Data is Back: MongoDB, Databricks, Snowflake

0:00What is happening right now, Molly, is that you see these hyperscalers, some of them, are running out of capacity. So this large customer in Texas, speaking to them, they have really good partnership with one of the hyperscalers. They wanted to move more workloads in cloud, public cloud, and also some AI workload. And the hyperscaler said, sorry, we don't have a capacity. And they are one of the top 50 customers for that hyperscaler. So now they are saying, we were decommissioning this data center, CJ. We can't do that. So we are going to now run workloads on-prem and we are going to run some AI workloads, whether it's data privacy issues or other issues.

0:34We are going to run those workloads. Frontier Labs are using us for a multitude of use cases. We cannot disclose which, what use cases. There's only a handful. I wonder who they could be. Because a lot of AI-native startups, whether it's Emergent, Base44, Metal.ai, specifically 11 Labs. When we look at the 11 lab story, they have north of 50 million agents, depending on when you look at it, all running on MongoDB. Data is the unsung hero and data is bad.

1:13CJ, welcome to Sorcery. It is fantastic to be here with you, Molly. All the way in Paris. All the way in Paris. It's really funny. I was telling people about everyone that we're interviewing this week and it just like, okay, where can you interview all the top Silicon Valley CEOs? It's in Paris. It doesn't make sense. That does not make sense. Scott at Cognition, we were talking and we said, next week, should we meet in California? He said, yeah, that would be a good idea rather than always meeting in Paris or other places. So I am with you. Scott was one of the first interviews that we did. And it was funny because we were talking about Devon and the comeback of Devon.

1:54And the first task that Devon did was spin up MongoDB. I was so proud. Scott said he couldn't sleep at night, that Devon could set up MongoDB. And he felt that was a great task by Devon. Okay. So we, again, we're here at Raise. You were on the stage earlier with Laura from OpenAI. So what were you guys talking about? Just, you know, Laura is very focused on from the startup ecosystem, founders, what OpenAI does, how do they work with the founders, and specifically, what is the approach MongoDB has taken? because a lot of AI native startups, whether it's Emergent, Base44, both are wide coding platforms built on MongoDB.

2:39But then you have also Metal.ai, specifically 11 Labs, that is currently running all agentic workloads on MongoDB. So we talked about how do you truly partner with founders? How do you stay close to them as they are scaling their enterprise or they're scaling just the hyper growth era. And how do you, what are some best practices in working with them when they need you the most? So that was basically the focus. And the data layer is typically the unsung hero, but we feel that models and data both are needed to create a great agent tech application. And we talked about that. Everett Randall had just tweeted out something on the token economy and how much tokens and agents are creating more data.

3:31And then I tweeted underneath it. I don't know why I'm talking in tweets right now, but I tweeted beneath it, data is back. And he said, big data. Big data. So you are a winner, a downstream winner of everything that's happening in AI. And I think it was like a little unexpected too, because people were just so focused on the models, but guess what? All these models and all these agents are creating so much data. So where does MongoDB fit in with the AI super cycle and all the agentic economy? Yes, I would say the way I see the world is pretty simple. So many people say it, but they are truly not behind it.

4:15And here's what I say. Like people like to say companies, they like to say, oh, we are truly customer obsessed culture. oh, we care about customers, customer focus, but I'm really, really customer obsessed. And in a typical week, Molly, I feel it's not a good week unless I have spoken individually to 10 to 12 customers, sometimes even more, even if it's a short week. So I'm constantly learning from customers on what they are trying to do with AI specifically. And when you look at MongoDB today, so MongoDB, Mongo stands for humongous, which most people don't know. So humongous database that as you scale, you should feel comfortable as an AI company that you can scale with MongoDB, right?

5:06That is the whole idea, that it is a scale-out humongous database that can store lots and lots of data. Now, in terms of customers, we have seen on the AI supercycle three classes of customers, specifically React. One, Frontier Labs are using us for a multitude of use cases. So that's Frontier Labs. We cannot disclose which, what use cases. There's only a handful. I wonder who they could be. Yes, but it's labs as in plural, and we are very proud of that, that they use us and really work closely with them. Then you have AI native companies, whether they are wipe coding companies. I met a couple of robotics companies in the physical AI world who are using MongoDB.

5:54Every robot is generating data as they sense things, as they act on things, and then all that data gets stored in MongoDB. So you have digital AI or software AI and physical AI. Companies are also using MongoDB. and then you have enterprises who are creating agentic workloads and that's where we feel we are still early because i have not seen agentic applications whether you look at airlines right you look at your banking applications whichever bank you use you don't see lots of agents on your app on your phone so not seen that transition but it's still early and there is a lot of tinkering prototyping going on so those are the three classes of customers that I speak to and learn a lot on how they are using MongoDB because we do unstructured data really well we are scale-out architecture we work in multi cloud and we also can work on-prem so we are here in France yesterday I was with a customer and they said hey I just want to tell you we are not and it's a large customer based in Paris we are not going to move our workloads to public cloud We are going to run it on-prem.

7:09Data sovereignty is highest priority. French regulations forces us to have that as the highest priority. And MongoDB has to work within our on-prem environment for a lack of better term, which we do. So that's the classes of customers that we speak to. It's really interesting that you mentioned that use case. On-prem has made a bit of a comeback. Comeback, that is 100%. Sovereign AI is like, I mean, Alex Karp, you know, rang true about this the other week on CNBC, but he did say something that a lot of people were kind of thinking and a lot of CEOs of enterprises see when they're working with these labs is you have to be careful with your data and your proprietary information.

7:53But it did ring to like a larger conversation on on-prem and owning that data. So what is going on with sovereignty there? Yeah, I would say, I mean, if you look at the second half of the first decade of this century, right, that's when AWS started making inroads both in enterprises as well as in digital natives, 2008, 9, 10-ish. Then you look at the whole 2010 to 2020. It was all about public cloud, which applications are moving to public cloud, are you cloud first, what's your cloud journey. These are the conversations that will be top of mind for the CEOs, CIOs, CTOs and so on. And now most people they don't get this right.

8:41What is actually happening today is that when you speak to enterprises or even actually AI native companies, Like one of the frontier labs that we are working with, they chose MongoDB because we work in multiple clouds. We are not just a database that works in only one cloud, and they really like that for resiliency purposes. But what is happening right now, Molly, is that you see these hyperscalers, some of them, are running out of capacity. And they are running out of capacity. So this large customer in Texas, speaking to them, they have really good partnership with one of the hyperscalers. They wanted to move more workloads in cloud, public cloud, and also some AI workloads.

9:28And the hyperscaler said, sorry, we don't have a capacity. And they are one of the top 50 customers for that hyperscaler. So now they are saying, we were decommissioning this data center, CJ. We can do that. So we are going to now run workloads on-prem and we are going to run some AI workloads, whether it's data privacy issues or other issues. We are going to run those workloads and this is not a regulated industry customer. They're just a large customer in Texas. So this phenomena, whether it's this particular firm, which is Fortune 100 in Texas, or whether they are banks or healthcare, data centers are back.

10:08Nobody is now saying, hey, especially in the Fortune 500, and most people get that wrong because everybody thought, hey, all the workloads are just going to move to these hyperscalers. So hyperscalers running out of capacity and making sure that the proprietary data stays within your own premises is driving the data center and on-prem demand. I didn't know that. I didn't know people were getting pushed out. Yes, that is correct. Holy crap. What else is happening? What else do you know? The other, you know, large telecommunications company in the United States was also similar. They had one particular hyperscaler, and they said this hyperscaler refused to give them additional regional capacity that they needed.

10:54So they signed with the second hyperscaler, and now they have to manage across multi-cloud their application real estate, which adds complexity for them. That sounds like a nightmare. Yeah. So for people who are not familiar with MongoDB, which I'd be very, very surprised, what is your differentiation between the likes of Snowflake and Databricks and the other people out there? So MongoDB is a operational database or a real-time database. So credit card transactions or anything real-time, that's what we do. The category is called online transaction processing or OLTP but we are a document database, modern database that was created in 2007 so we are 19 years old.

11:41Database industry MOLLE has existed for 60 plus years. And regardless of the internet era, mobile era, after iPhone, now the AI era, you always need a data layer. So if you want real-time data layer or online data layer, that's MongoDB and some other databases. If you want analytical data layer where you can ask a question, a business analyst internally will ask a question. For those kind of use cases, you use those other companies' databases. They are called online analytical processing. So they are not real time and we are real time. This episode is brought to you by Brex, my favorite. You become what you spend on.

12:26And I refuse to spend my time on work that shouldn't exist. Expense reports, receipt chasing, and manual closes. The companies building what's next from Vercel, OpenAI, Anthropic, Granola, and Deepgram all made the same call. They all run on Brex. Brex is the intelligent finance platform that combines cards, expenses, and banking into a single stack with agentic finance built in. AI agents that handle expenses automatically, enforce policy before spend happens, and close your books in minutes. That's why Sorcery runs on Brex, so I can spend time on building and not busy work. It's time to get Brex AF.

13:08Learn more at brex.com slash sorcery. That's B-R-E-X dot com slash S-O-U-R-C-E-R-Y. Bye. What excites you most about what's happening in AI right now and how you come up as the leader? Yeah, so when I see the Frontier Labs using us for different use cases, and they are creating agents, so one of the Frontier Labs, they use us for inference, they use us for certain voice products they have, they have also video and image creations going on, so they use us for that. That gives me a lot of confidence that we have the right architecture, that we can save all this unstructured data very efficiently, real time.

13:55We can act as a memory layer. So Frontier Labs is like, I would consider the holy grail for us that really we understand because their scale and how fast they are growing is like not even linear in some of the cases. when I see weekly active users or daily active users, it's almost literally a vertical spike like this, right? So that's one. But even when we look at the 11 lab story, I mean, it is a lab. So you could argue that's a frontier lab, great success story out of here in Europe, right, out of London. And they have north of 50 million agents, depending on when you look at it, all running on MongoDB.

14:37So that also gives us a lot of confidence that we have the right architecture for agentic workloads. And then coming back to Fortune 500 or Global 2000, of course, we have the right architecture. We have embedding pieces, vector pieces, a lot of cool technologies in our Atlas database or on-prem. And that's where it's still a lot of experimentation, but I have not seen end customer facing agentic applications run at scale. So very positive. You know, optimism nowadays is in short currency from my perspective. And when I see how things are moving fast, you know that you always will need that data layer.

15:20And we are very well positioned for this AI era. What do you think of the data labeling companies that are doing all the reinforcement learning behind the scenes? I mean, those are great companies. We don't compete with them. You know, they have their own market for it. Our use cases tend to be typically, whether it's an AI native company or whether it's a frontier lab or whether it's an enterprise, customer-facing agentic applications or just customer-facing applications. If you're a telecommunications company, whether you look at your online app, see the usage, whatever the case might be, that's where we get used.

15:58Food delivery places like Zomato in India uses MongoDB, Atlas. So we have a variety of use cases from travel, entertainment, many media companies use us as well. And those labeling companies is just complimentary. They grow very, very fast. They do also grow very, very fast. Yes, yes. It's exponential. Correct. So you have a very interesting background and story and career progression, which we'll say. We were going to say something else, but we're going to go with career progression. Sounds good, yes. Could you share a little bit more on your story and how you became CEO? Yes, of course. So I started at a different database company.

16:44So when I graduated from college in Illinois, I started with Oracle. And Oracle was a super high-growth company, a really cool company to work for, a very technical company, very technical products, and so on. So throughout my career, I have always focused on products and engineering. That has been my job, either software engineering, running engineering teams, or creating products, or sometimes doing both. Over time, I got involved in go-to-market teams and understanding the product-market fit and how we scale products in Global 2000, or sometimes create new products and understanding customers pain points so that you can create the right products for them where we feel we have a right to play so that's been my career throughout my career always been product engineering infrastructure MongoDB is the first time I became a CEO which was a few months ago in November David Tacharya is there for 11 years went through cloud transition and they did a really good job and he said CJ make sure that now you go through the AI transition properly well but Dave and the team did a phenomenal job so first time CEO but I will not change a lot of things that I used to do in the past which is what I shared whether it's meeting customers all the time really understanding how they are using the product if they don't use some of our features why not if they use our features and they love them why do they love them and what other pain points that they are dealing with right now so that we can really help them so from a career progression standpoint now CEO for eight plus months absolutely enjoying the job and being with customers who are driving innovation like even just today we talked about cognition but just talking to Scott and understanding what's his vision and how he's helping with migrations with financial services company whether it's database migration or software migration or software upgrades and the potential for MongoDB and cognition to really work well together those are very exciting opportunities the same thing with Frontier Labs so the transition has from my perspective into the CEO role has not change what I really value being close to the customers and building great teams.

19:18I mean you picked I don't know if you picked it but you joined as CEO at one of the most chaotic times ever which I would assume is one of the most fun and energizing times. So how do you like what is your strategy like how you come in every week and like do you have a plan do you work on a plan like What does the roadmap look like for you? Yeah. You know, one of the advice I got from somebody who was a great CEO before told me, CJ, make sure that things that you do, only CEOs can do. Right? Don't try to spend time on things that others can do. But there are things that only a CEO can do. Right?

20:09Whether it's spending time on strategy, working with the partners, making sure you stay close to the customers, largest customers, fast-growing customers, and so on. I would say the pace of innovation and the compounding change that is happening with customers. I have seen the internet transition, right, internet applications that got developed late 90s, early 2000s, then the mobile era, cloud era, now the AI era. I have never seen a pace like this. And customers have a lot of questions. And as you see, Molly, over the last eight weeks, a lot of announcements from hyperscalers on forward-deployed engineers.

20:53They are using Palantir's playbook from my perspective. right and the reason is because customers are going through so many changes they need help and that's what they are realizing and they see that as an opportunity every hyperscaler made an announcement in last i think eight weeks to do that so what i'm seeing is i have never seen it is truly unprecedented in terms of changes happening with customers but even the innovation speed. When software writes software, right, the incremental cost of innovation has gone down, right, because software is writing software. It's kind of crazy that people don't have to write software and machines can just write software.

21:38And then number two, the cost of experimentation is also very low because you can say, hey, let me figure out I should create a product in this area or should I innovate here or can I create an agent for this? You know, I am personally creating a few agents so that I can do a better job of managing my own calendar, my own time where I shouldn't be spending time, things like that. So this level of change with our customers is unprecedented and companies are trying to move fast. And I think the speed is paramount right now. and so every day I wake up I said yes I'm not gonna plan okay what is our three-year roadmap what is our two-year roadmap on product innovation but even customers are pivoting really fast like one of the large banks I was with a couple of weeks ago in New York City we saw agent so I said can you share with me your agentic architecture how are you building agents like show me all the way from hyperscaler databases agents runtime models you are using so they showed me this this particular picture had I want to say 55 different boxes for the agentic architecture different type of LLM's frameworks databases vector databases whatever it is embeddings this that and so on I'm like wow that's a lot and I said how would this picture would have looked a year ago they said half of these boxes would have not been there because we realize in trying to create agents observability, security, how you do eval, how you move things in production, guardrails, whatever the cases are and I said do you expect this boxes to change or you may add boxes and they said absolutely.

23:30I mean this is what we are dealing with in terms of AI adoption in the enterprises and then once the All those things are figured out, like happened with cloud in 2010, 2012, 2014. Then you will see inference take off. So this is the slowest it'll ever be. This is the slowest. I think it is going to get even faster, right? Yes, 100%. So taking the operator mindset into one of the fastest moving positions, what did you kind of have to shed and how do you help the team navigate that speed? Yes, one of the things that I would say, Molly, is that you really have to, at the end of the day, customer-driven innovation.

24:18It doesn't really matter. You can't come and think of if I build it, they will come. That era is gone. So you really have to understand what customers want. We sell to enterprises, we sell to AI native companies, we sell to frontier labs. And when they tell us either their pain points or scaling points, whatever issues they are having. This, I'll give you an example. One of our large customers, they are in prediction market. And yes. Interesting. Yes. And they, because of World Cup soccer, obviously they're doing very well. Lots and lots of predictions happening every minute. And the CEO reached out to me and he said, CJ, it looks like we may be out of capacity in a few days on MongoDB.

25:06And the World Cup soccer is still a couple of weeks away, meaning from finishing. We really need help because we can't be down, right? And you understand. And then when he gave that feedback saying, can you help? So the team got formed within minutes. And we said, yes, of course. We know you guys are doing really well. and this is CEO he's also a founder who reached out and I said to him okay let's figure out how we can help you and he's like Molly stressed out about like disk space I need this much this many terabytes otherwise we are going to be down and it like telling me all those specific details via text and I said okay we got it we'll figure it out but in that process you learn a lot that for somebody scaling at that level, right?

25:54We also sometimes see that with Frontier Labs, right? Frontier Labs, as they release a new version, and if they are building on MongoDB, they will tell me, hey, I thought I had more capacity. I want to scale out. Oh, I'm running out of capacity in AWS, so I want to move some of MongoDB to GCP. How do we make it easy for them? And what I notice, Molly, with all these folks, whether it's prediction market or AI native companies or They just do not have people, actual people to deal with infrastructure software because they are so focused on their models, rightfully so, or the actual prediction market software.

26:34The infra software, which is the foundation on which they need to scale, whether it's the hyperscalers, whether it's us, database providers, they are like, can you just make this thing auto scale? or why am I calling you and telling you that I'm going to run out of capacity? Why couldn't you see it? I said, hey, you are going through unprecedented growth that we have never seen, whether it's World Cup soccer or whatever the case might be. So there is only so much we can do, but that's what you learn. So then you also need to innovate super fast and make it super easy for customers. So what's the solution?

27:13There is no solution. you constantly have to pivot or change the roadmap, right? You have to say, this is becoming a priority. Customers do care about this particular thing. They do not have, you know, in database world, when I was at Oracle, the most expensive person you can get is called Oracle DBA, database administrator. And they were the most expensive people that you have to hire when you're running Oracle. And now these AI companies are saying, we don't have time to hire these people. If we are using your software it should have its own machine-based database administration. So the point being the solution is you learn, you pivot fast and you figure out how do you really help them auto scale, be proactive in how you tell them what's going on and make a decision for them because you cannot rely on this expensive human being that they will hire and have somebody manage them.

28:18It's really cool and also very interesting that you go so deep with the customers that you learn how they build agents. What are the biggest lessons that you learn from their agent architecture? Yeah, so one of the things that what I'm seeing is, first, I had a very simple viewpoint of the world from a customer perspective. Oh, are you standardizing on this particular model? Okay, from say OpenAI or Anthropic or whoever it is. And then as you really understand their architecture, you find out for certain use cases, they may have domain-specific large language model. For certain use cases, they may have open source models.

29:06For certain use cases, they have small language models. For certain use cases, they may use Cloud or OpenAI, you know, 5.5 or whatever the case might be. So I learned that there is no, like, the one thing. There is no, like, standardization even when it comes to these models, right? It goes anywhere from open source to closed source, small to large, horizontal to domain-specific based on use cases. So that was a big learning. Second is, do you get deterministic outcome? I was with a large insurance company that's a MongoDB customer. And first I said, hey, it must be really annoying to the CIO that people talk about agents.

29:53And the first insurance agent was a thing that your industry created. So it must be really confusing when you guys talk about agents. And he said, yes, it's really confusing and very annoying. and then CJ you are stating the obvious that's even more annoying. So I said I get it. So he said to me that for insurance industry when you give a quote or a policy they have to do state by state in the United States. There are a lot of regulation. They have to make sure these agents don't give something. There is no bias involved and give a deterministic outcome. And do you need human in the loop? That if I want to add real story, my son who just started learning driving, if I want to add policy, in my policy, add my son as a dependent and a new driver, how does that work automatically?

30:46And if agents can make that decision and say, CJ, okay, your son just started learning, you're in the state of California, here is how much would it cost, can I audit that, and can I have still human in the loop so that we are following all the regulation for state of California that agents didn't have bias or this agent was competing with the other agent. So when you really try to understand what they were solving for, like, wow, okay, this is a lot more complex than just some prototype that you can do in agentic world. Wow. Yes. Huh. That's crazy. I mean, a lot of the main chatter and debates right now are on the open and closed source topic.

31:32And the one thing that we're hearing over and over again is the increased use of open source because as these things get standardized, they realize they don't need the fanciest model beneath them. And so it'd be really interesting to see, I don't know if you're tracking the kind of conversations that you're having of how that changes over time and what model they're using. Correct. I would say it is still a very mixed world in terms of that. Like I don't get this strong opinions from customers that we must use open source everything when it comes to models or databases or the software infrastructure, Linux, versus all the way from operating system to the top of the layer.

32:19and the other thing is what are the true building blocks and if you are using open source how do you get support for it you do not get support for it like for coding agents okay you have cloud code you have codex there is a grog build many options cursor has other options available so what you find is in those scenarios you almost always use see or hear closed models not open source models, proprietary models. But in certain cases, people will tell me, customers will tell me, hey, we are using something from Hugging Face and this particular open source model, which was very domain-specific or use-case-specific.

33:03So it's still a mixed world out there. I have not seen a complete pivot towards open source or a complete pivot towards proprietary. If you're building what's next in AI, you need to know MongoDB, the database platform developers love and built for the agents you're running. MongoDB stores, searches, and reasons over your data in real time with vector search and embeddings from Voyage AI all in the same system. No separate pipelines, no stitching together 10 different tools. It's why 75 % of the Fortune 100 and leading AI native startups run on MongoDB. Build and scale from your first user to billions of vectors.