MongoDB .local NYC 2024 | Sahir Azam | Chief Product Officer

Clips
More from MongoDB .local NYC 2024

Sahir Azam

Chief Product Officer

MongoDB

play_circle_outline MongoDB's Developer Data Platform: Announcements, Enhancements, Generative AI, Vector Capabilities, & Partnerships with AWS, Azure, and Google

play_circle_outline Investment and updates in MongoDB Atlas

play_circle_outline Demand for private AI and data decentralization

play_circle_outline The Transformative Power of MongoDB's Document Model and API: Enabling Superior Performance, Security, and Resiliency

play_circle_outline Internal usage of AI for natural language extension and data efficiency

Info

Sahir Azam | Chief Product Officer

Sahir Azam

Chief Product Officer MongoDB

search

>> We are back at a local here in the Big Apple. Coming to you live from New York City, we're really excited to have Sahir Azam in the studio here, he's the Chief Product officer of theCUBE alum. Sahir, great to see you again.

Sahir Azam

>> Good to see you. Thank you for having

>> me.A big week.

Sahir Azam

>> Yes.

>> A big year for you guys.

Sahir Azam

>> Lots going

>> on.Wow, lots going on. I mean, the announcements are coming through, the ecosystem, the partnerships. Can you help us connect the dots on all these announcements and how do they help customers drive value from AI?

Sahir Azam

>> Yeah, there's a variety of different things we're doing. I mean, I was excited today to announce a bunch of just capabilities and enhancements across the broad developer data platform that we've built and expanded a lot over the last few years. So everything from core foundational stuff, continuously improving the performance of our platform, the security capabilities, the resiliency, which is crucially important to customers who build mission-critical applications. There's obviously new features that are newer to the market like stream processing getting to GA and really all the feedback we've gotten over the last year around that. On the AI front, I think there's really a few key areas. I think our vector capabilities as part of our platform have been one of the most fast, if not I think the fastest adopted set of features that we've added in the last few years. We're seeing a lot of buzz and excitement around customers trying to apply generative AI to their applications or build new applications from the ground up that are very AI driven. We also have been working closely with our partners like AWS, Azure, and now even Google on training their large foundational models to be really good at recommending quality MongoDB code. And then we've actually taken some of those models and implemented them in our own developer tools to make our own developer ecosystem more productive. And so there's a variety of things that we're doing, even legacy modernization, so helping AI assist people in getting off of old applications that they want to modernize, get onto a more modern technology like MongoDB. So really a lot of broad applicability. Obviously the buzz and excitement of people trying this all out after the announcement this morning is really great to see.

>> Is that legacy modernization, is that mostly code assist in migration?

Sahir Azam

>> I think over time it's going to touch the whole life cycle, but the areas that we've been focused on most directly right now first is how do we understand the pattern of queries that are going into an existing system to start helping analyze the existing schema and what it should look like in a document-oriented fashion. So actually models can help with that. We're also seeing that a lot of these applications were built perhaps decades ago where the developer teams that built them are no longer even at the company. So if you want to go modernize that app, the first thing is, how do I even understand it?

>> Where's

Sahir Azam

>> Yeah, so code analyzing and even understanding how an application was constructed is really interesting, models can assist with that process of understanding and we've definitely seen a lot of process also on the testing side. You can actually generate tests out of existing code and then reuse those same tests by refactoring them for the new application on the tail end. And a lot of what is the challenge of modernization, it's not just the raw effort, which is obviously code conversion and understanding, it's actually the trust that you actually know how to validate that things that are working well and that's where the testing comes in. So we have work going on in all these different areas. We're learning a lot. This is a new area for all of us in the industry and that is one promising area that we're spending more and more time on.

>> And just on vector, it was amazing to me, I think it was a re-invent, you been talking and there was a study, the developer uptake in vectors generally, but specifically MongoDB Vector, this is before you even went GA, it had the highest NPS and it had the highest adoption of any vector capability which-

Sahir Azam

>> Yah, and the Retool Study, which was at the front end of AI themselves, actually, which is cool.

>> Yeah, that was pretty amazing. Now I want to ask you, so you shared in the keynote that you added significant... Read performance up 60for time series queries improvement, over 40improvement of bulk loads. So how were you able to achieve that?

Sahir Azam

>> There's no one sort of silver bullet, the approach across different areas of a complex system like an operational database, we made a variety of different changes. So the first thing was obviously saying, we always, always are putting effort in performance. It's one of the key areas. The same with security, resiliency, durability obviously of data as always. It's not always the stuff I talk about on stage, the newer feature capabilities, but it's always a big part of investment. So I just want to level set there. But I was especially proud of the progress we made in this particular release. And the two things I'd call out is one on the query system, we really looked end to end. We looked at all of the different steps in the process from when a query gets submitted to how it gets planned to how it gets executed in the system. And it wasn't one change that unlocked that I think 25stat that I talked about. It was a series of small changes that when in aggregate when we looked at the result and we're doing our benchmarking in testing, we were like, Wow, this is really a big change and a good step forward.But on the other hand, something like, time series you mentioned, that was an approach that typically you see in analytics databases around block processing that we brought to time series data in an operational context which allows you to batch together updates of multiple documents at once versus versus processing each of them sequentially. So it's a technique that we apply that unlocked that gain. And so there's no kind of one thing, it's a variety of different things compounding across different areas of the platform. I really just want to take this opportunity to highlight how much of our investment goes into those areas. And I can tell you from talking to a lot of customers, it's not just the new products and new features they care about, they care a lot about raw price, performance, resiliency, security of the system for the product they already have on mission-critical applications, and that's why we wanted to really highlight that this year.

>> Now, I would say Atlas is the crown jewel. It's a big chunk of the business, two-thirds of the business now. And yet the non-Atlas stuff is actually still, it's pretty well-

>> and interest there. So two questions, a two part questions. One is help me understand the Atlas updates where you your resources there and are you seeing now, because we're hearing it from customers that they want to do kind of private AI, do things on-prem, for instance. You play in both camps, so I'm curious as to where you putting your investments in Atlas and what you're seeing in terms of that So many of the announcements, especially the stuff we just talked about, performance, resiliency, that affects all of our environments, whether that's on-premises, enterprise customers, whether it's our community, these are foundational improvements and of course we run that in Atlas, so Atlas customers benefit from it as well. So a lot of the stuff spans our entire portfolio. On the Atlas side, specifically, there are certain capabilities, newer products that are basically architect in a cloud-native way and therefore really only in Atlas. Things like stream processing, the expansion of that capability, getting it to GA some of the new feature announcements that we saw there. But anything also with our developer tooling hits both camps, the self-managed, self-hosted customers on-premises as well as those leveraging Atlas in the public cloud. So I want to be clear on most of the announcements we made are really about both sides of that. Now we are definitely hearing a lot of demand for customers who, they want private AI, now that can even be done in an Atlas context, but they also, there are applications that have data that's going to be operationalized in their own data center. And so one of the big announcements we made, an area that we're kicking off a new project on is bringing the search and vector capabilities that we've had for now, search, for years, vector for a little over a year to our Community Edition. And that in many ways is an investment in our community, making sure that the millions of builders across the world, they get started by downloading our free version of the product can get the breadth of capabilities of these AI applications. So we are very much a run-anywhere company, and I'd say we're even seeing that extend to the edge. We announced a new version of Mongo that can run in edge locations, retail stores, airplanes, boats, all of that because we are seeing data getting more distributed. And obviously the cloud is maybe the centerpiece of the architecture for a lot of these customers, but there are going to be corporate data centers we believe and some applications that are never going to move, and then there's going to be data even more distributed across the edge over time.

>> So tell me more about the edge piece of it. We had Nomic on earlier and we were just talking about the changes that are going on in inference. You're going to see a improvement in performance and at 110 the cost of running inference. I mean, it's just going to be exponential. What are you guys doing there specifically?

Sahir Azam

>> Yeah, so in particular, there's a few things. One, a lot of these are these partnerships. The AI stack is emerging. There's a bunch of innovative new companies that we've spent, Nomic being one of them, that we've spent the last 18 months getting integrated with, getting to learn what they're doing, building connections to all these amazing founders. And we actually have formed partnerships in particular with inference players, like Firefly was here I think this week with us, as well as others to make sure that all the surrounding components that you need to power a modern gen AI application works seamlessly with our data layer and data technology. Because I think it's pretty well known in the industry, you have to get your data at a good place and customer's IP, as in their data, before you can really unlock the true potential of AI. So you're seeing us really, I think, pivot to being even more ecosystem focused, even more technology integration, partnered focus so that we can empower these new inference engines or these new model or embedding providers like Nomic to work well with our data that people have been building applications on for almost now 15 years.

>> I want to ask you, thank you for that, by the way, I want to ask you about the relational model versus the document model. Explain to me why an API from relational is not good enough.

Sahir Azam

>> Sure. Yeah, I think it's a combination of both the data model and the API and certainly there are a lot of relational databases. A big portion of the ecosystem is going to always have relational databases. We're not crazy, but we've focused on really saying why does MongoDB exist? It exists because a lot of modern programming languages, first and foremost, are object-oriented. And in a relational database paradigm, the data model of rows and tables in database was built at a time where we were optimizing as an industry for expensive hardware. So everything was about storing as efficiently on disk. Fast-forward 30 years later to when MongoDB was coming around, hardware was, as we all know, coming down in prices, storage is now extremely cheap. Compute's gone down significantly. And so the cost of infrastructure is no longer the constraint, it's the cost and the time of the developers and the people that's the expensive piece. So our founders saw that the constraints had changed, and so therefore they were saying, How do we persist data in a way that's much more natural for developers because now we can store all the data that gets accessed together cheaply,and that was never really possible with the hardware paradigms back when relational databases were invented by IBM all those years ago. And so that means that now object-oriented programming languages, developers working with formats of data like JSON, which is like the lingua franca of the modern internet, are much more natural in terms of how you store that in this document-oriented structure. And that really frees developers to just move extremely quickly. We also can do things like schema adjustments, schema migrations without having to take the database down in an always-on digital experience world, that's really important. That's kind of one aspect of why the data model and then also the API... one of the things when we talk to developers is, What do you think about the MongoDB query language?And they're like, I don't think about the query language because for me it disappears into the code that I'm writing because it's so natural and idiomatic.Versus SQL, which is a whole different language you have to learn and inject in the middle of another language as you're coding. So that's the other aspect of why developers love MongoDB the experience around it. It's those two fundamental things

>> combined.Pretty ingenious when you go back to the founding premise and the whole open source community. Of course at the time people were like, Ah, I won't scale.Well, you've proven that wrong.

Sahir Azam

>> Yeah. And building an operational database system that not just can serve lightweight applications but can scale, deal with the security, the resiliency, the durability guarantees to be a general purpose database takes hundreds of millions of dollars of investment many years, and that's what we've been very fortunate to have a very long-term oriented management team board and also investor community that has saw the potential of us and what we've been able to deliver to market.

>> In the little analyst breakout, I asked you a question, I'd love to get it on the record. So I basically asked a question that was sort of a snarky question, but hey, RAG is really easy. It's simple. It's not driving tons of ROI. What do you see in terms of... First of all, how you applying RAG internally because you're learning things, and how you helping your customers get to bigger dollars?

Sahir Azam

>> Yeah, definitely. So in terms of internal usage, we have a variety of different use cases. I'm sure there's some that I'm not even aware of internally in but a couple that I can call out. We built our own sort of RAG framework as well as our own vector database was used for this to enable a natural language extension to our documentation. MongoDB is a very popular technology, our documentation is one of our most trafficked sites because developers download the technology, sign up for Atlas, they want to learn from our examples. So now it's just much easier to use these AI-powered ways of finding the right answer for whatever problem you're trying to solve. That was a RAG use case internally at MongoDB. But we also have a bunch of other use cases now that help drive efficiency in the way we support customers, from our technical support teams to our customer success teams where we're accessing a bunch of different information. So not just information in MongoDB, but information from our corporate systems, our CRM, our internal knowledge base where we store support data and all of that. And so we had to really think about with our CISO and our CIO, what is the right governance structure so we can enable APIs from any applications, but that can maintain and make sure that developers who are building all these systems internally and these apps internally aren't accidentally exposing information about whatever, it could be one employee's salary versus another, it could be some sensitive information about a customer. All of this needs to be protected. And so that's where RAG as a concept or proving out a simple use case can be pretty straightforward, and I agree with that. But doing it across many use cases in an environment where you have sensitive information, you got to govern that, that is actually quite challenging. And so we've learned a lot about that, and now actually taken that knowledge and some of those technology partners that we used integrated them into our partnership program, and now we're actually helping customers go down at a similar journey.

>> No shadow RAG, That's dangerous stuff.

Sahir Azam

>> It's a tricky balance, to one hand, we're a very innovation focused, bottoms-up kind of culture at MongoDB. So the goal has always been not to say no, but to say, here are your tools as fast as possible that are safe, but go at it, once you're within these constraints, go at it on the innovation, the hackathons, and it's been great.

>> And I'm kind of joking about shadow, but you think about the Hadoop era, the big data, that we're just experimenting and then a year and a half, two years later, the organization would say, Well, wait a minute. We have to govern this And so you're getting out ahead of that saying, Here's a tool set that you If you're going to experiment, these are the boundaries that you think, with AI in particular, we're seeing that customers in the enterprise especially they understand this, they're very much intrigued by the potential, all are in terms of where some of the use cases that are going to be possible, whether it's driving efficiency and improving the bottom line, or whether it's creating new business models, new customer experiences. I think everyone's really very much a board-level kind of thing around that. But at the same time, established brands have a reputation to protect and customer data and personal information to make sure it doesn't leak into the wrong places or that they're recommending the right best practices because they're also being very careful and cautious, which is making this governance and ethics around how these things get applied. But definitely is an important topic.an impossible to answer question, but I'll ask it anyway. And it's related to the future of apps, how do you see the future of apps? If I asked that question of you in 2000, you would've seen Salesforce and you may have predicted some kind of SaaS model emerging, but you probably wouldn't have predicted the iPhone, maybe. But what kind of framework should we think about in terms of the characteristics of the app in the future, the intelligent data app, if we can call it that. How should we think about the future of applications?

Sahir Azam

>> Yeah, I think it is a hard question, but I can proxy some of the stuff that I've been reading or watching or hearing from some of the experts and just our ecosystem and all of that. One, I do think the long-term potential is absolutely something we believe And I think the hype is warranted in terms of where things will go over the long term. But as you know, we've all seen multiple technology cycles. There's going to be some hype and then there's some disillusionment when things don't deliver against that hype. But over the long term, things really do get there if not exceed our expectations. And I think AI is going through its own journey around that. I don't want to try to predict where we are on that curve, but I definitely think-

>> Early, early.

Sahir Azam

>> Early is definitely the thing. I saw one of our investors, does an AI conference and was amazing talk where they compared when the iPhone came out. And when the iPhone first came out, it was a really great presentation. They're like, The first wave of apps were pretty simplistic,the flashlight app, those utility apps that had some value, but they were quite small. But only a few years later you saw the DoorDashes, Ubers, the Instagrams, these applications that created completely new business models or experiences that we couldn't conceive of. And I think with AI, we're seeing glimpses of that, but I don't think anyone can really say where is things really going to go, who are the winners? But I know it's coming, and so I think that's exciting. I definitely think also that one of the things we're looking at that's promising is the whole idea of agentic workflows. Like how does, not just a model solve a particular narrow problem, but how can it help one orchestration agent do more complex business workflows over time? And you've see some interesting demos come out from some startups of I think where things can go, very early, but we're excited about the potential of that and are watching that space as well the other thing I would add to that excellent answer would be the impact on industry. I mean, you think about how mobile banking just completely changed the which we interact with our financial institutions. Stock trading, even though you had internet trading, now, it's just like, Boom, boom, Robinhood.

Sahir Azam

>> And the power and ability to reason about all this unstructured data truly in a way that wasn't really mainstream or possible, except for maybe the pockets of the largest, most sophisticated tech companies in the world, that's going to be really empowering and democratizing for any organization that's building software with all the technologies that are emerging and that's really exciting.

>> Sahir, amazing interview as always, thanks so you. Thank you for having

>> us.You're very welcome. And thank you for your attention here. We'll be back right after this short break. MongoDB live from New York City. You're watching theCUBE.

Sahir Azam | Chief Product Officer

search

>> We are back at a local here in the Big Apple. Coming to you live from New York City, we're really excited to have Sahir Azam in the studio here, he's the Chief Product officer of theCUBE alum. Sahir, great to see you again.

Sahir Azam

>> Good to see you. Thank you for having

>> me.A big week.

Sahir Azam

>> Yes.

>> A big year for you guys.

Sahir Azam

>> Lots going

>> on.Wow, lots going on. I mean, the announcements are coming through, the ecosystem, the partnerships. Can you help us connect the dots on all these announcements and how do they help customers drive value from AI?

Sahir Azam

>> Yeah, there's a variety of different things we're doing. I mean, I was excited today to announce a bunch of just capabilities and enhancements across the broad developer data platform that we've built and expanded a lot over the last few years. So everything from core foundational stuff, continuously improving the performance of our platform, the security capabilities, the resiliency, which is crucially important to customers who build mission-critical applications. There's obviously new features that are newer to the market like stream processing getting to GA and really all the feedback we've gotten over the last year around that. On the AI front, I think there's really a few key areas. I think our vector capabilities as part of our platform have been one of the most fast, if not I think the fastest adopted set of features that we've added in the last few years. We're seeing a lot of buzz and excitement around customers trying to apply generative AI to their applications or build new applications from the ground up that are very AI driven. We also have been working closely with our partners like AWS, Azure, and now even Google on training their large foundational models to be really good at recommending quality MongoDB code. And then we've actually taken some of those models and implemented them in our own developer tools to make our own developer ecosystem more productive. And so there's a variety of things that we're doing, even legacy modernization, so helping AI assist people in getting off of old applications that they want to modernize, get onto a more modern technology like MongoDB. So really a lot of broad applicability. Obviously the buzz and excitement of people trying this all out after the announcement this morning is really great to see.

>> Is that legacy modernization, is that mostly code assist in migration?

Sahir Azam

>> I think over time it's going to touch the whole life cycle, but the areas that we've been focused on most directly right now first is how do we understand the pattern of queries that are going into an existing system to start helping analyze the existing schema and what it should look like in a document-oriented fashion. So actually models can help with that. We're also seeing that a lot of these applications were built perhaps decades ago where the developer teams that built them are no longer even at the company. So if you want to go modernize that app, the first thing is, how do I even understand it?

>> Where's

Sahir Azam

>> Yeah, so code analyzing and even understanding how an application was constructed is really interesting, models can assist with that process of understanding and we've definitely seen a lot of process also on the testing side. You can actually generate tests out of existing code and then reuse those same tests by refactoring them for the new application on the tail end. And a lot of what is the challenge of modernization, it's not just the raw effort, which is obviously code conversion and understanding, it's actually the trust that you actually know how to validate that things that are working well and that's where the testing comes in. So we have work going on in all these different areas. We're learning a lot. This is a new area for all of us in the industry and that is one promising area that we're spending more and more time on.

>> And just on vector, it was amazing to me, I think it was a re-invent, you been talking and there was a study, the developer uptake in vectors generally, but specifically MongoDB Vector, this is before you even went GA, it had the highest NPS and it had the highest adoption of any vector capability which-

Sahir Azam

>> Yah, and the Retool Study, which was at the front end of AI themselves, actually, which is cool.

>> Yeah, that was pretty amazing. Now I want to ask you, so you shared in the keynote that you added significant... Read performance up 60for time series queries improvement, over 40improvement of bulk loads. So how were you able to achieve that?

Sahir Azam

>> There's no one sort of silver bullet, the approach across different areas of a complex system like an operational database, we made a variety of different changes. So the first thing was obviously saying, we always, always are putting effort in performance. It's one of the key areas. The same with security, resiliency, durability obviously of data as always. It's not always the stuff I talk about on stage, the newer feature capabilities, but it's always a big part of investment. So I just want to level set there. But I was especially proud of the progress we made in this particular release. And the two things I'd call out is one on the query system, we really looked end to end. We looked at all of the different steps in the process from when a query gets submitted to how it gets planned to how it gets executed in the system. And it wasn't one change that unlocked that I think 25stat that I talked about. It was a series of small changes that when in aggregate when we looked at the result and we're doing our benchmarking in testing, we were like, Wow, this is really a big change and a good step forward.But on the other hand, something like, time series you mentioned, that was an approach that typically you see in analytics databases around block processing that we brought to time series data in an operational context which allows you to batch together updates of multiple documents at once versus versus processing each of them sequentially. So it's a technique that we apply that unlocked that gain. And so there's no kind of one thing, it's a variety of different things compounding across different areas of the platform. I really just want to take this opportunity to highlight how much of our investment goes into those areas. And I can tell you from talking to a lot of customers, it's not just the new products and new features they care about, they care a lot about raw price, performance, resiliency, security of the system for the product they already have on mission-critical applications, and that's why we wanted to really highlight that this year.

>> Now, I would say Atlas is the crown jewel. It's a big chunk of the business, two-thirds of the business now. And yet the non-Atlas stuff is actually still, it's pretty well-

>> and interest there. So two questions, a two part questions. One is help me understand the Atlas updates where you your resources there and are you seeing now, because we're hearing it from customers that they want to do kind of private AI, do things on-prem, for instance. You play in both camps, so I'm curious as to where you putting your investments in Atlas and what you're seeing in terms of that So many of the announcements, especially the stuff we just talked about, performance, resiliency, that affects all of our environments, whether that's on-premises, enterprise customers, whether it's our community, these are foundational improvements and of course we run that in Atlas, so Atlas customers benefit from it as well. So a lot of the stuff spans our entire portfolio. On the Atlas side, specifically, there are certain capabilities, newer products that are basically architect in a cloud-native way and therefore really only in Atlas. Things like stream processing, the expansion of that capability, getting it to GA some of the new feature announcements that we saw there. But anything also with our developer tooling hits both camps, the self-managed, self-hosted customers on-premises as well as those leveraging Atlas in the public cloud. So I want to be clear on most of the announcements we made are really about both sides of that. Now we are definitely hearing a lot of demand for customers who, they want private AI, now that can even be done in an Atlas context, but they also, there are applications that have data that's going to be operationalized in their own data center. And so one of the big announcements we made, an area that we're kicking off a new project on is bringing the search and vector capabilities that we've had for now, search, for years, vector for a little over a year to our Community Edition. And that in many ways is an investment in our community, making sure that the millions of builders across the world, they get started by downloading our free version of the product can get the breadth of capabilities of these AI applications. So we are very much a run-anywhere company, and I'd say we're even seeing that extend to the edge. We announced a new version of Mongo that can run in edge locations, retail stores, airplanes, boats, all of that because we are seeing data getting more distributed. And obviously the cloud is maybe the centerpiece of the architecture for a lot of these customers, but there are going to be corporate data centers we believe and some applications that are never going to move, and then there's going to be data even more distributed across the edge over time.

>> So tell me more about the edge piece of it. We had Nomic on earlier and we were just talking about the changes that are going on in inference. You're going to see a improvement in performance and at 110 the cost of running inference. I mean, it's just going to be exponential. What are you guys doing there specifically?

Sahir Azam

>> Yeah, so in particular, there's a few things. One, a lot of these are these partnerships. The AI stack is emerging. There's a bunch of innovative new companies that we've spent, Nomic being one of them, that we've spent the last 18 months getting integrated with, getting to learn what they're doing, building connections to all these amazing founders. And we actually have formed partnerships in particular with inference players, like Firefly was here I think this week with us, as well as others to make sure that all the surrounding components that you need to power a modern gen AI application works seamlessly with our data layer and data technology. Because I think it's pretty well known in the industry, you have to get your data at a good place and customer's IP, as in their data, before you can really unlock the true potential of AI. So you're seeing us really, I think, pivot to being even more ecosystem focused, even more technology integration, partnered focus so that we can empower these new inference engines or these new model or embedding providers like Nomic to work well with our data that people have been building applications on for almost now 15 years.

>> I want to ask you, thank you for that, by the way, I want to ask you about the relational model versus the document model. Explain to me why an API from relational is not good enough.

Sahir Azam

>> Sure. Yeah, I think it's a combination of both the data model and the API and certainly there are a lot of relational databases. A big portion of the ecosystem is going to always have relational databases. We're not crazy, but we've focused on really saying why does MongoDB exist? It exists because a lot of modern programming languages, first and foremost, are object-oriented. And in a relational database paradigm, the data model of rows and tables in database was built at a time where we were optimizing as an industry for expensive hardware. So everything was about storing as efficiently on disk. Fast-forward 30 years later to when MongoDB was coming around, hardware was, as we all know, coming down in prices, storage is now extremely cheap. Compute's gone down significantly. And so the cost of infrastructure is no longer the constraint, it's the cost and the time of the developers and the people that's the expensive piece. So our founders saw that the constraints had changed, and so therefore they were saying, How do we persist data in a way that's much more natural for developers because now we can store all the data that gets accessed together cheaply,and that was never really possible with the hardware paradigms back when relational databases were invented by IBM all those years ago. And so that means that now object-oriented programming languages, developers working with formats of data like JSON, which is like the lingua franca of the modern internet, are much more natural in terms of how you store that in this document-oriented structure. And that really frees developers to just move extremely quickly. We also can do things like schema adjustments, schema migrations without having to take the database down in an always-on digital experience world, that's really important. That's kind of one aspect of why the data model and then also the API... one of the things when we talk to developers is, What do you think about the MongoDB query language?And they're like, I don't think about the query language because for me it disappears into the code that I'm writing because it's so natural and idiomatic.Versus SQL, which is a whole different language you have to learn and inject in the middle of another language as you're coding. So that's the other aspect of why developers love MongoDB the experience around it. It's those two fundamental things

>> combined.Pretty ingenious when you go back to the founding premise and the whole open source community. Of course at the time people were like, Ah, I won't scale.Well, you've proven that wrong.

Sahir Azam

>> Yeah. And building an operational database system that not just can serve lightweight applications but can scale, deal with the security, the resiliency, the durability guarantees to be a general purpose database takes hundreds of millions of dollars of investment many years, and that's what we've been very fortunate to have a very long-term oriented management team board and also investor community that has saw the potential of us and what we've been able to deliver to market.

>> In the little analyst breakout, I asked you a question, I'd love to get it on the record. So I basically asked a question that was sort of a snarky question, but hey, RAG is really easy. It's simple. It's not driving tons of ROI. What do you see in terms of... First of all, how you applying RAG internally because you're learning things, and how you helping your customers get to bigger dollars?

Sahir Azam

>> Yeah, definitely. So in terms of internal usage, we have a variety of different use cases. I'm sure there's some that I'm not even aware of internally in but a couple that I can call out. We built our own sort of RAG framework as well as our own vector database was used for this to enable a natural language extension to our documentation. MongoDB is a very popular technology, our documentation is one of our most trafficked sites because developers download the technology, sign up for Atlas, they want to learn from our examples. So now it's just much easier to use these AI-powered ways of finding the right answer for whatever problem you're trying to solve. That was a RAG use case internally at MongoDB. But we also have a bunch of other use cases now that help drive efficiency in the way we support customers, from our technical support teams to our customer success teams where we're accessing a bunch of different information. So not just information in MongoDB, but information from our corporate systems, our CRM, our internal knowledge base where we store support data and all of that. And so we had to really think about with our CISO and our CIO, what is the right governance structure so we can enable APIs from any applications, but that can maintain and make sure that developers who are building all these systems internally and these apps internally aren't accidentally exposing information about whatever, it could be one employee's salary versus another, it could be some sensitive information about a customer. All of this needs to be protected. And so that's where RAG as a concept or proving out a simple use case can be pretty straightforward, and I agree with that. But doing it across many use cases in an environment where you have sensitive information, you got to govern that, that is actually quite challenging. And so we've learned a lot about that, and now actually taken that knowledge and some of those technology partners that we used integrated them into our partnership program, and now we're actually helping customers go down at a similar journey.

>> No shadow RAG, That's dangerous stuff.

Sahir Azam

>> It's a tricky balance, to one hand, we're a very innovation focused, bottoms-up kind of culture at MongoDB. So the goal has always been not to say no, but to say, here are your tools as fast as possible that are safe, but go at it, once you're within these constraints, go at it on the innovation, the hackathons, and it's been great.

>> And I'm kind of joking about shadow, but you think about the Hadoop era, the big data, that we're just experimenting and then a year and a half, two years later, the organization would say, Well, wait a minute. We have to govern this And so you're getting out ahead of that saying, Here's a tool set that you If you're going to experiment, these are the boundaries that you think, with AI in particular, we're seeing that customers in the enterprise especially they understand this, they're very much intrigued by the potential, all are in terms of where some of the use cases that are going to be possible, whether it's driving efficiency and improving the bottom line, or whether it's creating new business models, new customer experiences. I think everyone's really very much a board-level kind of thing around that. But at the same time, established brands have a reputation to protect and customer data and personal information to make sure it doesn't leak into the wrong places or that they're recommending the right best practices because they're also being very careful and cautious, which is making this governance and ethics around how these things get applied. But definitely is an important topic.an impossible to answer question, but I'll ask it anyway. And it's related to the future of apps, how do you see the future of apps? If I asked that question of you in 2000, you would've seen Salesforce and you may have predicted some kind of SaaS model emerging, but you probably wouldn't have predicted the iPhone, maybe. But what kind of framework should we think about in terms of the characteristics of the app in the future, the intelligent data app, if we can call it that. How should we think about the future of applications?

Sahir Azam

>> Yeah, I think it is a hard question, but I can proxy some of the stuff that I've been reading or watching or hearing from some of the experts and just our ecosystem and all of that. One, I do think the long-term potential is absolutely something we believe And I think the hype is warranted in terms of where things will go over the long term. But as you know, we've all seen multiple technology cycles. There's going to be some hype and then there's some disillusionment when things don't deliver against that hype. But over the long term, things really do get there if not exceed our expectations. And I think AI is going through its own journey around that. I don't want to try to predict where we are on that curve, but I definitely think-

>> Early, early.

Sahir Azam

>> Early is definitely the thing. I saw one of our investors, does an AI conference and was amazing talk where they compared when the iPhone came out. And when the iPhone first came out, it was a really great presentation. They're like, The first wave of apps were pretty simplistic,the flashlight app, those utility apps that had some value, but they were quite small. But only a few years later you saw the DoorDashes, Ubers, the Instagrams, these applications that created completely new business models or experiences that we couldn't conceive of. And I think with AI, we're seeing glimpses of that, but I don't think anyone can really say where is things really going to go, who are the winners? But I know it's coming, and so I think that's exciting. I definitely think also that one of the things we're looking at that's promising is the whole idea of agentic workflows. Like how does, not just a model solve a particular narrow problem, but how can it help one orchestration agent do more complex business workflows over time? And you've see some interesting demos come out from some startups of I think where things can go, very early, but we're excited about the potential of that and are watching that space as well the other thing I would add to that excellent answer would be the impact on industry. I mean, you think about how mobile banking just completely changed the which we interact with our financial institutions. Stock trading, even though you had internet trading, now, it's just like, Boom, boom, Robinhood.

Sahir Azam

>> And the power and ability to reason about all this unstructured data truly in a way that wasn't really mainstream or possible, except for maybe the pockets of the largest, most sophisticated tech companies in the world, that's going to be really empowering and democratizing for any organization that's building software with all the technologies that are emerging and that's really exciting.

>> Sahir, amazing interview as always, thanks so you. Thank you for having

>> us.You're very welcome. And thank you for your attention here. We'll be back right after this short break. MongoDB live from New York City. You're watching theCUBE.