Episode Transcript
[00:00:00] Okay, Good morning everyone. Today I want to talk about the SAP Connect conference this week, which was very interesting. I was there for a couple of days, met a lot of the executives, saw a lot of the demos, and then of course next week is the Workday Rising conference and there'll be lots of things going on there, including some announcements from us. And then I want to talk about multi agents and multi agent architectures, which is a piece of research and a guidebook we're announcing and releasing tomorrow. So. So SAP is a German software company and that's pretty much obvious from all of the announcements they made at the conference. The new theme for SAP is the Autonomous enterprise. So what they're basically doing, building and saying is that we're going to give you AI agents that work in every one of your five major business areas, finance, supply chain, hr, customer and manufacturing operations. And we're going to give you a set of tools, tools called the SAP Knowledge Graph that take all of the knowledge and information and content you have about your company and expose it to these agents that will automate the thousands of things that happen in your company.
[00:01:15] And the interface to that is Joule work. J O U L E A Joule is a unit of energy, like a Watt, if you know what a Watt is, or a btu.
[00:01:25] So what they demoed was all sorts of juul work. Examples of how a purchasing agent, a supply chain manager, an HR manager or recruiter or a marketing person could use juul work to use the information about the company to ask questions, identify anomalies, fix things, improve things, change suppliers, fix financial issues, clean up the payroll, et cetera, and the agent infrastructure. Under this is a set of tools that allow you to build agents and identify workflows in your company today that you might want to re engineer around a new agentic workflow. And they also acquired a company called Tech Wolf, which is an 8 year old AI startup that has been doing AI inference in jobs and skills and tasks. They have a incredible series of tools that include WalkMe and other tools to take tacit information about how things are working today so you can re engineer them. And the Jewel Work studio is actually quite impressive on how you can build an agent around the knowledge you currently have about how your company works. So basically what they're kind of saying is everything in SAP that you have today exists and remains. It doesn't matter what version you have. And we're going to give you all of these agents on top to build agentic applications on top of the existing SAP tools. And systems you currently have today, which is great. I mean it's exactly what people want. Now what I think is missing in the vision, and I actually talked to Gina the Chro about this and a few of the other execs is they don't really have a vision for what your company should look like in the future. We of course do with HR 2030 and I'll talk a little bit more about that, but you'll hear more about that next week. So they're not essentially explaining to you how to use the AI yet or what new business processes you should build. They're more or less assuming that you or a systems integrator or consultant will figure that out. But you will use their tools because they're safe. They have management interfaces, they have a knowledge graph. They can access non SAP data through their knowledge and data fabric.
[00:03:50] Really, really sort of sound industrial strength software. And they are very clearly in the position, which I am too, that they're going to give you the tools to protect agents from doing things they're not supposed to do. Now I just published an article today, this is coming out in a more formal form on Thursday, that it's a little bit up to you whether your agents can misbehave. Because if you don't architect your agents with specialization and clear guidelines and clear centers of responsibility and any agent could try to do anything.
[00:04:27] And not that it would necessarily have access to data that it's not supposed to, but maybe you could find out how to work around your current systems. So some of this is still in your hands as a corporate person, but SAP is giving you lots and lots of tools for that. And then in the SuccessFactors business they're starting to see the interactions between these different agents and how they might work together in more business scenarios. Now, SuccessFactors still has two agent stacks. They have the agents they just bought from Smart Recruiters and then they have the agents from Joule. They're not planning on merging those yet, but they will, I'm sure, over time because the agents from Smart Recruiters were designed for candidates, whereas the Joule agents only work for SAP users. So you can't give it to an external party because it's got all your company information in it. Anyway, it's a very nice software story. And the reason I say they're a software company, I don't mean that negatively at all. They're obviously going to monetize this through credits and tokens and usage fees and so forth, just like everybody else, is that when you Listen to Christian Klein talk about the company and of course they have very large customers and they probably provide software to the largest number of large companies in the world. Oracle and Workday are not really at their scale. I mean, they're in the same range, but they're not really as nearly as big as SAP. He really talks a lot about the software and what he wants to sell them. So they're not a problem solving company per se, but they're a really good engineering company and a really good software company. And historically, the way SAP has gone to market, I mean, going back to the beginning of SAP, I remember when the company was a mainframe company, they really did try to automate business itself and that's how the ERP was created. They're not really talking about that now, they're just talking about this incredibly rich, robust set of AI tools and agents that come with the system.
[00:06:23] So for companies that are SAP already in manufacturing, in oil and gas, in transportation, many of the industries that are telecommunications that are very heavily SAP dominated, where they have lots of industry expertise, this is going to be really good stuff. People are going to like it because it's a tool set that lays on top of what they already have. It's open, it's contemporary, it's modern, it's fast, appears to be fast. And we have a relationship with SAP to implement Galileo within SAP that's been in a prototype stage for a little while. So those of you that are SAP customers that want to get Galileo, just let us know and we'll show you where that is. And you know, when you walk around the conference, it's a pretty big conference. There's, you know, obviously a lot of SAP users. I get the feeling that most of the people who come to SAP's events are technologists or IT people. There were a lot of HR people at this conference because it used to be Success Connect, it used to be Successful Factors Conference and now it's covering all of SAP. So there are a lot of HR people and I think over the years watching SAP's presentations, most of the years I've gone to them, I was bored because they didn't really talk about hr at all. SuccessFactors was a separate company. I don't know if most of you know this, but SuccessFactors was an acquisition. It was a small, very fast moving, high growth software company founded by a very talented entrepreneur that was acquired by SAP many years ago and then ported over to all the SAP infrastructure over many years. And little by little it got more and more integrated into the SAP stack. And I think this year you can see that the hr part, the SuccessFactors part, really is an integral part of the whole SAP infrastructure. It's not a standalone company or a standalone product anymore, with the exception of all the acquisitions. And the other thing about SAP, which is true of Oracle and more and more, more of Workday is a lot of the functional areas of the system are acquisitions and they don't even have names that are SAP names. So a lot of the components and the products have the names of their original companies that they came from. And somehow SAP doesn't mind that. I'm not sure that would be the way I would do it, but. So you have to kind of get to know the jargon of what all these things are. I was in a lot of analyst meetings where people were throwing around product names and, and all of the, you know, IDC analysts who've been the supply chain guys know everything about SAP. I'm raising my hand, what is this? What is that?
[00:08:56] But eventually I figured out what it is. And in the area of AI, they, they've acquired several companies recently and they also acquired WalkMe, which was a little bit pre AI. So they've had lots of attempts to build intelligent systems on top of SAP for many, many decades. That because the SAP system is so powerful, it's been, it's always been a little bit hard to use. So they've had lots of things where they've done over the years to build more and more intelligence into the front end or into the back end and for analytics and so forth. So it's a really impressive system and it's a very impressive company. The management team is in a great mood. You can tell they're, they're feeling good about the world. And customers I talked to were sort of staggered by the amount of technology that they're seeing and which means there's room for consultants and implementers and educators and people like us to help organizations understand it. I didn't see anything there that was groundbreakingly new or market leading in the sense that it was going to propel the company forward at a rapid rate of speed compared to where it was in the past. SAP is a well run company, very well run company actually, I would say. And they're going through internal transformations inside just like everybody else, which are normal. And those of you that are SAP customers, unless there's some particular glitch in the system you have, I think you're going to be pretty happy with what they're doing.
[00:10:20] Okay, so that's the first thing for today, and if you want more information on that, just call us. Second thing is next week is Workday Rising. We have a lot of important things to talk about next week, which I will not pre announce. We've been doing a lot of work with Workday, and I'm a very big fan of Workday and have been for many years.
[00:10:39] And so stay tuned for that. We'll be at Workday Rising. If you're coming to Workday Rising, we're going to have a bunch of us there. You'll see us, and you'll see Galileo sprinkled around. The third thing is the architecture paper. So I talked about this in the podcast a week ago, but let me get into more detail just for a minute. The large Language Frontier Labs are really getting in trouble with alignment and safety. Regardless of what Trump or anybody else says, These systems, because they're not deterministic, can do things that you don't predict. And they're very good at cybersecurity and infiltration. They're very good at figuring out software problems. It means that if the system, for some reason, decides to do something that it thinks is the right thing, it could easily break a legal boundary. A perfect example is you're a pharmaceutical company. You have an AI doing research on genes or proteins or some form of science that's contributing to your products. And you tell the AI, do a literature research on such and such a protein and come back to me and tell me what all the major science research and competitors are doing, and then compare it to our research, which you know already, Mr. AI and then give me a recommendation as to where we should go based on what's going on in the outside world. And I would not be surprised if every single pharmaceutical company has agents doing stuff like that. I mean, we do that right now just to keep up with our own space. So what it's going to do, it's going to go out there and it's going to search the Internet, go through Google and everything. It's going to pick up a bunch of research papers, it's going to find all sorts of sources, and then it's going to find websites of your competitors, and it's going to read the white papers on those websites, and then if it's really ambitious, it might try to break into those websites. And maybe those websites are completely separate from the research websites where the real research is taking place. Or maybe they're not. You know, maybe some of these pharma companies don't have super duper high powered IT security yet, or maybe they're too small, maybe they're startups and they go in and the agent in its literature search goes out and finds some proprietary information from one of your competitors.
[00:12:59] And you didn't tell it what proprietary means. It doesn't know necessarily, unless you did a whole training session on that. And it sucks some of that information in and you get it and you build your product and do your thing and maybe you don't even know where this all came from. And because you didn't have time to check and you produce this thing and it comes to market and you find out that you have a huge legal liability. Now I don't think that's a very hard thing to conceive and that's just one of thousands of examples. So the Frontier Labs can't help you. They don't even know how to keep their agents aligned in their own world, which is way more complicated than a company. So forget about your little company. Your little one use case is tiny compared to the thousands of things they're dealing with that deal with nuclear attack and defense and missile systems and maybe hacking into security in the military and so forth. So how do you protect yourself? Now? If you believe what SAP workday servicenow Microsoft tell you, you're going to have this agent protector software that's going to monitor and harness your agent itself and prevent it from doing something it's not supposed to do. But that means that the agent protector software, and there's a bunch of flavors of them, they're in the white paper, has to know a lot about your competitors and what bad behavior looks like. So you got to spend some time. That's not going to be a simple project to prevent that from happening. But another maybe easier way to think about this problem is think back now about the human issue, which is the same issue in a human company with no AI. Any employee could do the same thing. They could get on a plane and just visit a competitor and just walk into a conference room and steal a bunch of stuff. But they know they're not supposed to do that because they're going to get fired or they're going to go to jail. So they're not going to do it. And the scientist or the researcher who's the expert on that protein also isn't going to do it because it's his career at stake or her. And they're not going to try it either because it would be their reputation for the rest of their life. But then, you know, Maybe if you're Enron, if you remember the story of Enron, you might have a few people who don't have clear ethical boundaries and they'll do some nefarious stuff and they actually will jump over the legal rules and they'll do some things they're not supposed to do. And you may or may not find out about it until later. And of course, that was the end of Enron. And you know, the, the death of Ken Lay and Andrew Fastow went to jail and so forth. By the way, I talked to Andrew Fastow about that. This is in the article I wrote. So you've got the same issue with agents that you do with people.
[00:15:37] How do you avoid, protect, prevent bad behavior from people who just don't quite understand the rules or don't want to obey the rules? Well, the way we do it in the human world is the same way we're going to have to do it in the agent world. We segregate information and we create specialization. I was at a presentation at SAP and somebody presented something that I completely disagreed with and they said that they were going, we were going to go from specialists to generalists. I don't think that's true at all. I think specialization in your job and your career career and your company and your roles and your HR organization and your org chart is more important than ever because specialists make up the value creation of your business. I don't know what's going on inside of Apple, but somebody over there knows how that glass works. They know how that folding thing works in the duo. They know how the software works, they know how the camera works. That person that's a human is a specialist in that technology and that engineering. And we want those specialists to be really, really great. We want them to be smart, we want them to learn all the time. We want them to look at lots and lots of use cases of their specialization. That's true in sales, that's true in marketing, that's true in finance, that's true in hr.
[00:16:54] And even if the AI could cover lots of grounds of lots of specialists, we don't want it to do that. We want the AIs to be organized into specializations also. Now, I won't go through the whole story in this podcast a little too complicated, but it's in the article and it's in our research. But in HR there are literally 151 or more, maybe more like 160 specializations that have to happen. We don't want one big massive super grok AI that's trying to do all of that, because if you created that, that thing would have to have access to every piece of information in your company and the number of exposure surfaces and risks would be massive. So what we're really building in the business world is a network of specialized agents. And so what I like to think of HR 2030 as is it's an operating system of agents. It isn't a bunch of individual agents working as assistants to one or two people doing one or two jobs. It's a highly interdependent, carefully architected network of agents that work together. And what you find when you start building an agent or whatever, even if you build your own personal agent, is you end up going down this exact path. Because the more details you put into the applications or the needs or the problem solving capabilities of any AI, the deeper and deeper you realize that domain is. And so what we want is we want sourcing agents, interviewing agents, assessment agents, recruiting agents that work on onboarding. All of those steps that form the basis of our HR capability model in HR at least, are in a sense agents. Now, five years from now, when you have a thousand agents in HR and 200 of them are corporate and the rest of them are individual and personal employees, PCs and phones, we need these things to work together and we need to design them so they work together.
[00:18:57] And that's exactly what we've done with HR 2030, is we've designed that network. And once you do that, you in some sense solve the safety problem. Because each agent has a limited boundary set of data and responsibilities. And it is told through its training or its pre prompt or its context layer what it is expected to do and what it is not expected to do. And if it wants to do something in another area, here's the agent it should talk to.
[00:19:29] Now I can't guarantee this is going to make everything 100% safe because these are still probabilistic systems. But then each specialist agent learns more and more and more about its domain and the company becomes infinitely smarter and more scalable for growth. Because I think what AI is all about is growth, not cost reduction or simply productivity. Productivity is in some sense a smaller component of growth. So that's what this multi architecture research is about. Now I did a lot of work on researching this. I read a lot of papers from a lot of IBM people and others that have worked on multi architectures. And when you look under the covers of a lot of sophisticated AI products, they are sub agents working together on what a soup, we call it a Super agent would do. This is all new for a lot of you. You're probably not thinking about it, but I think in our domains of hr, we're going to find that we're doing org design in a way. This is AI Org design. We're designing an infinitely scalable AI organization of agents that work together and do lots and lots of things. We have 220 use cases for this business. Use cases. And they all interoperate with each other in various forms depending on what your company's trying to accomplish or what the problem is you're dealing with at the moment. And it's going to take a few years. This isn't going to happen overnight. I mean, we're barely getting all these things working yet, and there's, you know, far too many vendors in the market. I was, it was funny. I was at the SAP meeting and I'm sitting down at the front and a woman came and sat next to me who I happened to work with at digital think 35 years ago. And it was so funny. We were just chatting around about the old days before the Internet, by the way. We didn't even have the Internet. It was barely, we barely knew what to call it. We were working on elearning at the time.
[00:21:22] And I, you know, as we were talking, I said to her, you know, Digital Think was a billion dollar market cap company and it went to zero.
[00:21:30] Many, many, many, many of the companies we worked with in those early days of the Internet are long, long gone. There. Nobody even knows who they are. Well, unfortunately, the same thing's going to happen now. A lot of the AI companies that seem very creative and ambitious and smart and really, really cool are going to just be. Poof, they're going to be gone. And we're in that early stage of what's called the J curve in this technology. The J Curve. I'm going to write an article about this too, so you guys can see it. I'll talk about it on Unleashed. The J Curve is an idea that in any new technology, for the first number of years, the value goes down, the productivity or opportunity gets worse because we don't really know how to use it yet. We see potential for it, but we haven't really learned how to use it. And then as the use cases grow, this goes for electricity. The barcode, the loom, the electric motor, all of the invent, the typewriter, the personal computer, all of these things. The first few years, we're fascinated with the technology. We don't know how to use it. We don't know how to maintain it, we don't know how to operate it, we don't know how to manage it. But we're sure there's something here.
[00:22:41] And so it kind of feels like a big waste of time and money. And then little by little, we figure it out and the curve goes up and up and up and up and up. And then we're at that sort of bottom half of the J coming up the other side slowly, and we're that basically, that's kind of our job here in our company, is try to help you guys move along that J a little more quickly. So that' our that paper is coming out tomorrow and you can get a quick overview of it, but you'll have to buy Galileo or join our membership to get your hands on the details. Or just call us up and we'll do some advisory work with you to help you understand it. So that's kind of my update for this week. Next week is going to be a very interesting week and lots of new things that I won't talk about yet. And we'll be at Workday Rising. So have a good week and maybe do one more podcast before the week is over. Bye for now.
