Snowflake Summit 2025 | Matt Glickman, Genesis Computing & Stefan Williams, Snowflake

Clips
More from Snowflake Summit 2025

Matt Glickman

Co-Founder & CEO

Genesis Computing

Stefan Williams

VP, Snowflake Ventures

Snowflake

play_circle_outline Snowflake Ventures investment in Genesis Computing

play_circle_outline Genesis Computing's AI use case for data pipeline automation

play_circle_outline Genesis Computing's approach to human-in-the-loop interaction

play_circle_outline Importance of agents learning from exceptions and accumulating knowledge

play_circle_outline Hiring needs for core engineers and sales engineers at Genesis Computing

Info
Transcript

Matt Glickman, Genesis Computing & Stefan Williams, Snowflake

Matt Glickman

Co-Founder & CEO Genesis Computing

Stefan Williams

VP, Snowflake Ventures Snowflake

In this on-site segment from Snowflake Summit 2025, Stefan Williams, head of corporate development at Snowflake, and Matt Glickman of Genesis Computing, join theCUBE’s Dave Vellante and John Furrier to unpack Snowflake Ventures’ recent investment in Genesis and what it means for enterprise data teams. Broadcasting live from the show floor, the conversation explores Snowflake’s vision of the AI Data Cloud as the connective tissue for AI-driven innovation and operational efficiency.

Williams details how Snowflake is extending its platform to support mu... Read more

explore Keep Exploring

What is the background and strategy of Snowflake Ventures in supporting companies like Genesis that integrate Snowflake into their infrastructure for customer technology delivery? add

What problem were users initially looking to solve last summer before realizing that the more urgent need was addressing messy, partially migrated data upstream from their business users? add

What approach is being taken to incorporate human input into AI processes, specifically in terms of decision-making and problem-solving? add

What is the secret sauce that makes the AI agents valuable assets to customers over time? add

What type of job is in high demand for individuals with experience as data engineers or product managers? add

bolt Powered by CUBE AI

Matt Glickman, Genesis Computing & Stefan Williams, Snowflake

search

Rebecca Knight

>> Good afternoon, Data Nation, and welcome back to theCUBE's live coverage of the Snowflake Summit 2025 here at the Moscone Center in San Francisco. I'm your host, Rebecca Knight, alongside my co-host and analyst, Dave Vellante. We've got a great segment coming up. We've got Matt Glickman. He is the CEO and co-founder of Genesis Computing. Welcome, Matt.

Matt Glickman

>> Thank you.

Rebecca Knight

>> And Stefan Williams, VP of Snowflake Ventures at Snowflake. Welcome back to theCUBE, Stefan.

Stefan Williams

>> Thank you.

Rebecca Knight

>> I'm actually going to start with you because Snowflake Ventures recently invested in Genesis Computing. I want to hear what sparked the interest in the investment because there are a lot of automation tools out there. What was it about Genesis that set it apart?

Stefan Williams

>> We've known Matthew and the team at Genesis for a long time. They previously, obviously, came from Snowflake. And part of what we're trying to do with Snowflake Ventures is really trying to lean into companies that are willing to think about Snowflake as a core component of their infrastructure as they look to deliver the technology to customers. And Matt and team from the ground up started to build their product as a native app on Snowflake, really unlocking this agentic AI use case that I'm sure Matt will get into in a little bit. And so we believe in the team. We obviously believe in the market opportunity. And because the technology is powered on top of Snowflake, they're going to have a data advantage when they go into some of the customers that they're trying to work with.

Dave Vellante

>> Is it fair to say that your objective when you invest in the company is to attract people into the ecosystem, grow that leverage, and exit value is not the primary motivator or how much does that factor in?

Stefan Williams

>> We have a fiduciary responsibility to invest the company's money into companies that we think have a great market opportunity.

Dave Vellante

>> Sure.

Stefan Williams

>> So I'm not going to say we don't invest for ROI reasons. We have one LP at Snowflake, it's the cash off of our balance sheet. We really try to use it as a way to lean into companies that are standing on Snowflake technologies and primitives above and beyond what a regular partner would do and really unlocking new use cases, new personas, new industries where we can go to our customers and actually allow them to use the data that they have in Snowflake across a much broader set of workflows, use cases that they would've done with just Snowflake in and of itself. And so to the extent we can invest in companies, we can get them to accelerate the innovation that they're doing on Snowflake and ultimately allow our customers to do that, the investments a really nice tool that we can use to unlock that opportunity.

Dave Vellante

>> You're not allergic to IRR, but you want to see participation in the ecosystem and leverage.

Stefan Williams

>> Yeah. So one of the things we try to do is we have a commercial agreement on product, so we try to have certain product integration milestones that we line up with these partners to go and fulfill that are actually going to be valuable for our customers collectively. It's not we are forcing you to do it, but actually it's something that a lot of sense. We'll tie some nice go-to-market tailwinds alongside of that and we'll actually formalize that in commercial agreement that actually is, hey, we've got some meat on the bone here, we're actually going to go and actually do something versus here's money and we'll see you in six, 12, when I get my money back.

Dave Vellante

>> So Matt, why did you and your co-founders start Genesis?

Matt Glickman

>> So we were long-term Snowflakes ourselves. We had basically helped build a lot of the core components that we saw an opportunity to take advantage of, particularly when the AI wave started. And we saw that from the marketplace to native applications, we realized with AI coming on board, my co-founder and I were effectively hoping that someone would figure out to build once what we saw, customer after customer, building themselves and realizing it was harder to go from demo to production than they thought. And we thought that hope was a pretty terrible strategy if that's what we were starting from, so we realized this was the time to do it. And at Summit last year we launched. We basically had a fixed timeline and said we have eight weeks to get our first version of product to market and to be part of the native app launch. We did that, we hit, and now obviously the AI wave and agents have taken on from there.

Dave Vellante

>> So please describe what those customers were building and what that last mile looked like.

Matt Glickman

>> So again, going back last summer, the first wave was how can I figure out how to enable my business users to talk to their data more without having to have it being a crunch of my data teams, and like everyone else, we started figuring out chasing that use case. Well, we soon realized a few months in every customer we started talking to they're saying like, "That's a good problem to help me with, but what would really help is all the data that's further upstream from that business user, that is a mess. It is hard to maintain. And by the way, it's only partially migrated from wherever I was coming from to come to Snowflake." And this kind of task we realized where AI really hits the mark is where you go after problems that human beings don't want to do. My time at Snowflake, my time before Snowflake, went around quant teams at Goldman, I've never met a data engineer who said, "Please don't help me take away the workload I have." It's a thankless job, it's a high burnout job, and it's highly complex and it has a very asymmetric kind of, if it works, no one notices. If it doesn't work, it's a disaster. So we realized if we can take particularly this next wave of reasoning models that understand business context, understand Snowflake, and understand coding, as we all know, become the killer use case, and can actually help customers build the pipelines, not reinvent the wheel, not reinvent dbt or Snowflake or, whatever, OpenFlow, use these tools, but take it off of the hands of the humans who have to basically code it and put their business logic inside. And the best case is, what typically happens now, is that all that logic gets put into code and becomes technical debt immediately. Nobody documents it, no one understands it. And the high attrition rate of data engineers makes it this terrible kind of corpus of growing angst by attacking this problem and having agents not only build the pipelines but learn the logic, learn the business, learn the style of a company. We actually presented today with GXS Bank who is a neobank, who's from the start is saying, "We are investing in AI building our pipelines so we don't have to have expensive teams of high attrition who are going to build and maintain these pipelines." AI is the perfect use case for that.

Dave Vellante

>> So really good problem to solve. My question is the, what I consider a largely linear data pipeline process, is that fundamentally an okay process that just needs automation, assistance, acceleration versus all the stuff that Zhamak Dehghani complains about versus rethinking that process? It sounds like the former, not the latter, if that makes sense.

Matt Glickman

>> Oh no, it's rethinking. So I think if you think about, even when people do migrations, they don't really want to migrate line-for-line, bad code for bad code. They want to take advantage of what's been learned, maybe it's not relevant anymore and redo it in a way that captures the knowledge but does it in a modern way. We actually had a customer we were working with in healthcare who basically had already signed on an SI consultant to do this kind of rethinking as you're describing and rebuilding into Snowflake. And they realized it was, actually, it was a hard problem, harder than they thought, and they were falling behind, so they reached out and said like, "Let's see how maybe Genesis with your agents could find things that maybe the SIs couldn't." And that was the most surprising thing. Obviously, it was going to be faster and cheaper, but it uncovered things that the human beings had missed, and that I think is doing it better. And then actually once it's built, now you have the knowledge to improve it over time, and now it becomes an asset instead of having it be a liability or a technical debt.

Dave Vellante

>> So you're not paving the cow path, if I can use that phrase, you're actually, maybe there's a little bit of that, but then you're enabling a new process that's-

Matt Glickman

>> Enabling a new process-...

Dave Vellante

>> 10X cheaper and 10X better kind of thing.

Matt Glickman

>> And capturing in natural language what is the intent of what this pipeline or what this transformation is doing so that next, when it's dbt today and PDQ tomorrow or Snowflake OpenFlow the next day, it's no longer a rebuild from scratch. We already captured the business logic, and now this agent has a knowledge base stored securely in Snowflake as an asset that can be used for the next iteration for detecting failures. A failure happens and now it's a disaster because it's like three in the morning, something happened, you're test didn't catch it, and all of a sudden you're trying to figure out what changed. What if an agent before the on-call person was called, the agent goes and says, "Well, I understand this pipeline. I have all this knowledge. I understand that this was the assumption that we made and it changed." An example was an identifier all of a sudden came in with a decimal point and it all of a sudden duplicated all the rows that we had before. These kind of things are just thankless, and every customer we've talked to, and again, and they have hybrid environments, they have some dbt, they have some ... It's saying, when this works, it's a no-brainer because it'll free up their data engineering team to then work with the business analyst. Instead of trying to replace a data engineer, free up their time so they can then try to find the next new opportunity instead of being bogged down by this high attrition, burnout, thankless job that only they get noticed when it fails.

Rebecca Knight

>> So that was going to be my next question is how does that change what the human data engineer is doing with his or her time? And what does that unlock for organizations too?

Matt Glickman

>> It's this massive augmentation for their entire data workforce who's basically relieved by not having to do that droll can now focus on higher order tasks, higher order work, finding what is an opportunity for a media company to have more subscribers instead of trying to say, "Why does this system A not connected system B and it's breaking between them?" So no one's losing their jobs, no one's losing their jobs.

Dave Vellante

>> I don't know about you guys, but GenAI has not taken away the time I spent working, it's actually created this really interesting backlog that I can't get to because it's taken away all the stuff I don't want to do. So now I spend time with all these great new ideas, and I have this huge backlog. I'm more busy than I've ever been.

Matt Glickman

>> And you get into the flow state. Because now that thing, that backlog thing is what draws you down. It takes away your energy. And this is where AI is most effective is freeing you up. I mean, again, we use AI for building out the actual system of Genesis. And when we were actually building it, I was amazed of just you enter a flow state because you're able to focus on a problem and not having to context switch and not having to be interrupted. It just frees you up. And I think that's the big opportunity here. And focusing on the problems that people don't want to do, instead of chasing the sexy idea that makes for a cool demo, focus on the things they don't want to do, let the backlog build up, apply AI to the backlog, free you up on the problems you want to solve.

Dave Vellante

>> Stefan, I'm just curious, do you have a board seat?

Stefan Williams

>> We don't take board seats. No.

Dave Vellante

>> You don't take board seats?

Stefan Williams

>> We don't, no.

Dave Vellante

>> Ever?

Stefan Williams

>> We don't take board seats ever. We sometimes take board observer seats, but as a public company, we don't want to take-

Dave Vellante

>> Yeah, I get it. Okay....

Stefan Williams

>> have influence.

Dave Vellante

>> But you still got good visibility on what's happening in the Snowflake customer base, so what's happening in the Snowflake customer base? I mean from your perspective as the lead investor, how's it going?

Stefan Williams

>> What's happening in the customer base?

Matt Glickman

>> They go-to-market alignment that we have with you.

Stefan Williams

>> I think what's really interesting with the way that Genesis is building on Snowflake with this native app principle, native app deployment is that, as a small startup, you're able to actually get a lot of access to bigger customers that you would've typically never had access to because they don't want to trust a startup. Big companies don't want to trust startups. But when you're containerizing your application and you're deploying the application in the customer Snowflake environment, you're actually not getting access to the data. And so you're still able to provide functionality, but you don't have to go through all the data governance, the security, the lengthy procurement cycles that can take six, nine months to go and actually deploy a solution. And so now as a startup, you get access to more data, you get a much more streamlined selling motion. You've got Snowflake reps that are looking to try to unlock new use cases within customers. So if I'm a rep at Snowflake and I have a customer, a big banking customer, I'm selling them one thing. But if I bring in Genesis, I'm actually going to unlock a new workload, a new persona and a use case that I wouldn't be able to drive myself. But if they come in, they drive it. And by the way, that helps me as a rep because I make commission on that. And so now there's this natural incentive to bring partners like Genesis into our accounts to go and actually help them unlock this new wave of innovation that's happening with AI, but across many different use cases. And so I think what I hear from customers when I'm speaking from Venture's perspective, I think the value that the ecosystem brings, the value that the hundreds if not thousands of companies building in and around Snowflake and the integrity that we can keep with the data, the governance, the security, I think is extremely powerful in allowing people to do more with the data.

Dave Vellante

>> So I'm really interested in this data engineering use case. If I go back, and correct me if I'm wrong, but if I go back to 2022, remember we shifted from spend at all costs, COVID, open the checkbooks to-

Stefan Williams

>> Good old days.

Dave Vellante

>> Yeah, good old days. They never last. But you remember the market shifted and everybody started to tighten up. Cost optimization was a big theme. AWS was big on that. You guys right there with them. And as well, customers were telling us, "We're taking the data engineering work out of Snowflake and we're doing it elsewhere because it's cheaper." Some Spark execution engine, maybe it was AWS, maybe it was Databricks, whatever, it doesn't matter. It sounds like you directly have addressed that, at least in part with Genesis, Snowflake has changed pricing. You're going after the data science persona, that is now a key workload. Are there examples that you can share with customers where that's the case?

Matt Glickman

>> Yeah, and I think this is exactly what we're seeing is that there's data engineering workloads that Snowflake was just not getting pulled into just because, for reasons in a lot of ways that consultants who are looking to do things that would justify their time, the complex is sellable. When you have an AI, you basically say, "How would you suggest we solve this problem?" The AI is not going to be biased by which ways mean more billable hours. The AI is going to basically say, "Well, for what you're trying to do, I think it actually, and the data is already in Snowflake, it actually is going to be easier to do it by actually keeping it where it is and it has the governance." So you get more from first principles the right way to do things. And similarly, if there's a workload that where Snowflake is not the right place to do so, you also have AI will be guiding that. I think that's this new kind of mode of agentic search engine optimization where the real target audience, if you want to think about of who Snowflake marketing or Databricks marketing is targeting, it's effectively agents. It's AI. If the AI understands a problem ... I guess Snowflake, as an example, has amazing documentation with one of the untold special sauces that Snowflake has that AIs love it. It's cross-referenced, has great examples. Given a choice, it will choose to use Snowflake SQL versus using something less known. So I think this idea of getting Snowflake to be back into the game and not have it be a bias that is going to skew people to make the wrong decision. And also having AI as a killer use case for AI and saying don't think of data engineering as which one has the better API, think of where is the most governance, where's the most cost performance and let the AI figure out how to build it there, and it really changes the game.

Dave Vellante

>> So multi-part question. It says, "Genesis is the first multi-agent system operating natively within Snowflake, designed specifically of enterprise high-quality data maturity." Okay, cool. What is that? And then how does it interact with humans, and can your agents learn from the reasoning traces of humans when there are exceptions?

Matt Glickman

>> Yes. Great, great question. The first one is, again, given our background, whether or not Ventures was going to invest or not, we were going to build on Snowflake. We helped build the kind of core components. It was obvious. And if we didn't do it, no one would. So we wanted to be out first out of the gate, and we saw opportunity for what Stefan said of go-to-market, get there quicker, get through procurement, security. It's a no-brainer, and we have the entire customer base. If we align ourselves, we go-to-market. The human in loop is the super interesting thing that most of the market still gets wrong. You don't save people time, you talk about the flow state, you don't save people time if I have to think about having to tell an AI, "Go chase my backlog. Go do things that I don't want to be tackling." You want it to be happening in the background. Your backlog is there, it's not owned by the AI. It's going to be tracking and doing it. That's how we're approaching the human in the loop where we give, just like you would for a junior employee, a good junior employee, a bad junior employee would wait to be told what to do. The good ones say like, "I got this. Okay? Give me the general parameters, and I'll come back to you when I have questions." That's how we do human in the loop. So you give it a project, "Let's migrate a pipeline from Informatica and SAP to Snowflake dbt, and here's the credentials and go." What I want to do is I want to be then in the loop when it gets stuck, when it's not sure of the answer. And this was the major breakthrough that we uncovered when the reasoning models hit early this year, 01, and now the slew of them. Before that point, no matter how much you prompt and no matter how much you begged, you gave it enough context, it will say, "I'm confident. That looks great," even if it was completely wrong. Where now because that extra thinking time, now we get it to be like it will say, "I'm 100 percent confident." Or if I'm not, "I need help. And these are my clarifying questions because I looked at ... In the example of what you were telling me to do, you were saying was going to be yes or no field. And I looked at the data and I saw a no, what do you want me to do with that?"

Rebecca Knight

>> And that's where you bring the human element.

Matt Glickman

>> That's where you bring the human. In that context saying, "As part of this research and part of this project, this is the point I'm stuck on, and I'm not going to proceed until you say, 'I don't care.' It may be it's such a minor point, whatever, I don't care. I'm counting subscriber count. I don't care if it's a no there, it's fine. Or this is, maybe, this is going to be the big qualifier if I'm going to send out that ad campaign. And if they said no, they opted out. If it's a no, I'm not going to send it." So this kind of escalation as the much more natural way to have AI engage, and most importantly, not force the users to have to learn how to talk to an agent. Have the agent learn how to talk to the human and their workflow. An example today was they want to have their customers, their users, to find in a Google Sheet the report they want and that's where they want it to be. And so we said, "Great, we'll teach the agents how to read Google Sheet and that'll be the point contention, of collaboration." Meet them where they are, escalate to them when you need them, and don't overload them with things that you're confident about. You do this correctly, it's magic. You miss it, it's a disaster and you turn it off. And I think a lot of the things that people now are unwinding, like the cloners of the world of, I think they were not thinking about how to-

Dave Vellante

>> That was-...

Matt Glickman

>> but I think they were just over their skis and not thinking about what is the right way to have the human in the loop. And if you say you take the human out of the loop and you expose the end user to bad results, bad things will happen.

Dave Vellante

>> Just to put a finer point on it. So your agents will learn from those exceptions? -

Matt Glickman

>> We will learn. And that's the secret sauce is that we learn, but it becomes an asset to our customers. The learning is stored in their Snowflake. It doesn't go to Genesis, it doesn't go somewhere else. It's becomes their asset that accumulates over time. So now, and this was actually, I mean Jensen at Summit keynote last year, he made this comment about how, he's like, "NVIDIA's a great place, but people leave. And when they leave," he's like, "we delete their emails. All their knowledge they had is lost." That no longer has to happen. The AIs should not only be trying to make things better, they should be accumulating knowledge so that technical debt becomes technical assets. And now the longer these agents go, the smarter they're going to get and the more valuable they're going to get for the enterprise that they're working in. It's untapped potential. And I think that if you can figure out the right way to integrate them into the environment, into these kind of teams where they're seen as a augmentation and not as a threat or annoyance, magic happens.

Dave Vellante

>> Are you hiring?

Matt Glickman

>> We are hiring.

Dave Vellante

>> What are you hiring?

Matt Glickman

>> We're hiring for core engineers and sales engineers. As you might imagine, the demand is high for this use case. And I've made this joke now, it's like we show up to a meeting, unless we show up drunk, and even those times, not that I show up drunk, but even if I did, it's going to land because it's such a pain point and the land is .

Rebecca Knight

>> You looking for a job, Dave?

Dave Vellante

>> No, but I'm curious. So you're not hiring burnt out data engineers for product managers?

Matt Glickman

>> No, no, no. We want-

Dave Vellante

>> Well, but wouldn't they make a good product manager because they know the problem inside and out?

Matt Glickman

>> No, no. They would, and I'd happily hire them to be a solution engineer working with customers to help them understand and say, "I've been there," and that worked with me. When I came from Goldman, I had been on the other side, I had seen the pain coming to Snowflake, and I was able to actually use that to work with customers and make them successful. I'd rather that them helping build my pipelines, which I don't want them to do.

Rebecca Knight

>> We could talk about this all day. Stefan, Matt, thank you both so much-

Matt Glickman

>> Thank you, guys....

Rebecca Knight

>> for coming on theCUBE, a really fun conversation.

Matt Glickman

>> Thanks for having me.

Rebecca Knight

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