Dreamforce 2024 | Muralidhar Krishnaprasad | Salesforce

Clips
News
More from Dreamforce 2024

Muralidhar Krishnaprasad

President and CTO, Einstein1, Data Cloud, MuleSoft and Tableau

Salesforce

Three insights you might have missed from Dreamforce ’24

Organizations in today’s digital landscape rely on interconnected systems, meaning integrating platforms and unifying operations is critical. As AI becomes a bigger part of modern systems, technologies such as Salesforce Agentforce AI are drawn more in focus.Practitioners, meanwhile, have proven to be the biggest benefactors and influencers of data-driven markets. With its leading role in automation integrations, Salesforce Inc. has had its eye on the broader enterprise AI market. The company’s plans around automation and for Salesforce Agentforce AI were a central focus of the recent Dreamforce event.Christophe Bertrand and John Furrier of theCUBE talk during Dreamforce 2024

Agentforce leads the next evolution for Salesforce in its AI offerings

Salesforce Inc. is making a major push to deploy AI agents on its CRM platform, an initiative the company views as the next step in enterprise artificial intelligence adoption.Building on its predictive Einstein platform for sales, service, marketing and commerce, Salesforce announced Agentforce this month as the latest chapter in the development of AI solutions.Muralidhar Krishnaprasad, president and CTO at Salesforce, talks about AI agents with theCUBE.“Think of it as the next evolution of our AI wave,” said Muralidhar Krishnaprasad (pictured), president and chief technology officer at Salesforce. “We had AI wave one, which was our Einstein predictive

play_circle_outline Introduction to Agentforce and its role in AI evolution

play_circle_outline Importance of harmonized data in agent success

play_circle_outline Agent Builder Platform for easy creation of agents

play_circle_outline Enhancing Application Development with Salesforce Integration and Semantic Model for Streamlined Analytics and Actioning

play_circle_outline Maximizing Efficiency Through Agent-Driven Marketing, Sales, Service, Commerce, and Analytics Integration with Tableau Demos

Info

Muralidhar Krishnaprasad | Salesforce

Muralidhar Krishnaprasad

President and CTO, Einstein1, Data Cloud, MuleSoft and Tableau Salesforce

search

Rob Strechay

>> Hello and welcome back to Dreamforce 2024 where we're really digging into what's been going on. There's been a ton of announcements last Thursday around Agentforce and Data Cloud and we're going to be digging into all of this. I'm Rob Stretchay, a managing director with theCUBEResearch, and I'm joined by George Gilbert who's been doing a ton into these data platforms. And to really bring it home, I'm so excited to be joined by MK, who's the president and CTO over Einstein 1, Data Cloud MuleSoft and Tableau. So great to have you on, MK, because I think again, when we start to dig into what some of these things that you've been announcing are, there's some really impressive stuff, I would say in the world of agents going on. And for us, we look at this as a huge piece of making AI a reality for customers. So why don't you help us understand what some of the things that have been announced, what may be some of the things coming today and what we will be talking about into the future here?

>> Awesome. First of all, thank you. Privilege to be here. It's a great, I guess audience to be able to talk to. So what you're going to be seeing all through Dreamforce is this Agentforce. What Agentforce is, think of it as a next evolution of our AI wave. We had AI wave one, which was our Einstein's predictive. The AI wave two is all our copilots with generative AI, and really we're now entering the age of agents. What agents in a very simplified term is just AI that's the next level, which can plan better and it can do more complex tasks for you and in many cases, it'll eventually to the more autonomous AI which can do things for you automatically. And the big announcements we have is that our platform is going to be one of the most comprehensive platforms for agent development, and that means think about any business in the world. What do you do normally? You do advertising, marketing, you do sales, you do online commerce, you do service, and you do analytics. Imagine all of them on a platform on which now you can have agents that can help you through all of these process and more. That's our big announcement. And part of the things, also all the developer tools around it, all the data that's going to go around it and all the interesting reasoning capabilities that make it possible in the line of work. Those are our big announcements.

>> So maybe let's elaborate on that. One of the things that would distinguish ... You did the foundational work with Data Cloud, which provides like harmonized context, and by harmonized I mean we get away from 60 years of silos of application automation and analytics and you've up-leveled that to the customer 360, the customer journey. But then when we build agent-based applications, there's a couple of things that seem to be unique. One is it's a breakthrough in enabling corporate developers to build sophisticated applications because they can teach them rather than specifying exactly what has to happen. You seem to have several unique differentiators. One is the harmonized data, so you're dealing with customer 360 journey, so the people, places and things and activities, not social security numbers, bits and bytes. The other is you're integrated within the surfaces of your operational applications-

>> Correct....

>> so that the outcomes of the agent's actions become teachable moments. Elaborate on that and elaborate on these differentiators that everyone who's coming at agents from different angles that can't put all those pieces together.

>> Beautiful. I think I can answer all that with one sentence, which is data plus agents plus CRM plus humans. Let me elaborate that. What I mean by that is, let's say if your ChatGPT or whatever says there are two Rs in strawberry, we all just laugh at it and move on, and yes, they have fixed it in GPT-4 now, but really imagine if you now rely on that agent to tell you what's your next quarter's results going to be. If you're going to rely on your agents to do your business, or even more, if you're putting the agents in front of your customers and the customer is asking a question, it better respond right. And how does it respond? It's not like people are asking some random questions about the internet or the world wide web. They're really asking about your business's data. And so data becomes a very, very important ingredient in an agent's success. But if you look at every company in the world, I can guarantee every company, including Salesforce, data is fragmented. It's just a simple customer information. If I say, "George," it's probably stored in 15 different systems. It's just natural evolution of these systems. And so what we did with Data Cloud was to be able to harmonize all that data first. And harmonize doesn't mean copy all that data because people will be like, "We are not trying to create a data gravity and suck all your data." We are saying some cases you'll pull the data to us, some cases you can leave the data where it is. We have zero copy. That's the first level. The second level is creating that harmonized canonical reference model because something as simple as a user is called a contact in sales call. It's called a profile on your website. It's called a subscriber marketing. Schema shapes are so different and so we harmonize it. Then the next step is identity resolution, because George, you probably have five different IDs across the systems. And an interesting example is we used to think in salesforce.com that we had 200 million visitors. It turned out after we did Data Cloud, it was really 98 or 100 million because people had different IDs. They're the same person. And so this then unifies it, and that helps us get the 360 view. I know exactly George or Rob, okay, you opened this email, you had this service case, sales thing and all that stuff. And also it's important that because with Salesforce we have all those applications, we have all that data flowing in, whether you open an email or not, whether you have a case or not, whether you purchase a sales or not. Then the next step is we want to be able to do RAG on top of them, and this is not just structured data with vector databases and embedding, we have all the unstructured pipe as well. So with this structured and unstructured data, you want to ground it with your agents. That's where the data plus agents come in, and this is where we give a lot of tools like Prompt Studio and others for admins to customize it, because your company's brand voice may be different than say my company's brand voice. So you want customize it, and so we give you all those tools as developers. And then finally you want to expose those agents in the line of work, whether it's a service agent in your service case, whether it's a salesperson or a marketing agent that is helping you with your campaigns. But there's one more thing too. Our agent's also differentiated because you can escalate to humans. Very, very important because we are built on the same bots and other foundation. The final thing is that feedback is flowing too. When your agent is making a sale recommendation, the sale actually closes. We have that data flowing back. So that's really that data plus agents plus CRM plus humans with the feedback that's closing the loop.

>> That seems to be, as far as I can tell, unique. There are other vendors more on the data and cloud application side that have long investment in low-code tools, but they don't have the feedback loop. They might have the supervision, they don't have the harmonized data.

>> That's right.

>> So maybe talk about how if a customer is trying to picture what's really a breakthrough both in accessibility of enterprise software and the ability to customize the 80% of activities we never could reach because you can learn now instead of specify. How do you get closer? It's not a, "Did I build out this process or application?" But how do I collect the teachable moments to cover more interactions more reliably?

>> Yeah, so the way it is because the data is flowing, first of all, that's important. The feedback data is flowing. And I say feedback data, it's not just whether you clicked on a website or not. That's certainly engagement data. It's all your business level data too. Whether your sales closed, your service closed or your CSAT scores are people are leaving reviews on your website. Are they good? Are they bad? All of those things are flowing, and so if you remember in predictive days, that loop was important for your ML algorithm to get better. But in this new world, you're not going to go modify OpenAI or other tools. What you're really doing is making your instructions better. A great example is, for example, the Dreamforce app itself. We've been fine-tuning it and we're not changing code. We're changing instructions and we are learning based on what people are asking, what responses. So you're fine-tuning that system, and many cases we are automating that fine-tuning too. And this is true in the security landscape as well where you have a lot of security signals flowing and we want to automate. And so the key thing we can do in this new world is giving better instructions and fine-tuning the instructions based on all of the feedback that is flowing, what is working, what is not working.

>> So it's a new development cycle where before you had an agile release cycle with microservices, but here you need to be collecting these ... You want to get best practice for each task to be a narrow distribution and higher reliability. And so part of the tools is collecting the human-supervised reactions where they might edit a plan or provide feedback or just feedback directly from the app as to whether the plan succeeded. And then in batch mode, I guess you periodically update.

>> So there's two parts to it. One is you want to have a dynamic plan because you don't know what questions the customers are going to ask. You're clearly opening up your agent to all of them. And so part of the thing is we have this reasoning engine, we call it the Atlas reasoning engine, where we use very interesting techniques called Ensemble, RAG and other things along with hierarchical planning. So which means you want to try out, you want to go do something, get some results and then keep learning from it. Actually, if you ask the LLM again, and there are more techniques coming in as well, where you can actually ask the question again to the LLM, it actually defines the answer. So there are interesting techniques that we are learning where you can create dynamic plans based on the questions that are coming in, the data that is relevant to that question and the feedback back from that user. So there is a little bit of a dynamism in that as well. As well as we want to learn from the bigger loop to say, "Okay, this is a common occurrence. Maybe our topic needs more fine-tuning. Maybe the instructions we give to the prompt needs to be tuned." That's the outer loop. So there's an inner loop that's happening in real-time as a conversations going on, there's an outer loop. And I guess, like you mentioned earlier, we have the goodness that we have all the application surfaces and that feedback data is super, super valuable, to make sure the sale is done, the service is handled correctly, or the marketing campaign is actually working as determined.

Rob Strechay

>> And I think one of the things that's really interesting is the fact that everybody, as you said, maybe they not only spell strawberry differently, but how they define revenue across all your customers has some intricacies to that. And when you start to look at when you're building for that persona that is going to build these agentic applications, these data products and data apps on top of Salesforce, how do you look at it and say, "Hey, here's how they're going to start." From, like you said, connecting up mainframes when we were talking earlier and all the way through. How do you look at that as well and say, "Yeah, this is how the personas are very different from the person who deals with the mainframe to the person who's dealing with Salesforce or the CRM aspect or Service Cloud." How do you really look at that so that they can build those easily, to build those -

>> Great question. So where we are starting off is a simple concept. We call it the agent builder platform. So all the 25 years that you may be building Apex code, APIs, data, everything else, we want to make sure all that knowledge can be used. So this is not something completely new stack where you were to start from scratch. We are saying start with your base. If you already have Sales and Service cloud, et cetera, you already have predefined business processes. You can expose them all now through these new agents, and this is where we have the notion of a topic where you explain, "This is what this thing does. This thing could be a business process that's about closing a sale or about creating a service case and so on." The second thing is with the power of something like MuleSoft that we have, we can actually make mainframe stock. Not even mainframes, even your old Visual Studio or Visual Basic website you may have, but nobody even is there probably to look at the code. Using RPAs and others, we can make all that talk. So that is on the API level. On the data level, we talked about Data Cloud, how we can now bring all the harmonious data together and actually surface it together. This along with all the flow and other actions can be used in your agent builder. And so all you have to do ... By the way, this matter of minutes or hours, you can easily create these agents, which is you just define the topic. You say, "You know what? I want something to go satisfy my service resolution." And it could be calling this particular flow, looking up these particular knowledge articles unstructured and maybe calling a MuleSoft action. With just that instruction, you have an agent ready and then you published your website. That's about it. So we're bringing forth every investment you have done in the last whatever, 20, 30 years to this new wave of AI.

>> Let me just ask pop back up a level of abstraction. There's been a dream for decades for applications to also be construction sets, but we only could limit it to configuration before we'd break it and you couldn't upgrade it. To what extent does the center of gravity of Salesforce now shift from application to application construction set where all the flows and the data models now are just building blocks and you can build unique operational applications that are formed by analytics?

>> Great question. So one of the things you will see also in Dreamforce, we are announcing this thing called Tableau Einstein, as you might have heard. So what we are doing is we are bringing the entire analytical platform also on the same data cloud and Agentforce platform. What it means is, to your earlier question, Rob, you said how people might define semantics differently. In fact, that's a problem because I might think ROI means this and then you have a different version of it. So we have defined this new thing called a semantic model. So you can actually have one semantic model defined that works not only for your flow but also for your agents across the stack. The second thing is we're bringing all of that visual analytic and actioning frame into the same stack. That way your analytics is not a siloed thing on one side, but it's also part of your line of work. That's like a big thing we're doing. To your earlier question, what we are looking at is agents going to redefine everything we're doing. Everything a marketer is doing, everything a service person is doing, everything a salesperson is doing and even beyond. And our goal is to make sure we have the common platforms so that you truly can have a super agent. Think about it. I go to a website, I don't care whether there's a marketer behind it, a salesperson, a service person, I want my job done. It could be a question that I ask saying, "Hey, where's my shipping order?" It could also be like, "Hey, do you have the latest inventory for what I'm looking for?" To closing a deal kind of thing. And so we want to make sure these things flow seamlessly into these agents and they're able to actually do multiple persona and answer the right things. So that's a big one. And from an application construction perspective, like we chatted earlier, I think we may need more people knowing English and MPhils and those kind of things because more and more it's going to be fine-tuning and asking the right question. Just like we all learned how to ask questions in Google when search came out. Same thing, application programming and everything is going to change to be instructing what to do and aligning with your business thing. That's the big change I'm seeing come.

Rob Strechay

>> Yeah. Being able to describe the architecture and how the people, places and things come together is going to be super important. I think one of the other things, and as we're getting towards the end here, you've had incredible growth of Data Cloud. It's year over year growing, I think at 130 plus percent.

>> Yeah, it's probably the fastest growing ever. Yeah.

Rob Strechay

>> So congratulations on that. When you start to look at this and say, "Hey, organizations are ..." Like you said, they're not going to go and use you to create their foundation model, that's not the business you're in. They're going to bring it, they're going to ground it, fine tune it.

>> That's right.

Rob Strechay

>> Do you see that as how you're helping the process? And are the agents helping them through that process or is it, "Hey, we've set it up inside the SaaS framework so that they can bring themselves through that"?

>> Great point. So we have both. We have made it all into SaaS so that it's easy. So you have your platforms you can connect. Now, we also recognize that people may have a lot of investments, whether it's on the data side or even the model side, maybe they're fine-tuning their own models in many cases. What we have done is an open and extensible platform. That's the zero copy for plugging in your data sources, that is the bring your own M for predictive, or bring your own LLM for LLMs too. So we allow you to plug and play, but then we have SaaSified that whole experience so that it's as simple as you don't need to know where that data came from. You don't need to know where that model came from. As a marketeer, sales, service, commerce person or analytics person, I just want to use the product. It works. And within all of those experiences, we obviously use agents to help it. If you see the new Tableau demos, you can use it yourself, Tableau agents, you can just ask and instruction. It creates a beautiful dashboard for you. You don't need to keep drag-dropping. Or you're in a flow. You can just say, "I want this flow to be created." Or MuleSoft or Sales Cloud, you can just say, "Create my agent for me." So we are doing it both ways. We're making it easy without pluggable things, so you can go on an existing enterprise landscape and integrate to it, and then we're using agents for configuring all of these applications itself to make it even more easier.

Rob Strechay

>> Yeah. That would seem to help across the entire set of personas that are actually using you. And I really appreciate you coming on board MK. Again, I know George and I have way more questions. We'll have to do a follow-up because we could dig in probably for about another hour, and I think you're so full of knowledgeable stuff.

>> Two, at least.

Rob Strechay

>> Maybe two.

>> IK would love to.

Rob Strechay

>> So thank you for coming on board.

>> Thank you.

Rob Strechay

>> And thank you for watching here at Dreamforce 2024. We'll be back shortly, so stay tuned and keep it right there.

Muralidhar Krishnaprasad | Salesforce

search

Rob Strechay

>> Hello and welcome back to Dreamforce 2024 where we're really digging into what's been going on. There's been a ton of announcements last Thursday around Agentforce and Data Cloud and we're going to be digging into all of this. I'm Rob Stretchay, a managing director with theCUBEResearch, and I'm joined by George Gilbert who's been doing a ton into these data platforms. And to really bring it home, I'm so excited to be joined by MK, who's the president and CTO over Einstein 1, Data Cloud MuleSoft and Tableau. So great to have you on, MK, because I think again, when we start to dig into what some of these things that you've been announcing are, there's some really impressive stuff, I would say in the world of agents going on. And for us, we look at this as a huge piece of making AI a reality for customers. So why don't you help us understand what some of the things that have been announced, what may be some of the things coming today and what we will be talking about into the future here?

>> Awesome. First of all, thank you. Privilege to be here. It's a great, I guess audience to be able to talk to. So what you're going to be seeing all through Dreamforce is this Agentforce. What Agentforce is, think of it as a next evolution of our AI wave. We had AI wave one, which was our Einstein's predictive. The AI wave two is all our copilots with generative AI, and really we're now entering the age of agents. What agents in a very simplified term is just AI that's the next level, which can plan better and it can do more complex tasks for you and in many cases, it'll eventually to the more autonomous AI which can do things for you automatically. And the big announcements we have is that our platform is going to be one of the most comprehensive platforms for agent development, and that means think about any business in the world. What do you do normally? You do advertising, marketing, you do sales, you do online commerce, you do service, and you do analytics. Imagine all of them on a platform on which now you can have agents that can help you through all of these process and more. That's our big announcement. And part of the things, also all the developer tools around it, all the data that's going to go around it and all the interesting reasoning capabilities that make it possible in the line of work. Those are our big announcements.

>> So maybe let's elaborate on that. One of the things that would distinguish ... You did the foundational work with Data Cloud, which provides like harmonized context, and by harmonized I mean we get away from 60 years of silos of application automation and analytics and you've up-leveled that to the customer 360, the customer journey. But then when we build agent-based applications, there's a couple of things that seem to be unique. One is it's a breakthrough in enabling corporate developers to build sophisticated applications because they can teach them rather than specifying exactly what has to happen. You seem to have several unique differentiators. One is the harmonized data, so you're dealing with customer 360 journey, so the people, places and things and activities, not social security numbers, bits and bytes. The other is you're integrated within the surfaces of your operational applications-

>> Correct....

>> so that the outcomes of the agent's actions become teachable moments. Elaborate on that and elaborate on these differentiators that everyone who's coming at agents from different angles that can't put all those pieces together.

>> Beautiful. I think I can answer all that with one sentence, which is data plus agents plus CRM plus humans. Let me elaborate that. What I mean by that is, let's say if your ChatGPT or whatever says there are two Rs in strawberry, we all just laugh at it and move on, and yes, they have fixed it in GPT-4 now, but really imagine if you now rely on that agent to tell you what's your next quarter's results going to be. If you're going to rely on your agents to do your business, or even more, if you're putting the agents in front of your customers and the customer is asking a question, it better respond right. And how does it respond? It's not like people are asking some random questions about the internet or the world wide web. They're really asking about your business's data. And so data becomes a very, very important ingredient in an agent's success. But if you look at every company in the world, I can guarantee every company, including Salesforce, data is fragmented. It's just a simple customer information. If I say, "George," it's probably stored in 15 different systems. It's just natural evolution of these systems. And so what we did with Data Cloud was to be able to harmonize all that data first. And harmonize doesn't mean copy all that data because people will be like, "We are not trying to create a data gravity and suck all your data." We are saying some cases you'll pull the data to us, some cases you can leave the data where it is. We have zero copy. That's the first level. The second level is creating that harmonized canonical reference model because something as simple as a user is called a contact in sales call. It's called a profile on your website. It's called a subscriber marketing. Schema shapes are so different and so we harmonize it. Then the next step is identity resolution, because George, you probably have five different IDs across the systems. And an interesting example is we used to think in salesforce.com that we had 200 million visitors. It turned out after we did Data Cloud, it was really 98 or 100 million because people had different IDs. They're the same person. And so this then unifies it, and that helps us get the 360 view. I know exactly George or Rob, okay, you opened this email, you had this service case, sales thing and all that stuff. And also it's important that because with Salesforce we have all those applications, we have all that data flowing in, whether you open an email or not, whether you have a case or not, whether you purchase a sales or not. Then the next step is we want to be able to do RAG on top of them, and this is not just structured data with vector databases and embedding, we have all the unstructured pipe as well. So with this structured and unstructured data, you want to ground it with your agents. That's where the data plus agents come in, and this is where we give a lot of tools like Prompt Studio and others for admins to customize it, because your company's brand voice may be different than say my company's brand voice. So you want customize it, and so we give you all those tools as developers. And then finally you want to expose those agents in the line of work, whether it's a service agent in your service case, whether it's a salesperson or a marketing agent that is helping you with your campaigns. But there's one more thing too. Our agent's also differentiated because you can escalate to humans. Very, very important because we are built on the same bots and other foundation. The final thing is that feedback is flowing too. When your agent is making a sale recommendation, the sale actually closes. We have that data flowing back. So that's really that data plus agents plus CRM plus humans with the feedback that's closing the loop.

>> That seems to be, as far as I can tell, unique. There are other vendors more on the data and cloud application side that have long investment in low-code tools, but they don't have the feedback loop. They might have the supervision, they don't have the harmonized data.

>> That's right.

>> So maybe talk about how if a customer is trying to picture what's really a breakthrough both in accessibility of enterprise software and the ability to customize the 80% of activities we never could reach because you can learn now instead of specify. How do you get closer? It's not a, "Did I build out this process or application?" But how do I collect the teachable moments to cover more interactions more reliably?

>> Yeah, so the way it is because the data is flowing, first of all, that's important. The feedback data is flowing. And I say feedback data, it's not just whether you clicked on a website or not. That's certainly engagement data. It's all your business level data too. Whether your sales closed, your service closed or your CSAT scores are people are leaving reviews on your website. Are they good? Are they bad? All of those things are flowing, and so if you remember in predictive days, that loop was important for your ML algorithm to get better. But in this new world, you're not going to go modify OpenAI or other tools. What you're really doing is making your instructions better. A great example is, for example, the Dreamforce app itself. We've been fine-tuning it and we're not changing code. We're changing instructions and we are learning based on what people are asking, what responses. So you're fine-tuning that system, and many cases we are automating that fine-tuning too. And this is true in the security landscape as well where you have a lot of security signals flowing and we want to automate. And so the key thing we can do in this new world is giving better instructions and fine-tuning the instructions based on all of the feedback that is flowing, what is working, what is not working.

>> So it's a new development cycle where before you had an agile release cycle with microservices, but here you need to be collecting these ... You want to get best practice for each task to be a narrow distribution and higher reliability. And so part of the tools is collecting the human-supervised reactions where they might edit a plan or provide feedback or just feedback directly from the app as to whether the plan succeeded. And then in batch mode, I guess you periodically update.

>> So there's two parts to it. One is you want to have a dynamic plan because you don't know what questions the customers are going to ask. You're clearly opening up your agent to all of them. And so part of the thing is we have this reasoning engine, we call it the Atlas reasoning engine, where we use very interesting techniques called Ensemble, RAG and other things along with hierarchical planning. So which means you want to try out, you want to go do something, get some results and then keep learning from it. Actually, if you ask the LLM again, and there are more techniques coming in as well, where you can actually ask the question again to the LLM, it actually defines the answer. So there are interesting techniques that we are learning where you can create dynamic plans based on the questions that are coming in, the data that is relevant to that question and the feedback back from that user. So there is a little bit of a dynamism in that as well. As well as we want to learn from the bigger loop to say, "Okay, this is a common occurrence. Maybe our topic needs more fine-tuning. Maybe the instructions we give to the prompt needs to be tuned." That's the outer loop. So there's an inner loop that's happening in real-time as a conversations going on, there's an outer loop. And I guess, like you mentioned earlier, we have the goodness that we have all the application surfaces and that feedback data is super, super valuable, to make sure the sale is done, the service is handled correctly, or the marketing campaign is actually working as determined.

Rob Strechay

>> And I think one of the things that's really interesting is the fact that everybody, as you said, maybe they not only spell strawberry differently, but how they define revenue across all your customers has some intricacies to that. And when you start to look at when you're building for that persona that is going to build these agentic applications, these data products and data apps on top of Salesforce, how do you look at it and say, "Hey, here's how they're going to start." From, like you said, connecting up mainframes when we were talking earlier and all the way through. How do you look at that as well and say, "Yeah, this is how the personas are very different from the person who deals with the mainframe to the person who's dealing with Salesforce or the CRM aspect or Service Cloud." How do you really look at that so that they can build those easily, to build those -

>> Great question. So where we are starting off is a simple concept. We call it the agent builder platform. So all the 25 years that you may be building Apex code, APIs, data, everything else, we want to make sure all that knowledge can be used. So this is not something completely new stack where you were to start from scratch. We are saying start with your base. If you already have Sales and Service cloud, et cetera, you already have predefined business processes. You can expose them all now through these new agents, and this is where we have the notion of a topic where you explain, "This is what this thing does. This thing could be a business process that's about closing a sale or about creating a service case and so on." The second thing is with the power of something like MuleSoft that we have, we can actually make mainframe stock. Not even mainframes, even your old Visual Studio or Visual Basic website you may have, but nobody even is there probably to look at the code. Using RPAs and others, we can make all that talk. So that is on the API level. On the data level, we talked about Data Cloud, how we can now bring all the harmonious data together and actually surface it together. This along with all the flow and other actions can be used in your agent builder. And so all you have to do ... By the way, this matter of minutes or hours, you can easily create these agents, which is you just define the topic. You say, "You know what? I want something to go satisfy my service resolution." And it could be calling this particular flow, looking up these particular knowledge articles unstructured and maybe calling a MuleSoft action. With just that instruction, you have an agent ready and then you published your website. That's about it. So we're bringing forth every investment you have done in the last whatever, 20, 30 years to this new wave of AI.

>> Let me just ask pop back up a level of abstraction. There's been a dream for decades for applications to also be construction sets, but we only could limit it to configuration before we'd break it and you couldn't upgrade it. To what extent does the center of gravity of Salesforce now shift from application to application construction set where all the flows and the data models now are just building blocks and you can build unique operational applications that are formed by analytics?

>> Great question. So one of the things you will see also in Dreamforce, we are announcing this thing called Tableau Einstein, as you might have heard. So what we are doing is we are bringing the entire analytical platform also on the same data cloud and Agentforce platform. What it means is, to your earlier question, Rob, you said how people might define semantics differently. In fact, that's a problem because I might think ROI means this and then you have a different version of it. So we have defined this new thing called a semantic model. So you can actually have one semantic model defined that works not only for your flow but also for your agents across the stack. The second thing is we're bringing all of that visual analytic and actioning frame into the same stack. That way your analytics is not a siloed thing on one side, but it's also part of your line of work. That's like a big thing we're doing. To your earlier question, what we are looking at is agents going to redefine everything we're doing. Everything a marketer is doing, everything a service person is doing, everything a salesperson is doing and even beyond. And our goal is to make sure we have the common platforms so that you truly can have a super agent. Think about it. I go to a website, I don't care whether there's a marketer behind it, a salesperson, a service person, I want my job done. It could be a question that I ask saying, "Hey, where's my shipping order?" It could also be like, "Hey, do you have the latest inventory for what I'm looking for?" To closing a deal kind of thing. And so we want to make sure these things flow seamlessly into these agents and they're able to actually do multiple persona and answer the right things. So that's a big one. And from an application construction perspective, like we chatted earlier, I think we may need more people knowing English and MPhils and those kind of things because more and more it's going to be fine-tuning and asking the right question. Just like we all learned how to ask questions in Google when search came out. Same thing, application programming and everything is going to change to be instructing what to do and aligning with your business thing. That's the big change I'm seeing come.

Rob Strechay

>> Yeah. Being able to describe the architecture and how the people, places and things come together is going to be super important. I think one of the other things, and as we're getting towards the end here, you've had incredible growth of Data Cloud. It's year over year growing, I think at 130 plus percent.

>> Yeah, it's probably the fastest growing ever. Yeah.

Rob Strechay

>> So congratulations on that. When you start to look at this and say, "Hey, organizations are ..." Like you said, they're not going to go and use you to create their foundation model, that's not the business you're in. They're going to bring it, they're going to ground it, fine tune it.

>> That's right.

Rob Strechay

>> Do you see that as how you're helping the process? And are the agents helping them through that process or is it, "Hey, we've set it up inside the SaaS framework so that they can bring themselves through that"?

>> Great point. So we have both. We have made it all into SaaS so that it's easy. So you have your platforms you can connect. Now, we also recognize that people may have a lot of investments, whether it's on the data side or even the model side, maybe they're fine-tuning their own models in many cases. What we have done is an open and extensible platform. That's the zero copy for plugging in your data sources, that is the bring your own M for predictive, or bring your own LLM for LLMs too. So we allow you to plug and play, but then we have SaaSified that whole experience so that it's as simple as you don't need to know where that data came from. You don't need to know where that model came from. As a marketeer, sales, service, commerce person or analytics person, I just want to use the product. It works. And within all of those experiences, we obviously use agents to help it. If you see the new Tableau demos, you can use it yourself, Tableau agents, you can just ask and instruction. It creates a beautiful dashboard for you. You don't need to keep drag-dropping. Or you're in a flow. You can just say, "I want this flow to be created." Or MuleSoft or Sales Cloud, you can just say, "Create my agent for me." So we are doing it both ways. We're making it easy without pluggable things, so you can go on an existing enterprise landscape and integrate to it, and then we're using agents for configuring all of these applications itself to make it even more easier.

Rob Strechay

>> Yeah. That would seem to help across the entire set of personas that are actually using you. And I really appreciate you coming on board MK. Again, I know George and I have way more questions. We'll have to do a follow-up because we could dig in probably for about another hour, and I think you're so full of knowledgeable stuff.

>> Two, at least.

Rob Strechay

>> Maybe two.

>> IK would love to.

Rob Strechay

>> So thank you for coming on board.

>> Thank you.

Rob Strechay

>> And thank you for watching here at Dreamforce 2024. We'll be back shortly, so stay tuned and keep it right there.