Resilience Universe with Sanjay Poonen | Chirantan “CJ” Desai

Great to see you. Great to see you.

Six months.

I joined MongoDB six months ago. After ServiceNow, I was at Cloudflare, and after Cloudflare, I joined MongoDB to be first-time CEO, like you are, and we're doing it together. MongoDB in 2007, the company was created in 2007 and the fundamentals were very simple. Relational database is very rigid. The Oracles of the world and SQL servers of the world are very rigid way to represent data. You should represent data in a natural way as in a document model. And cloud was just emerging as in AWS public cloud. So can you leverage data and scale data in a scale-out fashion rather than scale-up? Oracle and all other architectures are scale-up architectures that you can run on commodity hardware and you could be cloud agnostic. I mean literally that was the thing.

Yes, very similar to your story. And unstructured data, as you know Sanjay, you guys do this every day, continues to grow at a much higher rate than structured. And AI continues to have a lot of unstructured data. I want to create an image, I want to create a video, whatever it is. All of those things, we are the default data platform for many companies and we want to be the default data platform in the agentic world as well as how we have been in the application world.

Yeah, we go back and forth a lot internally, and I'm sure you do too. How do we segment customers in this AI transformation? So we have 65,200 customers. We have a product-led growth motion which helps us with acquisition of those customers in a very effective way. You have frontier model companies, there are very few, you can count them on fingers. Then you have AI-native companies. So these are the companies truly, Sanjay, built after 2022, ChatGPT era, where on top of LLM, whichever LLM you use, you build something, right? So those are AI-native companies. Then you had the digital natives, right here in the Bay Area, many companies that got created in early 2000s post-mobile era and so on. And the enterprises, but enterprises is definitely where MongoDB is used for mission-critical workloads, whether it's payments, commercial banking, healthcare, insurance policies, public sector. So within the enterprises, I would say banking or financial services, healthcare, tech, and then the fourth one I would put it as public sector.

Yeah, I mean the way I see it is, and you know, when I understand why they are using us and for what specific use case, there are two ways to do that. Because even if you're training a frontier model, there's so much unstructured data that you feed into LLM so that LLM can provide you better answers. So some frontier model companies use us for research because they put lots and lots of training data and then the new model comes out. Versus there is a frontier model company that uses us for inference. So you ask a question, create an image of Sanjay and CJ standing together. It creates an image and then you go back and forth and it keeps that long-term memory. So we have research as well as inference depending on who it is.

So first, I do want to say thanks to your teams. Even before the partnerships, for our joint customers, which really matters, is that Helios control plane is built on top of MongoDB. And MongoDB, and that is the brain behind all the different backup silos as you outlined. And that really, really means a lot to us that if we are part of the control plane on Helios, that just tells us that we are the most mission-critical even for you on how you serve. We are a customer, you're great, and we embed that everywhere we go.

Number two on the customers, Sanjay, it's very simple. Initially we started out, 2007, 2008, MongoDB, when the company was created, a lot of focus was, we were boxed as a NoSQL company, unstructured data company. And hey, if it's unstructured data, this is a great database for it because of document model, scale-out, etc. But over time we have become mission-critical at some of the firms, our joint customers, where they are running, like I said, payment processing workload or insurance policy claims. All of those mission-critical workloads run on MongoDB. And so then the request was, hey, this is our horizontal backup strategy. They are always trying to improve RTO and with ransomware and all that, that you guys have been focused on, all these regulated industries, both from an operational resilience perspective but just cyber protection, the ask is always the simple: how do you work with leaders such as Cohesity and what can you do? Right? So that partnership from my standpoint just gives them peace of mind. And if it gives our customers peace of mind, it gives me peace of mind. I mean, ransomware stats, you know, they're insane right now. All the way from core data to identity, immutable is such an important point. And this regulated industry where they have to show to regulators all the RTO objectives and so on, that is where the partnership is really, really meaningful. And for me, it's just a simple statement to that CIO or CTO at a bank, she can, we can just articulate joint partnership together so that they know, hey, if I made MongoDB as my standard for retail banking, and oh by the way, our horizontal standard for backup, restore, cyber res is Cohesity, how do you guys work together? And I should be able to answer that very simply.

I know. And Sanjay, you know, here is the interesting piece that I'm seeing. If you look at frontier model companies, even last week's announcement by somebody like Anthropic, they are now moving their workloads on-prem. And this is very interesting. You and I have worked together for many other companies in the past and we thought like 10 years ago everybody is moving to public cloud. And now when I'm speaking to customers, the joint customers we have, there is this fascination, hey, we may not move this application to actually public cloud. And we are actually expanding more. One of the large banks on the east coast, they're expanding their on-prem data center footprint even more now. And I know that's the one big advantage NetBackup had, Cohesity had, and that's the advantage MongoDB has. So I go and tell these customers, that's fine if you don't want to move to Atlas, our public consumption offering, you can use MongoDB on-prem. And I know it's the same codebase that works for Cohesity across multi-cloud.

The same words, yeah, hybrid.

In data, right? Data sovereignty. I think there's going to be some yin and yang where there's going to be data, especially in the Middle East, European countries that might have sensitivity to US public clouds, to having a sovereign cloud solution that works for them. Even banks, even Indian banks. I was talking to one of our sales teams recently and they said that the whole data sovereignty in parts of Asia is becoming even more important. And so we are not only just Europe, of course, like France, you see a lot, UK public sector, very, very meaningful way. But then you also go to Scandinavia, parts of Scandinavia, and like you said, Middle East, it almost has to work in a software environment.

Yeah. The amount of data, you know, when I was coming up to speed both on Cohesity and the Veritas data protection architecture that has evolved on top of file system over time, your file system is, you have a lot more data than we always will have. How do we make it easy to write that data, APIs and integration, and also from RTO perspective, how fast you can restore a MongoDB cluster at a bank or a public sector organization or whatever that is, like, critical. Because we are becoming mission-critical. And on the AI world, Sanjay, because of the examples I shared, now there are AI-native companies which are completely built on top of MongoDB or run on MongoDB. But even in banks, they are experimenting with agents at production where we could be the context layer or the memory layer. And that may be actually more data than the persistent data. And how Cohesity and us can partner, that would excite me for the next.

So 100% and 50% credit, 150% credit to David Tachari and the team. They thought of this idea that when you want to do agents in production, it is never an LLM issue. It is always an issue of how effectively you do your RAG. And that should be cheap enough because there is so much data everywhere. You needed to have real-time feed to give real-time answer because if I'm making a decision on your mortgage via agents, of course, I'm going to do RAG across the entire data states I have. How do we make that effective? So we call it the embedding model. So that was the acquisition done of Voyage in 2025 February. And where it's going right now is that we will help you the most effective way to retrieve the right data, and that's what then you feed with your LLM data. So even your cost of tokens go down. So phenomenal acquisition. And our vision, which already we have achieved now because it's been 15 months since that acquisition, is core operational data layer with search, vector search, and embeddings all in one so that you can build agent-take applications at scale. That's it, literally.

That's correct. GenAI app and you feed in text and now you can ask a question rather than just doing text-based search and all that. And it's all in the operational layer, as in MongoDB operational layer of real-time data. And we are really excited about this team.

Yes.

So far, really, really good. So AI-native companies, many at scale, are already using it and they are saying the quality of retrieval is extremely high and fast. And the models that we have created, embedding models, even the latest is called Retrieval something-something benchmark, I can even tell you, but based on that benchmark, we are ahead of Gemini and OpenAI models for embeddings. So for your key piece of innovation guy.

Yeah, we are excited if we can partner together. It's our partnership, early days. I had a good discussion with your Stamford team who was very smart a few weeks ago.

Yeah. And we're in early discussions there. But the problem statement is very similar, which is type a query and search unstructured data.

Right. Yeah. Because you have PDF files. I had a bank tell me just recently that, hey, we have all this stuff in PDF documents. Banks' charts and others are within that PDF document. Can I do that search effectively without having to pull everything out of that PDF document? So that's exactly right that you think about unstructured data. The problem, I think, Jensen, who I know you have spoken to also, talks about PDFs and other, which is the content repository of an organization. How do you do that search? And you have backups of pretty much every relevant Fortune 100 and 500 that matters. And that has always been the holy grail that I have all this data in the Cohesity backup. How can I leverage it?

Yeah, here is my take right now, and I may be wrong. But here is my intuition just based on looking at our joint customers piloting AI production at scale. The amount of memory or the size of the memory you require to have the context where agents can make real decisions. So how you work with us on the persistent storage layer of MongoDB, as we evolve our architecture, how you can be still the best-in-class in terms of providing protection for that data as well. So I can go back and look at why did the agent make this decision, or if something goes wrong or a tampering happened, how can we look that up? I think those are the opportunities that exist for us.

We are 5,500.

Same size.

So honest answer, which you always get from me, we are in early stages, okay? And I'll tell you my simple framework and you'll say, CJ, okay, that's not very insightful, but I'm just.

So my simple framework at a company level, because your question is at a company level, is I want four buckets, okay, on how I think about AI internally. Bucket number one, which is the lifeblood of a tech company, innovate faster, engineering.

Second, can you sell more?

Okay. So, whether it's your SDRs, how you run campaigns, how you help the seller, can you sell more? Third, once you source something, can you serve better? This is like customer support, customer success, how are you helping them serve better? And fourth is run efficiently. So across HR, IT, very simple, it's four good frameworks.

So we recently rolled out a coding assistant. And the teams have embraced it. They are giving us, you know, sometimes investors ask me, which I'm sure they ask you, what percent of CJ your code is written by AI. I'm not ready to give those kind of stats yet and I don't know if that really matters as long as it's a good.

Things moving faster. Absolutely. There are things that used to take us weeks, we can do that in days.

Right. Even the thing that you guys are all about, which is around cyber resiliency. I will look at, hey, can we find vulnerabilities internally for our MongoDB faster by doing hackathons and other using this coding assistance? 100%. So innovate faster, which is the killer use case for AI. We are looking at right now multiple KPIs and my intuition, having talked to other people who are like a year ahead of us or two years ahead of us, 20 to 30% higher productivity.

And then as a CEO, my question is very simple: does that mean I get 20 to 30% more features that we can sell to our customers, or we don't need to hire 20 to 30% more people and we can just have current team being more productive? I mean literally it's as simple as that. And I have to keep it simple otherwise you get too complicated. Third framework, because the first one is your engineering team, the second one's your sales and marketing team, the third one's your support team, and fourth is everybody.

Right, maybe the core productivity, knowledge worker, this and that.

So I was the first one among the peer group back in 2023, on the ask from Microsoft, and you work with them closely, to roll out GitHub Copilot.

And that was 2023. Today, not many companies even mention them and mention these other names. So one of the things that I told our CIO Deepa to do is that when we sign with these coding assistant companies or any AI company, I just sign, Sanjay, one year at a time.

And my team may benefit on innovate faster, they may think Codex.

I think if you go a year at a time, things are very dynamically changing.

Let's talk a little bit about the negative, everyone's asking about Mythos.

Yeah. So I do have a perspective just based on today's information because that continues to change. Is we were with a bank last week in the UK in London. Okay. And we were talking to the CIO there and he was very worried about Mythos, right? Because he said, I do not know what I do not know. Okay. So from our customer environment standpoint, they are really worried. And as you know, they have started taking this approach: we'll start with the operating system layer, Windows, Unix, Linux, the things that we used to do. And where it stays, Windows, Unix, Linux, we look at all the browser and the front-end interface and of course all the infrastructure software, whether it's MongoDB, other databases, software from you guys, we look at all of them. I can tell you without hesitation, you are going to find a part.

Yeah, time to exploit. So I think this is a real threat. And this bank individual told me that he thinks he may change the priority from innovation to cyber protection because he doesn't know what Mythos he's going to. So I don't know, I know Mythos has not been rolled out to every bank. I know some joint customers are really close stages of doing.

Thank you Sanjay and wishing you great luck. Thank you.