Snowflake Summit 2026 | Anahita Tafvizi, Snowflake & Durgesh Das, ICE/NYSE

Clips
More from Snowflake Summit 2026

Anahita Tafvizi

Chief Data and AI Officer

Snowflake

Durgesh Das

VP, Data, Analytics & Governance

ICE/NYSE

play_circle_outline Snowflake Summit 2026: ICE/NYSE Leaders, AI-Native CoCo/CoWork, Snowflake-on-Snowflake, and GTM Sales Assistant (1M+ Answers)

play_circle_outline Moving pilots to production by solving real user problems and measuring ROI

play_circle_outline ICE Data Services delivering data products on Snowflake (50+ stages)

play_circle_outline Governed Experimentation: How RBAC, Data Trust, and Semantic Layers Accelerate Safe Deployments

play_circle_outline From insights to action: automating workflows and AI agents with human oversight

Info
Transcript

Anahita Tafvizi, Snowflake & Durgesh Das, ICE/NYSE

Anahita Tafvizi

Chief Data and AI Officer Snowflake

Durgesh Das

VP, Data, Analytics & Governance ICE/NYSE

In this interview from Snowflake Summit 2026, Anahita Tafvizi, chief data and AI officer of Snowflake, joins Durgesh Das, vice president of data, analytics and governance at ICE and NYSE, to talk with theCUBE's Dave Vellante and Rebecca Knight about moving enterprise AI from fragmented experimentation to governed, production-ready agentic intelligence. Tafvizi details Snowflake's 18-month journey to becoming a fully AI-native organization, anchored by a go-to-market assistant now fielding more than 40,000 questions per week across the sales team. She walks th... Read more

explore Keep Exploring

How has Snowflake adopted and implemented its own AI products (such as CoCo and CoWork) internally, and what results have those initiatives produced? add

What factors made those deployments (e.g., the sales assistant and mortgage assistant) successful, and what lessons were learned about moving projects from concept/pilot to full execution? add

How is ICE Data Services using Snowflake to deliver governed, up-to-date data for customers (including support for AI tools like CoCo), and how many ICE Data Services products are available there? add

How do you encourage company-wide experimentation with AI and self-service development while preventing chaotic or inaccurate use of data and ensuring trusted, governed information? add

How can an enterprise move from beautiful dashboards and reports to making AI-driven decisions and automated actions at scale? add

bolt Powered by CUBE AI

Anahita Tafvizi, Snowflake & Durgesh Das, ICE/NYSE

search

Rebecca Knight

>> Hello everyone, and welcome back to theCUBE's live coverage of the Snowflake Summit 2026 here at the Moscone Center in San Francisco. I'm your host, Rebecca Knight, alongside Dave Vellante, Chief Analyst and Co-CEO of theCUBE. I would like to welcome our next guest to the show, Anahita Tafvizi, Chief Data and AI Officer at Snowflake. Welcome Anahita. Welcome back, Anahita.

Anahita Tafvizi

>> Thank you.

Rebecca Knight

>> And Durgesh Das, making your first ever appearance on theCUBE, VP Data Analytics and Governance at ICE and New York Stock Exchange. Thanks so much for coming on.

Durgesh Das

>> Thank you for having us.

Rebecca Knight

>> Durgesh, I'm going to start with you because ICE operates across very complex businesses, exchanges, mortgage technology, data services. As you think about enterprise AI adoption, where are you seeing the biggest opportunities to drive business value?

Durgesh Das

>> As you rightly said that we are in three major business units, mortgage technology, exchange business, and then data business. Christian today in the keynote actually, he brought one good point, which I actually wanted to emphasis that our role is actually to erase the friction. We were already doing right data for right people with governance, but now, the opportunity for us with AI is to provide that information on time without any friction so that the business can bring their own tools and get the answers quickly. I see that is the biggest change as a data head for the ICE. That's my responsibility.

Dave Vellante

>> And you lead the so-called Snowflake on Snowflake, right?

Anahita Tafvizi

>> That's right.

Dave Vellante

>> I call it eating your own dog food, but.

Anahita Tafvizi

>> Drinking your own champagne is a better one.

Dave Vellante

>> But the comms department doesn't like when I say that. Drinking your own champagne, okay. Tell us about what you've learned. We've been talking all week for the last several months about top-down mandates came down from the CEO, "Thou shalt be AI native." And then people look around and say, "Okay, how do we do that?" And then now there's this bottoms up activity going on with OpenClaws and NemoClaws and so forth. How would you describe where you're at in that maturity, things that you've learned?

Anahita Tafvizi

>> Yeah, it's been really incredible being part of Snowflake and running, as you said, Snowflake on Snowflake being the first customer of all these AI products that we have built over the past a year, year and a half I would say, and partnering very closely with Christian and team as we build these products, bringing the voice of the customer to these conversations. I would say Snowflake has really transformed into a truly AI-native organization over the past 12 to 18 months. And as you saw from a lot of the announcements today with the announcements of CoCo and CoWork, you see a lot of that coming to life and we're excited to bring to our customers. It's been incredible to partner with Durgesh and team as well as the other customers who have been part of our research preview and one of the first users of CoCo. So very excited that we are now bringing it back to all of our customers. At Snowflake, we started on this journey, as I mentioned about 18 months ago. One of the examples of what we have built on CoWork was a GTM AI assistant that has now answered more than a million questions across our sales organization. It answers about 40,000 questions a week, so in terms of the amount of excitement that it gets, and the amount of return that we have from our sales team coming back and trying that tool again as well. But also we have not only done that in our GTM, go-to-market organization but across finance, marketing, HR, across the entire company, we're really truly AI native and very excited about that.

Dave Vellante

>> Durgesh, thank you for that, but for a product company like Snowflake, they can move superfast. I know a little bit about ICE and NYSE because we're there quite often and you guys super buttoned down. Not that Snowflake's not, but you in exchange and in your business, governance is everything. So I would suggest that maybe you can't move as fast, but you still have to move fast. So how do you reconcile the move fast but don't break things?

Durgesh Das

>> Yeah, so we actually also move fast, but with governance, and we invested in Snowflake long time back. The way it works is that our governance as a foundation is very strong just because we are using Snowflake's, our back policies, we are using Snowflake's data quality. The trust in data is already there. We actually worked very closely with their, Snowflake's team and ITAS team in last few months actually since last summit actually and when I first met Anahita is that they are the customer zero, as you rightly said. They have solved these problems already, and so we didn't want to restart that solving that problem again. We went and worked with them and it accelerated our solutions. Anahita actually, with her team, we worked with some of the brightest mind in Snowflake, and that allowed us to actually build solutions which are actually being solving the problems inside ICE and NYSE today.

Anahita Tafvizi

>> Maybe something I will add on the governance and you heard this yesterday also from both Sridhar and Daniela. We actually think of governance as an accelerator, not necessarily something that is a trade-off to innovation, and that's something that you see for ICE and NYSE every day.

Rebecca Knight

>> Well, research really shows that actually having constraints makes teams more creative and more innovative because having to work within certain confines increases the amount of solutions you can come up with. I'm curious to hear about moving your deployments and making them go from concept to execution because this is where a lot of companies get stuck, and this is where a lot of people can stay in this pilot mode and stagnate. You've done some very notable ones, sales assistant, mortgage agents. What do you think has made those deployments so successful, and what have you learned along the way about it?

Durgesh Das

>> We wanted to solve the real problems. The sales assistant or one mortgage assistant for external customers, we want to solve the problems where the users are actually there, right there. I'll give you the example of sales assistant, which we work very closely with Anahita's team and Snowflake's forward engineering team. We actually build a customer 360 skill for them for sales analyst, which takes care of five different dashboards in Tableau, which was converted using CoCo and now not only they can get the answers quickly, they can do that preparation within minutes. What you used to take more than a day for them before they go to a big meeting with a customer. If we solve the problems which they are already facing, and the tools they are already using, they want that to be done quickly and that's why we were able to... Actually, it is live today in ICE and NYSE for sales mortgage team where they can get this answer so quickly, which used to take so much longer to solve that.

Dave Vellante

>> I mean, people think of ICE, they think of New York Stock Exchange and trading, but I think of you as a data company and a data products company. What's the relationship between how you use Snowflake and how it feeds data products?

Durgesh Das

>> Yeah, so this is a very good point. We have a business called ICE Data Services as you rightly know. We actually share data on Snowflake on more than 50 stable ages, which is a big number for Snowflake. The principle of what we adopted there is that we have many other delivery platform, but Snowflake is the delivery platform asked by most of our data services customer today. We wanted to give them something where the data is not stale, it is readily available, it is governed, and they can bring their own AI tool on top of that. If they want to use CoCo today, they need the data in Snowflake, which we are providing today in ICE Data Services business. I think we have seven or eight products already on ICE Data Services Snowflake platform.

Dave Vellante

>> One of the things we've been talking about in theCUBE and in theCUBE research is that we have these islands of automation today, and when you have this top down, bottom up sort of innovation, you risk having islands of intelligence and it may be a necessary step to go through. Andy Grove said, "Let the chaos rain and then reign in the chaos." So right now, I feel like a lot of people are learning, experimenting, and of course a lot of production work going on. How do you think about as data executives, data officers, AI visionaries, how do you ensure that you don't just rebuild islands of automation, that you actually have an enterprise substrate, if you will, and avoid just yet another IT industry set of stovepipes.

Anahita Tafvizi

>> For sure, yeah.

Durgesh Das

>> You take it.

Anahita Tafvizi

>> Okay, yeah, I'll take this one. First of all, we want to encourage experimentation because that's a lot of the ideas don't necessarily come from me or from the C-suite. They come from anybody across the company, right? CoCo was actually an idea that started with one engineer, right? We want to always encourage experimentation and we want to enable people to be able to encourage to do experimentation and for everyone to become developers. That's a lot of what we do is how do we build the right data foundation for them to have the right opportunity to experiment? What's important though that you don't end up with chaos because AI is actually lowering the barrier for access to information and if you don't do it in a well-governed and trusted way, you can end up with a lot of chaos. I've heard some really interesting examples from our customers where they've done a lot of vide coding and anybody has come up with its own dashboard that may have not been the correct information but then published it as the truth, and then the next person came in and started using those dashboards. The chaos can actually scale a lot faster with AI if you don't put the right data foundations in place. Our team's job is really how do we put that foundation of data in place the right way, that semantic layer, the right definitions, the right business logic, accurate information, the right governance and role-based access control as Durgesh suggested so that as teams experiment and try these different AI tools and build the new ones, they can trust that it's giving them always accurate information and access to the information that they should have and know more than that. By building that foundation, then you let them experiment but they still all come back to the same correct conclusion.

Durgesh Das

>> I would just add something to what Anahita was just saying that the reason we picked CoCo desktop is actually I was trying to annoy the informal setting and I started talking about what the problems we are facing today where the data is in Snowflake governed with all the access control and trusted. But the end users are in the tools which are completely away from Snowflake and they want to download the data and bring it to this new tool and they build some cool dashboards, and they thought this is the way to present this to their business and we wanted to solve this. Right now, if you think about CoCo desktop, which was announced just this morning.

Dave Vellante

>> Right.

Durgesh Das

>> And we had the privilege to work on that for last three months, it sits right on top of Snowflake's governance. You cannot log into CoCo desktop without having an account in Snowflake, and you have all the access control, role-based access control built in inside. We can even deliver skills to a specific set of users and knowing that analytics users will need this skill while finance will need that skill all built in inside desktop.

Dave Vellante

>> Best practice essentially is built in through those governing guidelines. Sridhar said something yesterday in the analyst session, made me laugh. He said, "I don't think token maxing is a good idea." What are you seeing in your organization in terms of just token consumption? Do you govern the token consumption? How does that work?

Durgesh Das

>> I think they announced that today that we have a per user quota. You have to understand who is actually delivering the result versus who is just consuming tokens. We have a very good AI principle policy which takes care of that. We have a dashboard running on top of usage so we actually control that usage, but what we do is that we look at the deliverable at the end of the day. We don't look at what you are consuming, what actually you are solving. We start the process right from Jira or ADO where the requirements are converted into actual project, and these tools are actually solving those problems, and in the end I see what is the total ROI getting generated. We just don't give you tokens just because you may get more tokens because you are actually solving more problems than the others.

Dave Vellante

>> So very deliberate approach. I was talking to a forward deployed engineer recently said, "What's your token budget?" And he said, "Unlimited." And I said, "Well, uh-oh." And the reason was because they were passing the cost onto the customer, and that's not going to last, right?

Anahita Tafvizi

>> Yeah.

Dave Vellante

>> You're going to have to have some kind of rules in place to govern that and it's got to be intentional, it's got to be tied to ROI.

Anahita Tafvizi

>> For sure.

Durgesh Das

>> On day one also.

Anahita Tafvizi

>> Yeah. I'd say over the last few months, maybe even starting February of this year, we've seen across the industry this encouragement to experiment, try AI, and maybe some competitions as you saw on token maxing. But I think over the past few weeks, we have seen more and more businesses asking a question of what's the ROI that I'm getting from this? And I think being more intentional, as Durgesh said, as the data teams, as ultimately measuring success based on the business value that you created, the amount of innovation that you created, the amount of efficiency that you created. Ultimately, I think Sridhar says it very well, you either help companies make more money or spend less money. So how are you actually creating value and then based on that, seeing what's the ROI of?

Durgesh Das

>> Yeah. By the way, cost intelligence is a skill inside CoCo.

Dave Vellante

>> Right.

Durgesh Das

>> Yeah, it is.

Rebecca Knight

>> But I mean, I'm curious about when you see these beautiful dashboards and they are beautiful and they can tell you so much, but how does an enterprise move from getting those dashboards and those reports to actually making AI driven decisions at scale? I mean, I think that's the big question here.

Durgesh Das

>> So next thing it has to happen is that it has to take action on that. So right now, I think everybody wants to have a human in the loop, just because sometime it can give you wrong answer and we want to make sure that you are not taking decision based on that. Once we saw that human in the loop problem in the sense that human in the loop is actually providing you approval to go and take action, I will see the actionable intelligence very soon coming out of those dashboards. Dashboard is giving you some information, you have to take some action on that, and I think CoCo desktop or too, CoWork is going to solve that problem where you say, "Okay, now go and take action on this." But right now, we are solving through human in the loop solution.

Anahita Tafvizi

>> Yeah. I'd say you want to go from data to insights to action, right? So data is what we spend a lot of our time making sure it's well governed, well articulated, it's accurate, it's fresh. Then you want to come up with the insights where, as you mentioned, like CoCo has been incredible in helping us to get there. But now even for action, you can actually have CoCo give you suggestions on what to do next. And now, you can even have entire workflow automations, that the action is taken on your behalf as well. I'll give you an example. Just a few weeks ago, if a data pipeline breaks, an analyst on my team in the middle of the night, on call has to come in, has to understand and diagnose what happened, propose a solution, submits a PR, then another person have to approve that for it to go in effect. Now that entire workflow is automated. The data pipeline breaks at two in the morning, it triggers an AI agent to wake up to start doing the analysis of what happened, to propose a solution, to submit a PR, so truly having an action in place. And a lot of the announcements that you saw today from Christian is now enabling that to our customers. A human just when they wake up in the morning, you have it in your inbox just to review and approve and then it goes into production.

Dave Vellante

>> This is interesting because we're going from a world of what happened to what happened, why did it happen, what's likely to happen next or what should we do next?

Anahita Tafvizi

>> Yeah.

Durgesh Das

>> Yeah.

Dave Vellante

>> Now when I think about CoWork and CoCo, it seems to me that initially AI is going to attack a lot of the coordination tasks, but those judgment tasks are where the human still needs to be involved. But over time, if I'm interacting with CoWork and that's my interface, it's not maybe the browser or maybe it's through a browser, but it's through CoCo or if I'm a developer or a CoWork, my agents are going to learn from that feedback loop. So is that how you see it playing out? There's a lot of different tools. There's Claude, there's OpenAI, Elon is buying Cursor, there's all these tools, but we're here talking about CoWork and CoCo, so who uses those? What's the persona of those, and how does it relate to all the other tooling in the organization and how do you bring all that together?

Durgesh Das

>> Yeah. I think CoCo desktop is designed mostly for developer community. All my team actually uses that to develop pipeline, make sure that the data is secure and getting loaded, trusted. And at the same time, if there is a migration project, they are using that to move data over to Snowflake. In my opinion, if we have data in a certain platform, like in this case Snowflake, it makes perfect sense to use a tool which is governed and accessible inside Snowflake. That's why my preference to use CoCo Desktop all the time, what I do is the whole day. The idea behind is that if you think about these tool, CoWork was designed for business users, but there is an interchangeability where you can see that a power user can use desktop while a junior developer can come and start making use of CoWork and still make impact.

Dave Vellante

>> Okay. We could go on for a while. I know we're getting tight on time, but I want to leave the audience with, you guys are two really deep data experts and it's very hard to predict where this business is going.

Rebecca Knight

>> But he's going to make you do it anyway.

Dave Vellante

>> Right. I won't. I won't make you go out years, but how do you... Sridhar said something else that was really interesting. We asked him a question in the analyst session, he said, "The answer to that is unknown." But here's what he said is known, that if we keep innovating and we build great products, we're going to be able to compete in the marketplace and do well. Is that the mindset that you're bringing to your jobs every single day?

Anahita Tafvizi

>> Absolutely. I mean, I think it's a very exciting time to be alive. It's incredible to see the rate of innovation that's happening in the industry and to keep up with that. But what we are striving is to continuously every day bring our best to the work, bring most innovation to the work and continue to compete as well and bring the best to our customers like New York Stock Exchange.

Durgesh Das

>> Thank you.

Dave Vellante

>> Because it's hard to predict, right? Jeff Bezos says, "Don't even try to think about what's going to change. Think about what's not going to change and optimize for that, right?" We'll give you the last word.

Durgesh Das

>> For me, I think driving that innovation down to your team is very satisfying. Every day I walk around by my team and I see them using what I told them may be useful for them and making a real impact is extremely satisfying. This is the best time to be actually a data leader because AI cannot go without data.

Anahita Tafvizi

>> That's right.

Rebecca Knight

>> Great note to end on, thank you so much-

Anahita Tafvizi

>> Thank you.

Dave Vellante

>> Thank you, guys.

Durgesh Das

>> Thank you....

Rebecca Knight

>> Anahita and Durgesh, thank you so much for coming on the show.

Anahita Tafvizi

>> Thanks for having us.

Durgesh Das

>> Thank you, it was a pleasure.

Rebecca Knight

>> I'm Rebecca Knight for Dave Vellante. Stay tuned for more of theCUBE's live coverage of the Snowflake Summit. You're watching theCUBE, the leader in enterprise tech news and analysis.