AWS Summit NYC 2024 | Keynote Analysis Pt 2 with Sarbjeet Johal | Stackpane

theCUBE.net

info Help
Sign in
Register

Clips
News
More from AWS Summit NYC 2024

From code to cloud, how is AWS shaping the gen AI landscape? TheCUBE keynote analysis from AWS Summit NY

The generative artificial intelligence wave has entered the infrastructure stage. Hyperscalers and solutions companies are laying out the gen AI blueprint for simplified, secure and scalable deployment.One such name is Amazon Web Services Inc., whose messaging signals an accelerating focus on enterprise AI, marked by rapid innovation, an expanding partner ecosystem and support for broadened app layers.On set with theCUBE at AWS Summit New York 2024.“I think [AWS’ Matt Wood] used the platform to really establish the fact that AWS is not just a participant in the gen AI wars, they are the leader in it,” said Zeus Kerravala (pictured, left), founder of ZK Research and CUBE Collective analyst

play_circle_outline Matt Wood's keynote highlights Generative AI and Amazon as a builders' cloud.

play_circle_outline Amazon's new announcements include Anthropic Claude 3 and AWS App Studio.

play_circle_outline Infrastructure layer almost settled, focus shifting to application layer.

play_circle_outline Discussion on application stack evolution and data cleansing for gen AI applications.

Info

Keynote Analysis Pt 2 with Sarbjeet Johal | Stackpane

search

>> Welcome back, everyone. This is theCUBE's coverage in New York City. I'm John Furrier, the host of theCUBE. I'm here with Sarbjeet Johal, who is theCUBE contributor, also Analyst with StackPane. Check him out on Twitter. He's actively watching all the action as well as reporting. And Sarbjeet is always kind enough to be doing keynote reviews with us. And he's got his notebook from the field, which he also does the segment called Notebook from the Field. Sarbjeet, get your notes. Good to see.

>> Got my notebook, yes.

>> We sat next each other in the keynote. Got a nice selfie.

>> Yes, -

>> What's your analysis? I mean, obviously Matt Wood was the star of the show. I mean, I had customers up there, but then they do their thing. "Hey, we're using Cube, blah, blah, blah." But Matt Wood delivered the payload. Generative AI. We are the AI company. What is your analysis?

>> I think they are the builders' cloud. They are the builders' company, if you will, overall. So when you talk about builder, they're ISVs on one side and then in-house developers at the enterprises. Hands down, they are favorite for ISVs. But when it comes to in-house development at enterprises, they are these days with some other competition. They have competition in that domain and a lot more competition than ... I mean, they have competition in ISV space as well, but they have the sort of stronghold on that and they're trying to keep that stronghold as strong. They were. And they had a couple of new announcements which are fresh preview that everybody ... It's industry trend now, right?

>> Yeah, -

>> Preview and then GA.

>> General availability, yeah.

>> But most of the announcements now -

>> Take us through the announcements. What do you got?

>> Yeah, Amazon. They're doing the ... I think the biggest one was Anthropic Claude 3 announcement that the first time you can do fine-tuning of that model. This is the only cloud they say where you can do that. So that was, I think, huge. And there were a lot of enhancements on the queue and they're expanding the scope of the queue from business to low code to pro code. So, spectrum of use cases there. I think the use cases in that segment and queue in the middle are kind of very overlapping. There are a couple of two, three low code sort of productization they have done of the queue, which is, to be honest, it's kind a little noisy. And contextual grounding checks of the models, that was a good one. AWS app studio, that's their low code. Another low code sort of offering, which I'm not a big fan of because I come from the developer background. I'm a pro developer, pro. Not -

>> They don't have a queue version for you yet. Okay. When you say builders' cloud, which love that term because I just love when I first heard that from Andy Jassy. It just makes so ... People are building applications, they're building companies and they want to do it right. When I'm in New York City, I see around, we're on the Javits Center, we're on the Hudson side. It's been booming here for decades. This building's going up and you look up and you see a big building being built. So when I hear building, I think steel, hanging iron, building the floors up and then finishing it up with walls and you put in the elevator, you have all these components. Someone's managing that job, some general contractor, some GC. So now I see we're in an era of cloud, post-cloud 1.0, you got all the hyper scale clouds pumping at all cylinders and you've got a cloud with all this resource and capabilities. And so, it's really a question of not just for builders, what do you build? So I like to see Amazon bring that, here's app studio, here's three steps to build an app. I'm not yet sold on, give me a prompt and I'll build an app for you. That sounds very cliche. oh, come on. Really? Okay, that's good hype and shows like the ChatGPT vibe, that's magic. But I think the real serious developers, I want to build an app. I want to build an app. I want to run it securely. And if I'm a consumer, I'm an entrepreneur, okay, I don't really care about data because I have my own data. Or if I'm a company, I care about my data too. So security matters. So it's a hard nut to crack because you want to be horizontal in that use case. Do you think he met the mark today on the keynote?

>> I think he met the mark on the keynote, but a lot of repetition from the other AI or AWS summits. But there's only very few fractional new things here. But having said that, I think Amazon is one of those companies, of course, but all vendors have this sort of blind spot. I've seen that. I've been in the industry for 30 years now. So they swing the pendulum to the new shiny object a little too much. I think they're doing that. This AI summit is all about AI. This is an AWS summit, not -

>> Who's , Amazon is?

>> Amazon is.

>> people.

>> Amazon is doing that as well.

>> You think they're groping for relevance or do you think they have it?

>> They have-

>> -...

>> substance in the cloud. But from the AI side, they are a little behind. Everybody knows that. And then they're trying to grope -

>> They got the , so they got the silicon, they got the horsepower, they got the platform. And to me, I think it's bedrock has to come together. To me, bedrock is the key.

>> Again for the builders. Builders like those three layers. And then they will embrace that. No doubt about that. I have almost zero doubt about that. But when it comes to the in-house developers at enterprises, I think they will be segregated. Tell me the vendor which doesn't have AI story, gen AI story. Every vendor -

>> Everyone's bolting gen AI in their story and-

>> Exactly....

>> hoping they get a bigger valuation. But right now, we're a year and a half in. I'd say a year past the hype in a full year of production. You're starting to see production workloads hit. Again, low hanging fruit of some nice winds out there. This year is going to be an integration year. And so, we had the partner folks on earlier this morning, Mona, and the person who's running the company center, I know you took a tour at Deloitte, which is another tell sign. Still early innings, but Amazon has an ecosystem. Their entire company was built with people using the platform. And so, to me, I want to see what comes out of the ecosystem. So the question is, what are partners doing? What do you see there? Do you see a robust set of building going on from the partners? I do. I mean, I see a lot of activity.

>> I see.

>> And they're not hand waving either. There's some real production. We're going to have some on this afternoon, so we're going to hear from them. But what's your take? What's the key for them to building gen ed? Do you see what Matt Wood presented, would resonate with partners?

>> Okay. You asked the question with the application stack, and then let's wrap that into-

>> There's five questions in there. Pick your choice.

>> I will answer in a way that it will cover both. So this is ... I think the point which we have to focus on, application stack is going through a big change right now with the introduction of gen AI. So, the way we write applications will change. So, we'll be using many models. It's a multi-model, multimodal world. So how we will weave those models into our applications, that's still to be seen. Right now, what we are cooking up is very rudimentary stuff. Most of the use cases Deloitte showed us yesterday, five of them, they're still ... They're good, solid, but still rudimentary. Under each, they showed us the C-suite AI for CFO, and they had the creative edge for CMOs. It's like marketer in a box. So, you want to do a campaign, just -

>> Was that Deloitte's lab or ... That was the new lab that they're doing?

>> Yeah, in New York they are.

>> In New York City.

>> They took all of us, their analysts there and then did like two hour nonstop session. And then-

>> Yeah, I saw the play from Deloitte on the competency center and also with Amazon. Now they've got the lab there.

>> And we recorded two videos, which you will see that.

>> I don't think that was illegal. I think there was no videos allowed, but send them to us. We'll publish them.

>> Somehow we can do that.

>> Like the Pentagon papers for the videos. Send them to us, we'll publish them.

>> So, application stack is going through a big change. And the -

>> Yeah, the data stack's going under big change too.

>> Of course.

>> I mean, all gen AI apps are fed data, so I get that. But what's the action point right now? You think it's the application layer or is it the infrastructure layer?

>> I think infrastructure layer is almost behind us. We are -

>> You think we're clear on the runway on that?

>> Yes. Yes, we are. And we are going into the application layer. And TCS in the morning yesterday was on a panel, one of their big partners. They were talking about how the data needs to be cleansed in a way or fed in a way which was not previously done. Even the relational database basis, if you just take the rows and columns and tables and feed it to AI, it's not going to work. So you have to label the data properly. And Symantec search or we talk about -

>> Well, that's good news. If you see that the infrastructure, we're clearing the runway on that, that's a game changer because we've been waiting for that. But that's going to cause some consequences because think about all those companies out there, the whole GPU clouds that are out there, or the other things, they might or may not make it. I mean, there is a trajectory here that says, if the app layer takes all the action, all those IAS providers that had those specialty systems may or may not pan out. And heavily funded areas too.

>> Definitely. Yeah.

>> CoreWeave's one of them.

>> of the world and Supermicros. Look where they're flying, their stock is-

>> Well, they're going to AI. They're doing AI effects on premises.

>> Yes.

>> The question is-

>> Do you need that on premise?

>> Well, that's the thing. What's over-provisioning? What does the word, over-provisioning, mean in today's world? Remember the old concept, too much bandwidth, we don't really use it.

>> Low -

>> I got all the servers under capacity. What happened? A glut?

>> Yeah. So I think it was talking earlier about average server utilization. He said 30%. I think it's even lower in private data centers. And the storage utilization was even lower because every business unit will get their and they will not let -

>> So, the stuff's sitting around.

>> Yeah. And same thing will happen here. If people procure-

>> You mean with GPUs?

>> In GPU world, it will happen. If people get all those GPUs, and especially if they get by the business unit.

>> They buy all the GPUs. It's the classic shark fin or the iceberg under the water. Oh my God, it's going to cost us all this to put it together. We didn't see those costs coming. So again, cost exposure works both ways. Not just in the cloud, but also on premise.

>> On premise as well. I see this ... I sound like a broken record when I say that public cloud is the industrialization of backend computing. There's no denial in that factor.

>> Yeah, and they did a good job doing it.

>> And they did a good job.

>> Fantastic job.

>> And then other people copied it, few ... We have top three -

>> And it's sub-optimized.

>> Yeah. sub-optimized.

>> That's called diseconomies of scale.

>> And we are at the AWS summit, let's talk about them a little bit more. They innovate on all strands. There are so many strands of cloud. At high level is network. There's storage and there's compute. Under network, there's so many strands. And policy-based networking is -

>> Is it's too complicated. I mean, the benefit of the cloud is horizontal scalability. Have we gotten too product focused? Because again, gen AI, we're seeing the same movie. Data has to be horizontally scalable, which is not a normal thing. No one ever ... Go back a decade ago. Are you kidding me? No one scales data horizontally. You have your little silos and you got your little data warehouses, in comes Snowflake, in comes Databricks. And Dupe failed. Okay, now it's like, okay, I got to make my data horizontally scalable. I think Amazon has that opportunity to make that layer, pre-app layer absolutely fantastically scalable. And I think, I haven't seen that yet. I haven't seen that emerge, but that's where the puck has to go. I mean, don't you think?

>> I totally agree with you. I think data being in a high horizontal sort of construct, if you will, we have been dreaming about that for the last two decades. But we are not there yet. But having said that, on the data side, Google is doing a good job as well. And so is Microsoft. Amazon is, yes. I think on the data side, Amazon is like, they have a little silo problem.

>> Let me get your reaction. I know we only have a couple minutes left, but I want to get your reaction. So one thing that Matt Wood said, I'm going to press him on this, is that I love ... Because I've been saying it for over two years. Models will talk to each other. In fact, we put up the power law on SiliconANGLE, theCUBE research. He's validating everything we said a year and a half ago. So one, I'm happy about that. Just pat myself on the back. But the issue is that, that brings up an issue that's going to change the conversation instantly. One, we're all connected through APIs. 85% of all traffic's API-based.

>> True.

>> If models talk to each other, that means they need to have a mechanism to talk to each other. That's a glue layer. A software has to make that happen. That's a control plane. So we are now moving into that next level conversation. If you believe Matt Wood's statement to be true, models need to talk to each other, and small language models have been all the rage in the past six months. Again, we've been talking about that for over a year. I think we coined the term small language models. Okay, small talks to big, big talks to small. I mean, this is a system. This is coming.

>> It is.

>> This is the app layer then, isn't it?

>> It is. I mean, if you just go to the basic primitive level, first principles thinking level, what is a large language model or even a small language model or smart model, what is it? It's a database which contains intelligence. And you get to it through API. And once you can get to it through API, then it's just traditional database and that sort of-

>> It's just data and it has-

>> -...

>> tons of hallucinations. Swiss cheese called it. And by the way, that's never going to go away because just by the sheer power law analysis, it has to be that way. In fact, the conversation on Twitter last night amongst our group was, most people over the IQ of 130 don't even use large language models. That means smart people don't even use LMs. Why? Because it's too dumb.

>> Yeah.

>> They're smarter than the language model.

>> It's vanilla.

>> It's vanilla. And it's like, why would I want to go down in fourth grade?

>> For whatever reason-

>> I won't say fourth grade. It's got good writing. But the point is, the insights aren't great. But if you mash models up and they interact because our Cube AI talks to OpenAI just for one thing, sentence structure formatting. That's all it does. That's a good use of their model for us. We don't tap it. We don't need their insight. Doesn't know anything about ... OpenAI doesn't know anything about enterprise tech or Amazon Cloud or all the stuff we talk about.

>> Yeah, they are very generic models. So when everybody's using generic models, where will the differentiation come from? Because businesses thrive on differentiation. The differentiation will come from your own data, your own IP, the way you innovate, the way your company's laid out, the way you hire. All these things will still matter.

>> Final question for you. We got a great example of a customer coming on and partners here from Amazon ecosystem. Question for you, Sarbjeet. As we move to the next level now as this next year comes upon us, we're expecting a Cambrian explosion of applications, gen AI applications. What will you see coming, agents first? I mean the chatbot's the joke and that joke is old now. It's not just your yesterday's chatbot. Okay. Get that. Agents are hot right now. Okay. What do you think

>> Just like chatbots were.

>> I know. I was going to say that. What do you think is going to come out next? Where do you see the action on the application side coming? Process improvements and net new, kind of new functionality experiences? What's your tea leaves tell you when you read them?

>> I believe that the application stack, anybody who's working on the primitives of application stack in this new world will be doing better. I think Amazon is doing that, but they look kind of little behind because-

>> I think that's optics, but I get that. But you're saying-

>> -

>> You're saying fix the primitives.

>> Yes.

>> And let everything shake itself out and let the people build.

>> Yeah. I mean, so Werner, they talk about primitives. I think primitives are very important.

>> Werner -

>> -

>> Warner Brothers. Okay.

>> No, not Warner Brothers. So primitives are very important and we have to retrofit the primitives to this new world. That's the-

>> I think you're right on....

>> key takeaway.

>> Great. Let's keep it going. Thanks for coming on. We'll take it offline later. We'll get back on Twitter. Follow Sarbjeet on Twitter. He's got lots of goodies coming every day. He's following all the action. Of course, follow theCUBE, siliconangle.com, theCUBE Research where all the top research goes. We're here in New York City for more action. Here is the post-keynote analysis. We're going to continue on with more interviews. I'm John Furrier, your host of theCUBE. We'll be right back.

Keynote Analysis Pt 2 with Sarbjeet Johal | Stackpane

search

>> Welcome back, everyone. This is theCUBE's coverage in New York City. I'm John Furrier, the host of theCUBE. I'm here with Sarbjeet Johal, who is theCUBE contributor, also Analyst with StackPane. Check him out on Twitter. He's actively watching all the action as well as reporting. And Sarbjeet is always kind enough to be doing keynote reviews with us. And he's got his notebook from the field, which he also does the segment called Notebook from the Field. Sarbjeet, get your notes. Good to see.

>> Got my notebook, yes.

>> We sat next each other in the keynote. Got a nice selfie.

>> Yes, -

>> What's your analysis? I mean, obviously Matt Wood was the star of the show. I mean, I had customers up there, but then they do their thing. "Hey, we're using Cube, blah, blah, blah." But Matt Wood delivered the payload. Generative AI. We are the AI company. What is your analysis?

>> I think they are the builders' cloud. They are the builders' company, if you will, overall. So when you talk about builder, they're ISVs on one side and then in-house developers at the enterprises. Hands down, they are favorite for ISVs. But when it comes to in-house development at enterprises, they are these days with some other competition. They have competition in that domain and a lot more competition than ... I mean, they have competition in ISV space as well, but they have the sort of stronghold on that and they're trying to keep that stronghold as strong. They were. And they had a couple of new announcements which are fresh preview that everybody ... It's industry trend now, right?

>> Yeah, -

>> Preview and then GA.

>> General availability, yeah.

>> But most of the announcements now -

>> Take us through the announcements. What do you got?

>> Yeah, Amazon. They're doing the ... I think the biggest one was Anthropic Claude 3 announcement that the first time you can do fine-tuning of that model. This is the only cloud they say where you can do that. So that was, I think, huge. And there were a lot of enhancements on the queue and they're expanding the scope of the queue from business to low code to pro code. So, spectrum of use cases there. I think the use cases in that segment and queue in the middle are kind of very overlapping. There are a couple of two, three low code sort of productization they have done of the queue, which is, to be honest, it's kind a little noisy. And contextual grounding checks of the models, that was a good one. AWS app studio, that's their low code. Another low code sort of offering, which I'm not a big fan of because I come from the developer background. I'm a pro developer, pro. Not -

>> They don't have a queue version for you yet. Okay. When you say builders' cloud, which love that term because I just love when I first heard that from Andy Jassy. It just makes so ... People are building applications, they're building companies and they want to do it right. When I'm in New York City, I see around, we're on the Javits Center, we're on the Hudson side. It's been booming here for decades. This building's going up and you look up and you see a big building being built. So when I hear building, I think steel, hanging iron, building the floors up and then finishing it up with walls and you put in the elevator, you have all these components. Someone's managing that job, some general contractor, some GC. So now I see we're in an era of cloud, post-cloud 1.0, you got all the hyper scale clouds pumping at all cylinders and you've got a cloud with all this resource and capabilities. And so, it's really a question of not just for builders, what do you build? So I like to see Amazon bring that, here's app studio, here's three steps to build an app. I'm not yet sold on, give me a prompt and I'll build an app for you. That sounds very cliche. oh, come on. Really? Okay, that's good hype and shows like the ChatGPT vibe, that's magic. But I think the real serious developers, I want to build an app. I want to build an app. I want to run it securely. And if I'm a consumer, I'm an entrepreneur, okay, I don't really care about data because I have my own data. Or if I'm a company, I care about my data too. So security matters. So it's a hard nut to crack because you want to be horizontal in that use case. Do you think he met the mark today on the keynote?

>> I think he met the mark on the keynote, but a lot of repetition from the other AI or AWS summits. But there's only very few fractional new things here. But having said that, I think Amazon is one of those companies, of course, but all vendors have this sort of blind spot. I've seen that. I've been in the industry for 30 years now. So they swing the pendulum to the new shiny object a little too much. I think they're doing that. This AI summit is all about AI. This is an AWS summit, not -

>> Who's , Amazon is?

>> Amazon is.

>> people.

>> Amazon is doing that as well.

>> You think they're groping for relevance or do you think they have it?

>> They have-

>> -...

>> substance in the cloud. But from the AI side, they are a little behind. Everybody knows that. And then they're trying to grope -

>> They got the , so they got the silicon, they got the horsepower, they got the platform. And to me, I think it's bedrock has to come together. To me, bedrock is the key.

>> Again for the builders. Builders like those three layers. And then they will embrace that. No doubt about that. I have almost zero doubt about that. But when it comes to the in-house developers at enterprises, I think they will be segregated. Tell me the vendor which doesn't have AI story, gen AI story. Every vendor -

>> Everyone's bolting gen AI in their story and-

>> Exactly....

>> hoping they get a bigger valuation. But right now, we're a year and a half in. I'd say a year past the hype in a full year of production. You're starting to see production workloads hit. Again, low hanging fruit of some nice winds out there. This year is going to be an integration year. And so, we had the partner folks on earlier this morning, Mona, and the person who's running the company center, I know you took a tour at Deloitte, which is another tell sign. Still early innings, but Amazon has an ecosystem. Their entire company was built with people using the platform. And so, to me, I want to see what comes out of the ecosystem. So the question is, what are partners doing? What do you see there? Do you see a robust set of building going on from the partners? I do. I mean, I see a lot of activity.

>> I see.

>> And they're not hand waving either. There's some real production. We're going to have some on this afternoon, so we're going to hear from them. But what's your take? What's the key for them to building gen ed? Do you see what Matt Wood presented, would resonate with partners?

>> Okay. You asked the question with the application stack, and then let's wrap that into-

>> There's five questions in there. Pick your choice.

>> I will answer in a way that it will cover both. So this is ... I think the point which we have to focus on, application stack is going through a big change right now with the introduction of gen AI. So, the way we write applications will change. So, we'll be using many models. It's a multi-model, multimodal world. So how we will weave those models into our applications, that's still to be seen. Right now, what we are cooking up is very rudimentary stuff. Most of the use cases Deloitte showed us yesterday, five of them, they're still ... They're good, solid, but still rudimentary. Under each, they showed us the C-suite AI for CFO, and they had the creative edge for CMOs. It's like marketer in a box. So, you want to do a campaign, just -

>> Was that Deloitte's lab or ... That was the new lab that they're doing?

>> Yeah, in New York they are.

>> In New York City.

>> They took all of us, their analysts there and then did like two hour nonstop session. And then-

>> Yeah, I saw the play from Deloitte on the competency center and also with Amazon. Now they've got the lab there.

>> And we recorded two videos, which you will see that.

>> I don't think that was illegal. I think there was no videos allowed, but send them to us. We'll publish them.

>> Somehow we can do that.

>> Like the Pentagon papers for the videos. Send them to us, we'll publish them.

>> So, application stack is going through a big change. And the -

>> Yeah, the data stack's going under big change too.

>> Of course.

>> I mean, all gen AI apps are fed data, so I get that. But what's the action point right now? You think it's the application layer or is it the infrastructure layer?

>> I think infrastructure layer is almost behind us. We are -

>> You think we're clear on the runway on that?

>> Yes. Yes, we are. And we are going into the application layer. And TCS in the morning yesterday was on a panel, one of their big partners. They were talking about how the data needs to be cleansed in a way or fed in a way which was not previously done. Even the relational database basis, if you just take the rows and columns and tables and feed it to AI, it's not going to work. So you have to label the data properly. And Symantec search or we talk about -

>> Well, that's good news. If you see that the infrastructure, we're clearing the runway on that, that's a game changer because we've been waiting for that. But that's going to cause some consequences because think about all those companies out there, the whole GPU clouds that are out there, or the other things, they might or may not make it. I mean, there is a trajectory here that says, if the app layer takes all the action, all those IAS providers that had those specialty systems may or may not pan out. And heavily funded areas too.

>> Definitely. Yeah.

>> CoreWeave's one of them.

>> of the world and Supermicros. Look where they're flying, their stock is-

>> Well, they're going to AI. They're doing AI effects on premises.

>> Yes.

>> The question is-

>> Do you need that on premise?

>> Well, that's the thing. What's over-provisioning? What does the word, over-provisioning, mean in today's world? Remember the old concept, too much bandwidth, we don't really use it.

>> Low -

>> I got all the servers under capacity. What happened? A glut?

>> Yeah. So I think it was talking earlier about average server utilization. He said 30%. I think it's even lower in private data centers. And the storage utilization was even lower because every business unit will get their and they will not let -

>> So, the stuff's sitting around.

>> Yeah. And same thing will happen here. If people procure-

>> You mean with GPUs?

>> In GPU world, it will happen. If people get all those GPUs, and especially if they get by the business unit.

>> They buy all the GPUs. It's the classic shark fin or the iceberg under the water. Oh my God, it's going to cost us all this to put it together. We didn't see those costs coming. So again, cost exposure works both ways. Not just in the cloud, but also on premise.

>> On premise as well. I see this ... I sound like a broken record when I say that public cloud is the industrialization of backend computing. There's no denial in that factor.

>> Yeah, and they did a good job doing it.

>> And they did a good job.

>> Fantastic job.

>> And then other people copied it, few ... We have top three -

>> And it's sub-optimized.

>> Yeah. sub-optimized.

>> That's called diseconomies of scale.

>> And we are at the AWS summit, let's talk about them a little bit more. They innovate on all strands. There are so many strands of cloud. At high level is network. There's storage and there's compute. Under network, there's so many strands. And policy-based networking is -

>> Is it's too complicated. I mean, the benefit of the cloud is horizontal scalability. Have we gotten too product focused? Because again, gen AI, we're seeing the same movie. Data has to be horizontally scalable, which is not a normal thing. No one ever ... Go back a decade ago. Are you kidding me? No one scales data horizontally. You have your little silos and you got your little data warehouses, in comes Snowflake, in comes Databricks. And Dupe failed. Okay, now it's like, okay, I got to make my data horizontally scalable. I think Amazon has that opportunity to make that layer, pre-app layer absolutely fantastically scalable. And I think, I haven't seen that yet. I haven't seen that emerge, but that's where the puck has to go. I mean, don't you think?

>> I totally agree with you. I think data being in a high horizontal sort of construct, if you will, we have been dreaming about that for the last two decades. But we are not there yet. But having said that, on the data side, Google is doing a good job as well. And so is Microsoft. Amazon is, yes. I think on the data side, Amazon is like, they have a little silo problem.

>> Let me get your reaction. I know we only have a couple minutes left, but I want to get your reaction. So one thing that Matt Wood said, I'm going to press him on this, is that I love ... Because I've been saying it for over two years. Models will talk to each other. In fact, we put up the power law on SiliconANGLE, theCUBE research. He's validating everything we said a year and a half ago. So one, I'm happy about that. Just pat myself on the back. But the issue is that, that brings up an issue that's going to change the conversation instantly. One, we're all connected through APIs. 85% of all traffic's API-based.

>> True.

>> If models talk to each other, that means they need to have a mechanism to talk to each other. That's a glue layer. A software has to make that happen. That's a control plane. So we are now moving into that next level conversation. If you believe Matt Wood's statement to be true, models need to talk to each other, and small language models have been all the rage in the past six months. Again, we've been talking about that for over a year. I think we coined the term small language models. Okay, small talks to big, big talks to small. I mean, this is a system. This is coming.

>> It is.

>> This is the app layer then, isn't it?

>> It is. I mean, if you just go to the basic primitive level, first principles thinking level, what is a large language model or even a small language model or smart model, what is it? It's a database which contains intelligence. And you get to it through API. And once you can get to it through API, then it's just traditional database and that sort of-

>> It's just data and it has-

>> -...

>> tons of hallucinations. Swiss cheese called it. And by the way, that's never going to go away because just by the sheer power law analysis, it has to be that way. In fact, the conversation on Twitter last night amongst our group was, most people over the IQ of 130 don't even use large language models. That means smart people don't even use LMs. Why? Because it's too dumb.

>> Yeah.

>> They're smarter than the language model.

>> It's vanilla.

>> It's vanilla. And it's like, why would I want to go down in fourth grade?

>> For whatever reason-

>> I won't say fourth grade. It's got good writing. But the point is, the insights aren't great. But if you mash models up and they interact because our Cube AI talks to OpenAI just for one thing, sentence structure formatting. That's all it does. That's a good use of their model for us. We don't tap it. We don't need their insight. Doesn't know anything about ... OpenAI doesn't know anything about enterprise tech or Amazon Cloud or all the stuff we talk about.

>> Yeah, they are very generic models. So when everybody's using generic models, where will the differentiation come from? Because businesses thrive on differentiation. The differentiation will come from your own data, your own IP, the way you innovate, the way your company's laid out, the way you hire. All these things will still matter.

>> Final question for you. We got a great example of a customer coming on and partners here from Amazon ecosystem. Question for you, Sarbjeet. As we move to the next level now as this next year comes upon us, we're expecting a Cambrian explosion of applications, gen AI applications. What will you see coming, agents first? I mean the chatbot's the joke and that joke is old now. It's not just your yesterday's chatbot. Okay. Get that. Agents are hot right now. Okay. What do you think

>> Just like chatbots were.

>> I know. I was going to say that. What do you think is going to come out next? Where do you see the action on the application side coming? Process improvements and net new, kind of new functionality experiences? What's your tea leaves tell you when you read them?

>> I believe that the application stack, anybody who's working on the primitives of application stack in this new world will be doing better. I think Amazon is doing that, but they look kind of little behind because-

>> I think that's optics, but I get that. But you're saying-

>> -

>> You're saying fix the primitives.

>> Yes.

>> And let everything shake itself out and let the people build.

>> Yeah. I mean, so Werner, they talk about primitives. I think primitives are very important.

>> Werner -

>> -

>> Warner Brothers. Okay.

>> No, not Warner Brothers. So primitives are very important and we have to retrofit the primitives to this new world. That's the-

>> I think you're right on....

>> key takeaway.

>> Great. Let's keep it going. Thanks for coming on. We'll take it offline later. We'll get back on Twitter. Follow Sarbjeet on Twitter. He's got lots of goodies coming every day. He's following all the action. Of course, follow theCUBE, siliconangle.com, theCUBE Research where all the top research goes. We're here in New York City for more action. Here is the post-keynote analysis. We're going to continue on with more interviews. I'm John Furrier, your host of theCUBE. We'll be right back.