MongoDB .local NYC 2024 | Benny Chen | Fireworks AI & Gregory Maxson | MongoDB

Clips
News
More from MongoDB .local NYC 2024

Gregory Maxson

Global Lead of AI GTM

MongoDB

Benny Chen

Co-Founder

Fireworks AI

Fireworks AI and MongoDB aim to redefine generative AI platforms

Generative artificial intelligence as a platform ultimately succeeds when it’s a marriage between collaboration and innovation....

play_circle_outline Synergizing Fireworks AI and MongoDB: A Dynamic Partnership for Complementary Approaches and Revolutionary AI App Program

Info

Benny Chen | Fireworks AI & Gregory Maxson | MongoDB

Gregory Maxson

Global Lead of AI GTM MongoDB

Benny Chen

Co-Founder Fireworks AI

search

>> Hi, everybody. Welcome back to the Big Apple. We're here at MongoDB.local. Come on inside theCUBE, we are really excited to have Benny Chen here. He's a co-founder of Fireworks AI and Gregory, Greg Maxson, excuse me, MongoDB Global lead of AI, GTM, that's go-to-market. Greg, Benny, good to see you. Thanks so much for coming on.

Benny Chen

>> Thank you.

>> Okay, first question. I always love to ask co-founders and founders this question. Why'd you start the company?

Benny Chen

>> Well, I asked myself if not with this team, then with who? And I had no one else. I respected Lin and Dimitro so much that I was like, I'm going to go for it.

>> Okay, so you really didn't have an idea as to what problem you were going to solve at that point. You just knew you'd figure it out. Is that right?

Benny Chen

>> We had some idea. We were a bunch of PyTorch veterans, so we knew that people out there need a lot of help on PyTorch. Exactly how, we didn't really figure out completely. And honestly, when we founded the company, ChatGPT wasn't out yet, so we had to pivot somewhere in the middle as well.

>> So, tell us what you do, what the company's focused on.

Benny Chen

>> So, Fireworks is a inference provider that helps people serve their open source models and we also serve proprietary models, like we help stability and upstage. We also help with inference on fine-tuning as well as embedding. So, then businesses can really control the LMs productionize those cases.

>> And you are sort of Meta DNA. You came out of Meta, and Meta likes to build stuff. So, it's interesting to me that you partnering with Mongo because Mongo is all about making things simple, right? You guys like hard things typically. So, why Mongo? What's that partnership about?

Benny Chen

>> Number one, I do see that Mongo is actually one of the most sophisticated database providers. So, that part, actually, I think that has strongly resonated with me because it's very easy to use. And of the day, we want developers to be able to leverage these tools. We want to work with providers who enable and empower developers. And that's the secret sauce for PyTorch. Back then, it wasn't the most powerful, most efficient model framework, but it was the most easy to use. And then, the developers have gravitated towards the things that's easy to use, and then, we can figure out the performance and all that stuff later. I don't know if Greg has-

>> And add value to you have anything to add there, Greg?

Gregory Maxson

>> Yeah, we really valued the partnership with Fireworks. We feel that together they create a complete offering for our developers. And that's where we're just obsessed thinking about is what's the full end-to-end stack that they need to be successful? And so, the services that Fireworks provides around inference, around fine-tuning combined with your proprietary data being easily accessible with is why we think this is such a great perfect fit partnership for our customers.

>> And the big news here, MongoDB announced its AI app program, MAP. I think it actually announced it earlier this week. How's that fit in to the partnership?

Gregory Maxson

>> Sure. I think it's a perfect extension of the work we were already doing with Fireworks. The reason we wanted to partner with Fireworks was because they were highly complementary. We thought about what are all the needs our customers have? We solve some of them, but we don't solve the other parts. And Fireworks comes in as that perfectly complementary component. However, there's a lot of other parts to the stack that a developer would need in order to build a modern AI application. And so, we wanted to provide that complete stack to our customers. What we were hearing from our customers was, This market's moving so fast, it's changing so quickly that it's hard for me to keep pace. I find it overwhelming to know how to get started. And I oftentimes don't have the in-house expertise to integrate and implement all of these things. So, I need a trusted advisor to help show me what's the full stack I need. What does great look like? What's worked?And that's what MAP is intended to provide is that comprehensive set of offerings. And we're really excited that Fireworks was an integral launch partner as part of that,

>> In a recent survey with our partner ETR, it was a new York-based data platform and they do a lot of surveys, surveys about 800 predominantly CXOs and IT decision makers, 20said they're not pursuing any gen AI. And when I saw that number, Why? You got to be kidding me. Everybody's doing gen AI.And when we dug into the data and we talked to some of these folks, they said exactly what you just said, You know what? It's moving too fast. It's actually too risky for us. So, we're going to sit back...I like it. It's a Forest Gump, Bubba Gump Shrimp when the hurricane came in and wiped out everybody and then they were left. We're going to wait and see where all the damage is. And we're going to come in.But that's perhaps not the best strategy because you could be left behind. So, if I understand it correctly, you would obviously appeal to those folks, and also, help accelerate those that aren't quite seeing the ROI out of the chatty use cases. Is that a fair assessment?

Gregory Maxson

>> I think that's right. We've heard that same problem. We've heard it's really hard to get started. That first mile feels overwhelming, but a lot of companies also recognize that it's a very much disrupt or be disrupted period right now. And so, there's a lot of risk with waiting and being on the sideline and saying, Okay, in a year, we'll come back and see if the water's warm.And so, we want to alleviate some of that risk and make it easy to get started, show proven repeatable reference architectures, give the best parts of the stack, help provide the expertise in professional services to make sure all of these parts of the stack work well together and implement it. And hopefully, that helps bring down your 20number to a much lower number, so that everyone knows they can get started and it's not as daunting as they might think it is.

>> But the other part of that equation, which is perhaps more interesting, is aligning with the 80Because 80doing stuff, but when you ask them what they're doing, it's tech summary, it's code generation, it's chat, it's basic stuff that's nice, and it's actually quick ROI, but it's not a big net present value. It's not a big chunky revenue driver. So, I'm trying to understand where you guys fit, how you can help not only that 20that aren't doing anything, but that 80who want to take the next step and do more value-oriented work.

Benny Chen

>> Yeah, I think for Fireworks, we serve two really big sets of use cases that helps to add value. The one is what I would categorize as co-pilots where the model need to think as fast as you type or think as fast as you speak. So, in those cases, we are very low latency, we're very good at serving things at production that is stable, highly available. So, in those cases, we will do well with very small models that already exist today. I think what's yet to happen is you need the assistant that shows up every day, that helps you do your day-to-day work, and the models are just not there yet. And I think it's up to us and the people in the industry to help push forward to get to a place where the model is smart enough to actually answer your emails. And at that point, we will have model that runs more slowly, we'll have more agents that's running the system on your behalf. But then, that I think has to happen later on in the next two years.

>> So, Benny, this is a model problem, not so much a data problem, or is it a combination?

Benny Chen

>> I think a combination. And also I think on top of that, there has to be frameworks like agents, and people set up those agents correctly to be able to do your day-to-day job. For us, I don't know, whenever I come into a new job, I get yelled at for two weeks at least before I can actually do the job. And the yelling that has to happen to the model, it's the data training, it's the guardrails, it's the agents that's put in place to get the model to behave the way you want it to behave before you can actually do the job. So, the developers are actually the most interesting part of the puzzle here. They need to be empowered to build these things, and be empowered to put these models the way they actually help you do the work, and we are here to help the developers.

>> Well, and of course, again, this scares people. This is where governance comes into play because Automobile Manufacturer A releases a RAG or a chatbot recommends the competitor's truck. You've seen things like. we've seen people being able to prompt engineer and get a better price, actually, get something for a dollar that should have cost thousands of dollars, so we've seen those types of corner cases. So, governance is a huge, huge concern. And we always talk about guardrails and things like that. I'm interested in both of how are you guys helping solve that problem? Presumably, that's part of what Mongo is delivering with this curated set of LLMs and tools. And then, I'm curious if you guys can help the problem, why don't you start?

Gregory Maxson

>> Yeah, definitely. We recognize the challenge with that. I think there's so much uncertainty still and skepticism around the ability for AI to function exactly as intended. And that's a big portion why we think that data is such an important component of it because the models that you're building, the applications that you're building that are enriched with AI need to be grounded in your proprietary data. They need to be talking the way your brand voice talks. They need to follow your corporate guidelines. And right now, if you're just using generic models and not grounding it in your data, then you're going to get generic responses that don't align with the way that your organization would want to operate and would traditionally train thousands of people to operate on, but now can be automated. So, that's why we believe that it hasn't yet come to fruition as much as we think it will. But your proprietary data is such an important component in ensuring that AI reflects, like your brand wants to be reflected, in these modern applications.

>> Benny, do you guys help solve that governance issue? And if so, how?

Benny Chen

>> I think one key point we help with is speed. And with fast inference, you can also enforce these rules in the LLM form and ask the LLM whether... If I was a sales agent from MongoDB, am I adhering to the corporate policies? Am I being warm? Am I being welcoming? And is actually really good at assessing content. So, if they're fast for inference, you can sneak in those guardrail components by asking the LLM whether you're doing these things right. And with faster speed, you can sneak in more and more of these guardrails. And eventually, with enough guardrails in place, you could be able to have these agents representing your company, but it will take a while.

>> So, thinking about Dave's three layers. He talks about the infrastructure layer, the training and tools layer, and then the apps layer. I'm curious as to how you guys see the future of applications. Obviously, AI plays a role, a shift in activity from training to inference. You've got GPUs, ASICs, LPUs now, all kinds of specialized processors that are super powerful that application developers can take advantage of. If I asked you in 1999, What's the future of apps look like?And you said, Oh, look at Salesforce, they're doing SaaS or cloud or cloud-based.They're really not cloud at the time. We weren't using that term, but software as a service, you'd point to that. But Salesforce wasn't the future of apps. They predated cloud and mobile. So, you look at Salesforce today, it's kind of not the greatest mobile experience, but it's survived. But the future of apps was what you see in social media, what you see Uber. So, with that is this long-winded background. How do you guys envision the future of apps? Maybe you could start, Benny.

Benny Chen

>> Yeah, I think the first thing is the set of co-pilots are going to proliferate. Right now, a lot of co-pilots are in your coding environments. They're probably also in your photo editors, things like that. A lot of tools will get up-leveled for more co-pilot use cases. So, then it's not so difficult to use these tools. If I have to change the hue and saturation for this photo, I have to go into this drop down and remember that, oh, these are the set of things I need to change for this set of photos. Those are just way too difficult for a human to remember, and co-pilots will help with those. I think the second set of things will be, for lack of better words, like agents. So, for me, honestly, it's like if one day I can email something and it comes back with satisfactory results, then that has passed the agent test for me. It doesn't happen yet. Eventually, those, we may or may not call them applications, but things I would imagine I can email with.

>> But their business impact activities, right? I bet you in the flywheel between what Mongo's doing, this whole partnership strategy and the developer community that you have, how that will lead us into the future and your vision, Greg?

Gregory Maxson

>> Yeah, when we think about the future of apps, I mean I think they're going to be highly iterative. There isn't a future app, it's going to be incremental in nature. But we expect to see significantly better personalization, so that apps aren't going to be generic one-size-fits-all experiences for everyone. I mean, think about a retail experience today. If you're on the Nordstrom app or something like that, you go through a click down and you select men's clothing and you select jackets, and you get this nice blazer that you have on. And you have to search through 50 pages of blazers to find something you like. However, if you have, in the future, some autonomous agent on there and you say, Hey, I'm going to Paris in April for a wedding. And the dress code is this. And then, here's my background of things I purchased, here's what I look like, here's what I like,some recommendation can be highly personalized to exactly who you are, how much you want to spend, where you're located, when you need it, what the weather's going to be in Paris at that time. All of these other details that previously you would've just had to go through your own discovery process. There's so many different ways in which that can show up. And I think we talk about apps and you think about an app on your phone, but then there's also apps for a business. Think about a major consulting firm. How do I build some sort of application so that my consultants don't have to spend hundreds of hours thinking through, What's the right customer experience workflow for a telco in South America based on all the diligence we've done historically? Comb through all of that information and build me a deck.That's an app for a different type of user. And so, no matter who your customer is, whether it's an internal employee looking for information about the company, whether it's a consumer, whether it's a B2B, there be all of these applications that will make people smarter, give them better information based on what their specific needs are.

>> That's a good vision because today we see glimpses of it. You can go into ChatGPT, you can ask it to whatever, You build me a business plan,or, A project plan.And it'll give you something that's generic. It's not that great, but it's helpful. But you're envisioning a world where the quality of the output is much better. And on the former example, like consumer examples, my takeaway there is you're talking about personalization at scale, highly-customized experience, but at scale. So, my question is coming back, I presume it's high-quality, speed, that's the role that you guys play, correct?

Benny Chen

>> Yes, yes, yes. And talking about personalization. So, I used to work at Meta as recommendations. A lot of our colleagues also came from recommendation type of workload before. And the recommendation we use to serve are given a set of inventory, how can I find things that is interesting for you? I think what will be changing is the inventory itself may also be the AI generated. And there could be cases where businesses can identify things that people are interested in but doesn't yet exist, and then use AIs to fill those inventories. And I think that's probably going to be where a lot of changes are, especially for social networks and for video sites in the next few years.

>> So, in the MAP program, there's some curated set of LLMs. We saw Anthropic on stage, Llama I'm pretty sure is in there. I think Cohere. I saw Cohere, they're in there. Okay, so that's cool. Are you comfortable with Mongo's choices? If Dave were right here, would you want him to add another one? What's your take on all this?

Benny Chen

>> I think it was like Warren Buffett's quote or something like, If Yao Ming walks into the bar, you don't ask if he's tall or not. You just recognize he's tall.If someone's tall, you just recognize he's tall. And we strive to be the best. And one day, if we are like Yao Ming, we're like, oh, we walk into the bar, and, This is one of the LLM providers,then we achieved our goal. And I think Mongo should definitely pick and choose whatever that's best for the customers. We're here to get the developers to the stage where they can productionize the applications. And honestly, as long as it's on that path and how Fireworks is helping people on that path, Fireworks should just be part of the choice. Mongo should pick and choose. Yeah.

>> You don't feel limited by Mongo's choices. This is clear. All right, Greg, I'll give you the last word on MongoDB.local 2024. Takeaways from the show? Anything you want to say about the partnerships?

Gregory Maxson

>> Yeah, we're really excited about all the partnerships that we're bringing to life here. I think it shows amazing momentum. As part of the MAP program, we had 15 great launch partners. But we're really just beginning in this journey and it's all about helping our customers be successful. We recognize that we're super early in this stage of AI development and we're really excited to bring some of these experiences to life with our partners, like Fireworks, over the next year.

>> We're excited too to be covering this. Thanks guys. Benny, Greg, thanks for coming to theCUBE.

Gregory Maxson

>> Thank you.

>> Okay, keep it right there. We'll be back. Suraj Patel is coming up. He's the head of MongoDB Ventures. And we're going to be featuring a company called Gnomic. Keep it right there. You're watching theCUBE at MongoDB.local from New York City. Right back.

Benny Chen | Fireworks AI & Gregory Maxson | MongoDB

search

>> Hi, everybody. Welcome back to the Big Apple. We're here at MongoDB.local. Come on inside theCUBE, we are really excited to have Benny Chen here. He's a co-founder of Fireworks AI and Gregory, Greg Maxson, excuse me, MongoDB Global lead of AI, GTM, that's go-to-market. Greg, Benny, good to see you. Thanks so much for coming on.

Benny Chen

>> Thank you.

>> Okay, first question. I always love to ask co-founders and founders this question. Why'd you start the company?

Benny Chen

>> Well, I asked myself if not with this team, then with who? And I had no one else. I respected Lin and Dimitro so much that I was like, I'm going to go for it.

>> Okay, so you really didn't have an idea as to what problem you were going to solve at that point. You just knew you'd figure it out. Is that right?

Benny Chen

>> We had some idea. We were a bunch of PyTorch veterans, so we knew that people out there need a lot of help on PyTorch. Exactly how, we didn't really figure out completely. And honestly, when we founded the company, ChatGPT wasn't out yet, so we had to pivot somewhere in the middle as well.

>> So, tell us what you do, what the company's focused on.

Benny Chen

>> So, Fireworks is a inference provider that helps people serve their open source models and we also serve proprietary models, like we help stability and upstage. We also help with inference on fine-tuning as well as embedding. So, then businesses can really control the LMs productionize those cases.

>> And you are sort of Meta DNA. You came out of Meta, and Meta likes to build stuff. So, it's interesting to me that you partnering with Mongo because Mongo is all about making things simple, right? You guys like hard things typically. So, why Mongo? What's that partnership about?

Benny Chen

>> Number one, I do see that Mongo is actually one of the most sophisticated database providers. So, that part, actually, I think that has strongly resonated with me because it's very easy to use. And of the day, we want developers to be able to leverage these tools. We want to work with providers who enable and empower developers. And that's the secret sauce for PyTorch. Back then, it wasn't the most powerful, most efficient model framework, but it was the most easy to use. And then, the developers have gravitated towards the things that's easy to use, and then, we can figure out the performance and all that stuff later. I don't know if Greg has-

>> And add value to you have anything to add there, Greg?

Gregory Maxson

>> Yeah, we really valued the partnership with Fireworks. We feel that together they create a complete offering for our developers. And that's where we're just obsessed thinking about is what's the full end-to-end stack that they need to be successful? And so, the services that Fireworks provides around inference, around fine-tuning combined with your proprietary data being easily accessible with is why we think this is such a great perfect fit partnership for our customers.

>> And the big news here, MongoDB announced its AI app program, MAP. I think it actually announced it earlier this week. How's that fit in to the partnership?

Gregory Maxson

>> Sure. I think it's a perfect extension of the work we were already doing with Fireworks. The reason we wanted to partner with Fireworks was because they were highly complementary. We thought about what are all the needs our customers have? We solve some of them, but we don't solve the other parts. And Fireworks comes in as that perfectly complementary component. However, there's a lot of other parts to the stack that a developer would need in order to build a modern AI application. And so, we wanted to provide that complete stack to our customers. What we were hearing from our customers was, This market's moving so fast, it's changing so quickly that it's hard for me to keep pace. I find it overwhelming to know how to get started. And I oftentimes don't have the in-house expertise to integrate and implement all of these things. So, I need a trusted advisor to help show me what's the full stack I need. What does great look like? What's worked?And that's what MAP is intended to provide is that comprehensive set of offerings. And we're really excited that Fireworks was an integral launch partner as part of that,

>> In a recent survey with our partner ETR, it was a new York-based data platform and they do a lot of surveys, surveys about 800 predominantly CXOs and IT decision makers, 20said they're not pursuing any gen AI. And when I saw that number, Why? You got to be kidding me. Everybody's doing gen AI.And when we dug into the data and we talked to some of these folks, they said exactly what you just said, You know what? It's moving too fast. It's actually too risky for us. So, we're going to sit back...I like it. It's a Forest Gump, Bubba Gump Shrimp when the hurricane came in and wiped out everybody and then they were left. We're going to wait and see where all the damage is. And we're going to come in.But that's perhaps not the best strategy because you could be left behind. So, if I understand it correctly, you would obviously appeal to those folks, and also, help accelerate those that aren't quite seeing the ROI out of the chatty use cases. Is that a fair assessment?

Gregory Maxson

>> I think that's right. We've heard that same problem. We've heard it's really hard to get started. That first mile feels overwhelming, but a lot of companies also recognize that it's a very much disrupt or be disrupted period right now. And so, there's a lot of risk with waiting and being on the sideline and saying, Okay, in a year, we'll come back and see if the water's warm.And so, we want to alleviate some of that risk and make it easy to get started, show proven repeatable reference architectures, give the best parts of the stack, help provide the expertise in professional services to make sure all of these parts of the stack work well together and implement it. And hopefully, that helps bring down your 20number to a much lower number, so that everyone knows they can get started and it's not as daunting as they might think it is.

>> But the other part of that equation, which is perhaps more interesting, is aligning with the 80Because 80doing stuff, but when you ask them what they're doing, it's tech summary, it's code generation, it's chat, it's basic stuff that's nice, and it's actually quick ROI, but it's not a big net present value. It's not a big chunky revenue driver. So, I'm trying to understand where you guys fit, how you can help not only that 20that aren't doing anything, but that 80who want to take the next step and do more value-oriented work.

Benny Chen

>> Yeah, I think for Fireworks, we serve two really big sets of use cases that helps to add value. The one is what I would categorize as co-pilots where the model need to think as fast as you type or think as fast as you speak. So, in those cases, we are very low latency, we're very good at serving things at production that is stable, highly available. So, in those cases, we will do well with very small models that already exist today. I think what's yet to happen is you need the assistant that shows up every day, that helps you do your day-to-day work, and the models are just not there yet. And I think it's up to us and the people in the industry to help push forward to get to a place where the model is smart enough to actually answer your emails. And at that point, we will have model that runs more slowly, we'll have more agents that's running the system on your behalf. But then, that I think has to happen later on in the next two years.

>> So, Benny, this is a model problem, not so much a data problem, or is it a combination?

Benny Chen

>> I think a combination. And also I think on top of that, there has to be frameworks like agents, and people set up those agents correctly to be able to do your day-to-day job. For us, I don't know, whenever I come into a new job, I get yelled at for two weeks at least before I can actually do the job. And the yelling that has to happen to the model, it's the data training, it's the guardrails, it's the agents that's put in place to get the model to behave the way you want it to behave before you can actually do the job. So, the developers are actually the most interesting part of the puzzle here. They need to be empowered to build these things, and be empowered to put these models the way they actually help you do the work, and we are here to help the developers.

>> Well, and of course, again, this scares people. This is where governance comes into play because Automobile Manufacturer A releases a RAG or a chatbot recommends the competitor's truck. You've seen things like. we've seen people being able to prompt engineer and get a better price, actually, get something for a dollar that should have cost thousands of dollars, so we've seen those types of corner cases. So, governance is a huge, huge concern. And we always talk about guardrails and things like that. I'm interested in both of how are you guys helping solve that problem? Presumably, that's part of what Mongo is delivering with this curated set of LLMs and tools. And then, I'm curious if you guys can help the problem, why don't you start?

Gregory Maxson

>> Yeah, definitely. We recognize the challenge with that. I think there's so much uncertainty still and skepticism around the ability for AI to function exactly as intended. And that's a big portion why we think that data is such an important component of it because the models that you're building, the applications that you're building that are enriched with AI need to be grounded in your proprietary data. They need to be talking the way your brand voice talks. They need to follow your corporate guidelines. And right now, if you're just using generic models and not grounding it in your data, then you're going to get generic responses that don't align with the way that your organization would want to operate and would traditionally train thousands of people to operate on, but now can be automated. So, that's why we believe that it hasn't yet come to fruition as much as we think it will. But your proprietary data is such an important component in ensuring that AI reflects, like your brand wants to be reflected, in these modern applications.

>> Benny, do you guys help solve that governance issue? And if so, how?

Benny Chen

>> I think one key point we help with is speed. And with fast inference, you can also enforce these rules in the LLM form and ask the LLM whether... If I was a sales agent from MongoDB, am I adhering to the corporate policies? Am I being warm? Am I being welcoming? And is actually really good at assessing content. So, if they're fast for inference, you can sneak in those guardrail components by asking the LLM whether you're doing these things right. And with faster speed, you can sneak in more and more of these guardrails. And eventually, with enough guardrails in place, you could be able to have these agents representing your company, but it will take a while.

>> So, thinking about Dave's three layers. He talks about the infrastructure layer, the training and tools layer, and then the apps layer. I'm curious as to how you guys see the future of applications. Obviously, AI plays a role, a shift in activity from training to inference. You've got GPUs, ASICs, LPUs now, all kinds of specialized processors that are super powerful that application developers can take advantage of. If I asked you in 1999, What's the future of apps look like?And you said, Oh, look at Salesforce, they're doing SaaS or cloud or cloud-based.They're really not cloud at the time. We weren't using that term, but software as a service, you'd point to that. But Salesforce wasn't the future of apps. They predated cloud and mobile. So, you look at Salesforce today, it's kind of not the greatest mobile experience, but it's survived. But the future of apps was what you see in social media, what you see Uber. So, with that is this long-winded background. How do you guys envision the future of apps? Maybe you could start, Benny.

Benny Chen

>> Yeah, I think the first thing is the set of co-pilots are going to proliferate. Right now, a lot of co-pilots are in your coding environments. They're probably also in your photo editors, things like that. A lot of tools will get up-leveled for more co-pilot use cases. So, then it's not so difficult to use these tools. If I have to change the hue and saturation for this photo, I have to go into this drop down and remember that, oh, these are the set of things I need to change for this set of photos. Those are just way too difficult for a human to remember, and co-pilots will help with those. I think the second set of things will be, for lack of better words, like agents. So, for me, honestly, it's like if one day I can email something and it comes back with satisfactory results, then that has passed the agent test for me. It doesn't happen yet. Eventually, those, we may or may not call them applications, but things I would imagine I can email with.

>> But their business impact activities, right? I bet you in the flywheel between what Mongo's doing, this whole partnership strategy and the developer community that you have, how that will lead us into the future and your vision, Greg?

Gregory Maxson

>> Yeah, when we think about the future of apps, I mean I think they're going to be highly iterative. There isn't a future app, it's going to be incremental in nature. But we expect to see significantly better personalization, so that apps aren't going to be generic one-size-fits-all experiences for everyone. I mean, think about a retail experience today. If you're on the Nordstrom app or something like that, you go through a click down and you select men's clothing and you select jackets, and you get this nice blazer that you have on. And you have to search through 50 pages of blazers to find something you like. However, if you have, in the future, some autonomous agent on there and you say, Hey, I'm going to Paris in April for a wedding. And the dress code is this. And then, here's my background of things I purchased, here's what I look like, here's what I like,some recommendation can be highly personalized to exactly who you are, how much you want to spend, where you're located, when you need it, what the weather's going to be in Paris at that time. All of these other details that previously you would've just had to go through your own discovery process. There's so many different ways in which that can show up. And I think we talk about apps and you think about an app on your phone, but then there's also apps for a business. Think about a major consulting firm. How do I build some sort of application so that my consultants don't have to spend hundreds of hours thinking through, What's the right customer experience workflow for a telco in South America based on all the diligence we've done historically? Comb through all of that information and build me a deck.That's an app for a different type of user. And so, no matter who your customer is, whether it's an internal employee looking for information about the company, whether it's a consumer, whether it's a B2B, there be all of these applications that will make people smarter, give them better information based on what their specific needs are.

>> That's a good vision because today we see glimpses of it. You can go into ChatGPT, you can ask it to whatever, You build me a business plan,or, A project plan.And it'll give you something that's generic. It's not that great, but it's helpful. But you're envisioning a world where the quality of the output is much better. And on the former example, like consumer examples, my takeaway there is you're talking about personalization at scale, highly-customized experience, but at scale. So, my question is coming back, I presume it's high-quality, speed, that's the role that you guys play, correct?

Benny Chen

>> Yes, yes, yes. And talking about personalization. So, I used to work at Meta as recommendations. A lot of our colleagues also came from recommendation type of workload before. And the recommendation we use to serve are given a set of inventory, how can I find things that is interesting for you? I think what will be changing is the inventory itself may also be the AI generated. And there could be cases where businesses can identify things that people are interested in but doesn't yet exist, and then use AIs to fill those inventories. And I think that's probably going to be where a lot of changes are, especially for social networks and for video sites in the next few years.

>> So, in the MAP program, there's some curated set of LLMs. We saw Anthropic on stage, Llama I'm pretty sure is in there. I think Cohere. I saw Cohere, they're in there. Okay, so that's cool. Are you comfortable with Mongo's choices? If Dave were right here, would you want him to add another one? What's your take on all this?

Benny Chen

>> I think it was like Warren Buffett's quote or something like, If Yao Ming walks into the bar, you don't ask if he's tall or not. You just recognize he's tall.If someone's tall, you just recognize he's tall. And we strive to be the best. And one day, if we are like Yao Ming, we're like, oh, we walk into the bar, and, This is one of the LLM providers,then we achieved our goal. And I think Mongo should definitely pick and choose whatever that's best for the customers. We're here to get the developers to the stage where they can productionize the applications. And honestly, as long as it's on that path and how Fireworks is helping people on that path, Fireworks should just be part of the choice. Mongo should pick and choose. Yeah.

>> You don't feel limited by Mongo's choices. This is clear. All right, Greg, I'll give you the last word on MongoDB.local 2024. Takeaways from the show? Anything you want to say about the partnerships?

Gregory Maxson

>> Yeah, we're really excited about all the partnerships that we're bringing to life here. I think it shows amazing momentum. As part of the MAP program, we had 15 great launch partners. But we're really just beginning in this journey and it's all about helping our customers be successful. We recognize that we're super early in this stage of AI development and we're really excited to bring some of these experiences to life with our partners, like Fireworks, over the next year.

>> We're excited too to be covering this. Thanks guys. Benny, Greg, thanks for coming to theCUBE.

Gregory Maxson

>> Thank you.

>> Okay, keep it right there. We'll be back. Suraj Patel is coming up. He's the head of MongoDB Ventures. And we're going to be featuring a company called Gnomic. Keep it right there. You're watching theCUBE at MongoDB.local from New York City. Right back.