out via MCP connectors to different sources of data, and you can connect to that agent via an MCP or just directly via API through whatever surface you might have. So you might say, great. I want to build this into Teams or to Slack or an internal portal or into a customer website. All those things are possible. You just use the API slash MCP if you want to to be able to connect back to that, and it does that same sort of authentication layer. So that's really important. Even when you're accessing that data sort of outside of Snowflake Intelligence, if you're accessing the API, you have sort of the roles that gets passed through, and all of that RBAC sort of carries through with those answers.

Santiago Giraldo: Awesome. Excellent. Here's a question, that we have, and I think we have time for maybe one more. So this one is pertaining to Cortex Code and observability. So Umesh, how robust is observability within the Snowflake platform for Cortex Code? Is it enterprise ready like other platforms, such as Datadog or Lang

Umesh Unnikrishnan: Yeah. It's a great question. We are in PRPR right now with in private preview right now with some of the observability features that we have. And, you know, we'll eventually have usage views where you can look at the specific prompts and token usage and things like that. But right now, we do publish all of it to the AI observability table. And it's in preview. We are looking for feedback on how to make it more robust and make sure it meets your observability needs.

Santiago Giraldo: Awesome. Okay. I think we have time for one more. And then I think that this is an important one. Think it was actually answered in line, but I think it's great for the broader group. And it's around semantics. What's the best way or fastest way to build semantic layers? And is it required for use of certain Cortex Code skills or using Snowflake Intelligence for those business users?

Will Allen: Best way by far is to use Cortex Code. Yeah. And if you guys can sort of talk a little bit more about that. But all of that demo that I just showed, built a semantic view, and sort of pulled raw log data, helped me build a semantic view, deploy the agent, get it into Snowflake Intelligence. All of that was done through Cortex Code. And and so I we've built special skills to run inside of Cortex Code to make that process faster and easier. So I I would love for you to give it a try if you haven't and and give us feedback. So that's definitely, you know, my view of, like, one of the the fastest way to do it. So use Cortex Code. You can also import those from other spots if you you have them in different areas. So you you've got sort of semantic autopilot that helps you sort of kick start that same work, which is really helpful. But I definitely tend to lean towards just using Cortex Code because it's very, very smart. But, Umesh, anything you want to add to that?

Umesh Unnikrishnan: No, I think you covered all of that.

Santiago Giraldo: Awesome. Well, folks, we're at time here. I wanted to thank everybody and remind you again, this is a monthly series where we're going to be showcasing the latest and greatest innovations from Snowflake. Please tune in. We'll be following up with other resources. Make sure you check out the press release from this morning as well as our new blogs expanding on the capabilities that you've seen today. Thank you very much.

Will Allen: Thanks, everyone.

Umesh Unnikrishnan: Thanks. Thank you.