Subscribe
Share
Share
Embed
Snowflake has been a stalwart of the SaaS economy and a leader in enterprise data for the past decade. But the company is deep in the middle of a transformation that most people haven't recognized yet.
In this episode of The Deep View Conversations, senior reporter Nat Rubio-Licht talks with Baris Gultekin, vice president of AI at Snowflake, for a candid look at how the company is navigating the AI era and what it's learning in real time.
Gultekin talks openly about how the entire team inside Snowflake is now using coding tools to build skills and automate their work. That includes non-developers who are using Project SnowWork, an AI agent for professionals across all roles.
Baris joined Snowflake in 2023 through the acquisition of blockchain startup nxyz and has spent the past three years building and running the AI teams inside the enterprise tech giant. He also brings a rare perspective from his time working on Google Assistant in the pre-LLM era, which gives him a unique lens on how much has changed.
Topics covered:
+ Why there is no AI strategy without a data strategy, and what enterprises keep getting wrong
+ How agentic AI has shifted enterprise data from question-answering to automation
+ The SaaSpocalypse and why Snowflake sees AI as a tailwind rather than a threat
+ Cortex Code (CoCo), Snowflake's coding agent that lets customers query their data in plain language instead of SQL
+ The governance and security challenges that come with multi-agent systems
+ How Baris uses coding agents in his own life
If you want to understand how a mature SaaS company reinvents itself inside an AI revolution, this one is worth your time.
Subscribe to the podcast for more conversations with the leaders, builders, and researchers shaping the future of AI.
And don't forget to sign up for The Deep View daily newsletter. We don’t just cover AI, we decode it. In a world flooded with hype, we deliver sharp, no-nonsense insights to keep you ahead of the curve and help you put AI to work every day: subscribe.thedeepview.com
Creators and Guests
What is The Deep View: Conversations?
From frontier labs and enterprise platforms to emerging startups reshaping entire industries, The Deep View: Conversations podcast interviews the brightest minds and the most influential leaders in AI.
Jason Hiner (00:01.302)
In this episode, senior reporter Nat Rubio-Licht talks to Baris Gultekin, vice president of AI at Snowflake. You probably think of Snowflake as one the stalwart SaaS companies of the past decade. What you probably don't know is the transformation the company's been undergoing in the current AI revolution. Baris talked candidly about how all tech companies are racing to capture their share of the AI momentum right now, and he gave us an inside look at some of the things Snowflake is doing and learning.
Baris came to Snowflake in 2023 in the acquisition of blockchain startup, NXYZ. He's had a wild ride in the past three years running the AI teams inside Snowflake. He also shared some of his perspectives from working on Google Assistant in the pre-LLM days. And he talked about how Snowflake customers are now mining their data for a lot more insights now that they aren't stuck using SQL, but can get the data points they need using natural language with Snowflake's Cortex code, affectionately known as Coco. Baris also revealed that the entire team inside Snowflake is now using coding tools to build skills and automate their work. That includes non-developers who are using Project SnowWork, an AI agent for professionals in any role. Okay, so here it is, our conversation with Baris Gultekin of Snowflake.
Nat Rubio-Licht (00:05.4)
Baris, thank you so much for joining us today. Obviously you guys have a lot of exposure to enterprises and sort of the issues that they're facing today with their AI deployment. So I'm going to kick it off with a question that I like to ask leaders in enterprise AI. What do you think businesses are getting right and getting wrong in their AI deployments?
Baris Gultekin (00:28.969)
So first, thanks for having me. It's great question. So what they're getting right is over the last few years, there's been a lot of focus on AI governance, making sure that you're selecting right models, you have the environment set up. So that's still a huge focus. And I think that is really right to make sure you have your governance set properly.
You know your data is already and all governed and then you only share. I the other one is there's a big shift in how people are thinking about AI to from You know just experimenting with it to that being existential for everyone's business. So everyone is prioritizing AI very Very much. So what they're not getting right? You know the flip side of what I just said, which is
You know, there is no AI strategy without a data strategy is what we'd like to say. Right. So you need to have your data in the right state to be able to unleash AI on top of it. Right. AI is very data hungry. So you need to have it kind of ready, cleaned up, governed, you know, have your data silos all broken so that AI can operate across a large data set and be very, very effective. you know,
Getting that ready first before you kind of bring AI to run next to data is really...
Nat Rubio-Licht (01:59.896)
Yeah, absolutely. And another shift that I've obviously seen and you've probably seen as well is this trend that's completely taken over, which is agents. Have agents caused a shift in how enterprises are thinking about their data? And if so, how?
Baris Gultekin (02:16.959)
Yeah, definitely. Agents have been a huge step function change in what's possible. Things that used to not work absolutely start working because AI can think about what are the steps to follow, then follow these steps, reason about what the outcomes are, and then figure out the next steps. So overall, that ability to kind of...
reason and look at the steps and then take action has been a big enabler. Anywhere from moving from just a pure question answering systems to automation, these agents have been very effective, as well as just being able to solve harder and harder problems, being able to run more and more complex things on a long period of time.
agents have unlocked a lot of capabilities. The ability to use, not just kind of give answers, but use tools, take action, all of those are capabilities that agents have brought in.
Nat Rubio-Licht (03:23.896)
Yeah, absolutely. To my understanding though, agents have also maybe caused sort of a shift in the sort of nexus of problems related to data security. How should enterprises sort of address that while staying ahead of the curve on agents?
Baris Gultekin (03:42.219)
Yeah, mean, exactly. comes back to what I was saying. Ultimately, the AI strategy and data strategy need to go hand in hand together. So when you're building agents, governance is incredibly important. Governance on the data, governance on the agents, governance on the tools that it has access to, setting a series of guardrails around what the agent can and cannot do. Those are core design principles as people are building agents.
From a Snowflake perspective, we'd like to think that bringing AI to run next to data gives you lot of advantages, for instance, because a lot of the data platform capabilities that platforms like Snowflake offers is all about governing and managing large amounts of data, as well as when you run AI next to it, all of those governance capabilities, all of the kind
Data governance gets respected by the systems that you can build. those are the key unlocks.
Nat Rubio-Licht (04:46.818)
Yeah, absolutely. So while data governance is incredibly important and should go hand in hand, do you think that broadly it's something that enterprises are paying attention to enough? Or do you think that people are more so concerned with staying ahead of the curve, so to speak?
Baris Gultekin (05:07.092)
think, yeah.
Baris Gultekin (05:11.37)
Data governance is non-negotiable when you're deploying in, especially in enterprise environments, especially in cases where there's a lot of regulation and a lot of sensitivity on data access. At the same time, you do not want to have that be a big reason for you not to be thinking about what AI can do to transform your business.
So what we see is in many cases, we have customers who would have a kind of test environment where they are more able to quickly build, iterate, and then experiment what is possible. But then of course, when it comes to building broad scale deployments, the data governance needs to be in place. All of the production readiness needs to be there. All of the evaluation observability, the full stack needs to be there for you to deploy broadly.
Nat Rubio-Licht (06:08.396)
Yeah. On the topic of agents, I'm sure you've heard about this so-called SaaSpocalypse, this idea that AI is sort of eating software as a service. Do you see this as a legitimate concern? Do you think it's overstated? And where does Snowflake sort of sit in that shift?
Baris Gultekin (06:28.202)
Yeah, so I think it's both. It's a legitimate concern and also it feels a bit overstated. You know, it's a legitimate concern because what we're seeing is AI is incredibly capable to very quickly build user experiences, very quickly build workflows. And there are many solutions out there that are predominantly offering a user experience and the workflow on top of data.
Nat Rubio-Licht (06:33.56)
Okay.
Baris Gultekin (06:57.99)
And then those need to think about how do you reinvent yourself, how do you bring AI into the experience to accelerate and to change, transform these businesses. For Snowflake, Snowflake sits in a different place. Snowflake is an AI data cloud, AI data platform. So what that means is as AI becomes a core way of interacting with data and with services,
A data platform is a very, very helpful product because as I mentioned, you want to kind of think about the context you're offering to the agents. Ultimately, the agents that you build are as good as the data you provide to them. And by bringing your data in a platform like Snowflake, where you're breaking these data silos, you're governing that data, you are essentially able to bring to life higher quality agents that are very, capable.
So Snowflake really benefits from AI from that perspective. The way I think about it is with AI, there is more data that gets unlocked. So a data platform kind of benefits from that. There is more users who can now benefit from this data because it's natural language. And also you get more insight from the data. So overall, all of those are compounding benefits of AI on our product, on our business.
Nat Rubio-Licht (08:22.87)
Yeah. Let's talk a bit about the data itself. As we know, data is the core of making AI that can actually do things. It's the core of making an AI that is functional and useful. Do you think that enterprises broadly have a clear picture of the power of the data that's at their disposal? And how big do you think is the gap between what AI could do and what enterprise data actually allows for?
Baris Gultekin (08:51.796)
Yeah, I mean, so we've been seeing a pretty substantial democratization of access to data and also democratization of intelligence actually on top of that data. So one example is traditionally in enterprises, dashboards are really important. You have to run your business by looking at what's going on.
And the issue is when you see a dashboard and let's say that some trend is going down, you have three more questions. Why is this happening? What can I do about it? And usually that results in you asking a question to a data scientist. Data scientist puts it in their queue and when it's time, they'll go do some research. And in a couple of days, hopefully you have some kind of an answer. So going from that world to a world where you have all of the data at your fingertips, you can ask natural language questions, you can get insights, information.
just changes the way businesses will be run. So that kind of democratization of access so that any business user can directly ask questions of their data, can get insights, changes the way they operate. So I've been incredibly excited about that. And then you also mentioned the ability to take action again becomes possible because you're getting clear insights from the data. Agents can now take action on your behalf as well, automating series of the processes.
So overall, seeing a lot of traction and impact on.
Nat Rubio-Licht (10:24.398)
Yeah. Do you have any advice for enterprises that are maybe sort of unaware of their data hygiene of the data that they have and of what AI could do for them if they were to get that in order?
Baris Gultekin (10:39.635)
Yeah, certainly. there is definitely a spectrum where many companies have been on a journey to modernize their AI, their data stack. And you have companies that are not quite there and there's still a lot of kind of work to be done in terms of cleaning data, governing that data to make it ready for AI to make use of it.
crucial for folks to really do that step, right? Because doing that step accelerates AI development. Otherwise, are hampering what you can do with AI. I've seen companies that, because their data is not quite ready, they are limited to only certain access. Because they're concerned if AI has access to more data, it's not quite governed yet.
It's not in a place, a state that AI can use, which really limits the kinds of things they can do. On the other hand, if you have been investing in this for a long time, the kinds of use cases that you could do span the full spectrum of not only just automating the types of things people are doing today, but also unleashing a lot of new opportunities, new areas to go tackle, just getting a lot more insights from the data that they were able to get before.
Nat Rubio-Licht (12:06.67)
Yeah. Where do you and Snowflake see the biggest opportunity for AI in the next 12 months?
Baris Gultekin (12:20.777)
Um, yes, it is.
Even 12 months is actually quite a long time. The world is changing so rapidly. There's a couple of things that are clearly happening. One is the world of BI is changing. How quickly are you able to get insights from your data is changing. So we're seeing bringing the ability to talk to your data in a world where you have a lot of dashboards that now transforms that experience. The second one is
Nat Rubio-Licht (12:29.17)
You
Baris Gultekin (12:56.199)
you know, how work happens is changing completely, where now you can just automate series of your workflows and bringing all the different parts of your business context together to make decisions, to automate tasks. so how people operate is changing with AI essentially from a work perspective. And the third pillar is just where
In the next 12 months, think we will start seeing more of the autonomous agents becoming real. How do you not just ask a question and get answer or run a quick automation, but rather have your agents work on your behalf on a regular basis, doing things, monitoring things, and then letting you know when you should pay attention. So we're going towards that direction and it's absolutely going to be a reality, already a reality today and will be more important in the coming.
Nat Rubio-Licht (13:56.354)
Yeah. What do you think is the biggest, I guess, risk around enterprise data and AI right now?
Baris Gultekin (14:05.683)
Yeah, I I think it really does boil down to governance and security because we're talking about having agents take action. These agents are now starting to be not just a single action, a single agent, but multiple agents. And when you have multiple agents, they need to have a shared understanding of reality to operate on this reality. So to avoid
hallucinations to avoid incorrect assessments. In a world where you have a governed data and context ready for these agents, you're able to ensure that they are operating on a shared reality, that the people who have access to the data are the ones that are seeing the data and no one else, and you're securing it in a way that important. essentially this layer of kind
agent governance, agent guardrails and control becomes really, important.
Nat Rubio-Licht (15:11.512)
Do you think that we have the tools to, those guardrails, do think that those are sturdy enough right now to prevent significant risk?
Baris Gultekin (15:20.787)
think this is an evolving space. We have a lot of tools already, and as AI evolves, the tools will have to evolve as well. So I wouldn't say this is a done deal. It's an evolving space.
Nat Rubio-Licht (15:34.222)
Absolutely. I also want to talk a little bit about your background. You came from a blockchain startup called NXYZ that was acquired by Snowflake a few years ago. What lessons, if any, can be taken from the world of blockchain into this new age of AI?
Baris Gultekin (15:53.491)
Yeah, I mean, it was a fascinating journey for me. I was at Google for a long time, working on the Google Assistant product in a world before LLMs. That world was more heuristics-based, and then we were putting together all these different use cases. And I said, you know what, AI doesn't quite work yet. And instead, blockchain is where it's at. And I started the company on blockchain. And right when I did that, ChatGPD happened.
Nat Rubio-Licht (16:17.73)
Thank
Baris Gultekin (16:23.144)
You know, with blockchain, blockchain is a fascinating technology. I was very, very excited about it. I was very excited about the data part of blockchain. So my startup was indexing the blockchain and making it accessible to everyone who were building on top of it. from that perspective, you know, I've always loved building platforms and building data oriented products and kind of
AI is absolutely kind of taking that to the next level. I mean, there's a ton of new opportunities that are starting to emerge at the intersection of blockchain and AI. I haven't focused on that recently, but I'll say.
Baris Gultekin (17:15.154)
Yeah, I think, my perspective, the transferable part has been more around focusing on the data, making that data available and accessible to the higher layers of the stack that ends up using it.
Nat Rubio-Licht (17:35.608)
Yeah, mean, data is the core of both technologies for sure. And you've been at Snowflake now for a couple of years, and you joined kind of in a very transformative time for AI, if I'm remembering correctly. What has surprised you in the last couple of years about this massive shift that we've seen?
Baris Gultekin (17:38.566)
Right, right.
Baris Gultekin (17:48.998)
Mm-hmm. Yeah.
Baris Gultekin (17:58.225)
I feel incredibly fortunate and I feel gratitude to be here at this time. I essentially was, my company was acquired right when the big shift into AI was happening. So we were able to build Snowflakes AI products from the ground up over the last three years. And what has surprised me is
the pace of innovation, like how fast everything is moving, how fast we're moving. Coming from a startup into a large company like Snowflake, but accelerating how fast we were moving, even faster than we were at the startup is just fascinating to me. And that's exactly where the industry is. The industry has...
created this tremendous opportunity and then everyone is working very hard in AI to capture the momentum and deliver a great set of products. So I've been very fortunate to be here.
Nat Rubio-Licht (19:04.28)
Yeah. Have there been any, I guess, shifts in the way that customers and your clients are using Snowflake that have impressed you or surprised you?
Baris Gultekin (19:19.816)
Definitely. have, I mean, the way customers are using Snowflake is changing dramatically. We've launched our coding agent called Cortex Code. We affectionately call it Coco. We love Coco. It's been a transformational experience because instead of interacting with Snowflake through SQL and through scripts, you're just interacting in natural language.
Nat Rubio-Licht (19:30.232)
Not.
You
Baris Gultekin (19:47.565)
substantially accelerate, you know, path to production for our customers. So we're seeing customers are doing a lot more and getting a ton of value from Snowflake because the tools make it very, very easy to do so. We're also seeing our customers build very capable agents across the board, anything from building, know, go to market and sales agents to kind of building dashboards that you can talk to.
It's been transformative for many customers.
Nat Rubio-Licht (20:21.76)
Yeah. You mentioned coding agents, and that's something that I'd like to hit on as well. How is AI being leveraged inside of Snowflake? Are your engineers using coding agents? Are your marketing people leveraging chatbots, et cetera, et cetera?
Baris Gultekin (20:41.042)
We are actually the whole company is now using coding agents. Cortex code, our product, because we're eating our own dog food, we have our data estate very readily available and in a governed way. So once we brought our coding agent and then made it accessible to the whole company, it unleashed tremendous opportunities. And we're seeing not just our engineers building, writing code with the coding agent, but also anyone from
Nat Rubio-Licht (20:44.588)
and
Nat Rubio-Licht (20:50.68)
Thank
Baris Gultekin (21:10.344)
marketing and sales are using these coding agents to develop pretty substantial products. I'll give you a couple of examples. We built a sales assistant for the whole organization and then deployed it broadly to 6,000 sellers. And this has been very, very impactful. Our sellers started using it for checking quota attainment, checking their customers' usage or
any type of competitive analysis. But then once we brought kind of a coding agent and then gave that to our sellers, we have a product called SnowWork, which is a coding agent, but for kind of for non-developers. That product essentially allows our whole team to automate a lot of what they do by just writing
automation skills for themselves and then running these on a regular basis. I've had account reps build anomaly detection for their customers so that they get notified when something happens. There's a lot of very creative uses of AI across the company.
Nat Rubio-Licht (22:24.014)
Creative How.
Baris Gultekin (22:26.055)
You know, essentially, whatever you need and you can think of, you can make happen. It is literally that. you know, our marketing team is now using AI to take concepts across the board, do series of comparisons, write substantial analysis and research by bringing all of that content together.
We have folks who are building applications that are really, powerful for their workflows. We have teams that are automating those workflows. Anywhere from a finance team that builds a very, very rich set of data capabilities to analyze what's going on in the business to building applications in HR for doing very specific tasks. So seeing broad use of it.
Nat Rubio-Licht (23:25.08)
Yeah. Has there been, whether it's within Snowflake or with your clients, any apprehension in deploying these systems?
Baris Gultekin (23:41.671)
You there is a...
There's a question around, we ready to deploy this broadly? And that question has a couple of components. One is, is it high quality or is it going to make mistakes? Second one is, is it governed? Governed has many sub-subcomponents. Will I get my ROI? What is the cost? Does it have the right access to data? Does it respect privacy? Is it secure? So all of those are.
elements that our customers need to go through and make sure that they feel ready before they deploy these broadly.
Nat Rubio-Licht (24:25.176)
Yeah, absolutely. We've obviously talked a lot about the creative ways that you can use AI tools. I want to talk a little bit about how you use them. How do you spend your day and what is the best tip you have for leaders who want to start deploying AI agents or AI tools in their day to day?
Baris Gultekin (24:46.631)
Yeah, I've I noticed that, you know, they is very busy. unless you really create the space to think about what are your what are the things that you tend to do on a regular basis? What are your routines? You'll you'll you'll you'll you'll end up using AI only as a talk lookup tool, right? Oh, I have this question. Let me go ask or I have this thing need. Let me go ask. But using, you know, coding agents.
is very powerful, but it requires space to think about how do you set this up for yourself? Meaning, how do you create the context so that you create your own co-pilot that really understands what you're doing and can automate a lot of your tasks? And it requires that upfront investment. Once you make that investment, you get a lot of value out of it. So for me personally, I've built my work co-pilot by kind of telling my agent a series of tasks that I do.
Here is my content, here's all of my documents. I connected it to series of data sources. And now that it has all of that context, I can now create automations for my work. Anything that I do more than a few times, I'll go turn that into an automation and I can go and run it and save time. So that's how I use it. I use it to anything from brainstorming new ideas to reviewing content like blog posts, press releases.
to just writing content. Overall, I have my regular routines about kind of like we have a, we write our weekly updates, weekly snippets. Now my agent knows everything I do. So I can just say, write me what I've done last week and what should I focus on next week? And it has all of that content to be able to give me that information.
Nat Rubio-Licht (26:38.646)
Yeah, I'm curious, do you use any AI tools in your personal day-to-day life unrelated to work?
Baris Gultekin (26:47.047)
I do, I do. Very similarly, I use coding agents for that as well. know, for one example is, my daughter is in seventh grade. So we were just talking about the path to high school and how we should kind of go charter that. using AI to look up
Nat Rubio-Licht (26:48.258)
I need it.
Interesting.
Baris Gultekin (27:13.583)
all the different schools out there, give us information about it, put it all in one place so that we can kind of reason about all the different kind of possibilities. It's quite rich. I've also used it for kind of tracking my health. Health records is all over the place. If you can just bring it all in one place, you have a lot of insight about what's going on. So all of those are possible with coding agents.
Nat Rubio-Licht (27:40.632)
Yeah. I want to go back to something you mentioned earlier, which is the rapid pace of change. What is something that you want enterprises, your clients, people that you work with to take away in keeping up with that pace of change?
Baris Gultekin (28:04.612)
I change is hard. Keeping up with the change requires...
first and openness and willingness to kind of drop the old. And that requires also, as I mentioned, space to think about it in the first place. But keeping up with the changes, I think there's a lot of information out there. And for me personally, that is a series of data sources that I collect. I use AI to summarize it for me because it knows what's important for me.
but it's...
change itself is more of a, how do you drive that behavior change in the organization? That becomes a more important question. And then for us, driving behavior change has been about communicating the importance, tracking how it's done, sharing across the company of best practices and examples of how AI is impacting everyone's life. But overall,
making it clear to everyone that that is a it's a it's an expectation of everyone's job to to be living and breathing AI.
Nat Rubio-Licht (29:25.39)
Wow, okay, yeah. And I know that we've talked about how hard it is to predict what is going to happen. Obviously, it's like a fire hose, just the amount of things that are changing and being announced every single day. But in the next year or a few years, where do you hope Snowflake sort of fits into the broader AI landscape?
Baris Gultekin (29:52.358)
Yeah. So as I mentioned, think, again, AI is as good as the context it's given. And Snowflake is in a very unique position with our customers trusting their data with Snowflake. So my hope is to enable our customers to make the most use of their data by making it available for them to build very high quality agents, what I'll call kind of data-oriented agents.
So that we're on that journey and there is so much more to be done because these models are getting better. So for me, it's all about enabling our customers to build high quality agents that are very capable to help them take action, get insights from their data.
Nat Rubio-Licht (30:37.432)
Yeah. All right. The last question that I've got for you. What is your hottest AI take?
Baris Gultekin (30:47.366)
How does AI take?
Baris Gultekin (30:55.258)
My hardest day I take is that.
Baris Gultekin (31:01.86)
What makes AI work is data. What differentiates what everyone does is data. So models are incredibly important. They are huge enablers. And once that enablement is there, what makes the difference is how well are you using your data estate to differentiate and to build high quality experiences. So again, to me, it's all about the data and the context for the next phase.
Nat Rubio-Licht (31:31.086)
Absolutely. Well, Barish, thank you so much for taking the time to chat. This has been a pleasure and I appreciate your time.
Baris Gultekin (31:40.772)
Of thanks for having me.