Episode Transcript
[00:00:00] Speaker A: Good morning, everybody. Today I'd like to start with some observations about the marketplace and the workforce. Then I would like to talk about HR 2030, just briefly. And then I want to talk about data security rules, skills and other technology architectural issues we have to think about in the whole world of AI. So the market for AI, in some sense commoditizing. And what I mean by that, and I explained this in an article I wrote on Substack, is that the large language models, the frontier models from Anthropic or OpenAI or Gemini, the models from Microsoft, the models from open source models, there's a lot of models. So if you're even slightly technical and you want to build an agent, you can download your own model and run it on your own computer. Not as easy as installing a piece of software would normally be, but it's getting very, very close. Which means that where we're moving to is not a world in the corporate world where you're going to build your business around Anthropic or OpenAI or Gemini or Copilot, that you're going to have a model or series of models in your company that are going to be unique to your company. Because the. The value of AI is not its functional capability at its core, it's the way it personalizes itself to your company. Your data, your rules, your processes, your people. So why would you make yourself dependent on one frontier vendor, one platform to do that when it's really your system, not their system? Now, that's very different from the old days of workday, SAP, Oracle and adp, where it was their system and you were essentially accommodating their business rules, their product managers, their features. So this is a very different world because as we flex our muscle and get into HR 2030 with more and more companies, what we're realizing is that many of these similar applications are different in different companies. The way you do recruiting, the way you do training, the way you do development, the way you do careers, succession, et cetera, is unique to you and you're going to want to change it over time. And so the AI infrastructure that we're building is much more personalized. Now, the reason I mentioned this in the terms of the market is when Anthropic and OpenAI go public, they're going to be massive valuations. But what they're really doing in some sense is sucking money out of the market for other things. And I'm not against, I don't have anything against them. They're very important research companies, but we don't want to spend more money on our. I mean, the fact that Apple is raising the prices on all of their products by 20% just because of memory prices, just because of storage prices, driven by data centers so that we can learn how to use AI and play with AI sort of leads me to believe that these giant profit centers that we believe are going to exist actually aren't going to exist. It's going to be much more of a commodity market. Now, by commodity, I don't mean it's low value. I mean the automobile market is kind of a commodity. You can switch cars in five minutes, really, if you want. If you want to take your car into a dealer and get a new one. And that's the way it's going to be with these. But you're not going to want to switch them that often because they're going to be so personalized to you. So from a market standpoint, the excitement and enamorment we have, that's a word with the frontier vendors. I think it's going to kind of go away. And I think what we saw from OpenAI this week was they're going to be putting ads at the bottom of their pages and they're going to make money with ads. So these things are going to look more and more like advertising platforms and less and less like corporate tools. And most of you are probably like me. You don't want to buy software from a company that sells ads for your corporate business. You want a company that focuses on enterprise applications. It's very interesting how this is evolving. And I think the reason this is all happening is so much money got poured into this so fast. There was no time for the market to mature. So these vendors are chasing revenue at a sort of superhuman or maybe extraordinarily reckless way. And you see the prices go up and down and there's going to be a, you know, cost per token for everything, and then there's going to be no cost, and then there's going to be embedded costs and so forth. The other thing I noticed this week that I think relevant to all of us is the AI that we buy for our companies and for our recruiting and our training and all that is also now embedded into glasses. I've been talking about this for a couple of years, or at least a year. Meta went crazy at the World cup advertising their glasses. I also saw it at a Valkyries game on tv. What that gets to is the embedding of AI into our personal lives and our personal devices. And there'll be lot of legislation and debate about this. Quite a few studies have been coming out in the last several weeks or months interviewing young people in their 20s who are really against AI. They don't like it, they're afraid of it, they don't want it, they think it's bad. Whether you agree with them or not, there's maybe a political pressure building that AI embedded into all these devices is not a good thing. I think it's interesting that Mark Zuckerberg has a tendency to go exactly into the most, maybe promiscuous that get into our personal lives more than anybody else. So we're going to kind of have a little bit of a replay of what happened in social media where this stuff on the consumer side gets a little bit out of hand. And at least in the United States, the administration has no interest in regulating anything. So I don't think they're going to do anything about it. So my point for us in the corporate world is we're going to probably be served by vendors that focus on corporate solutions and don't focus on this broader, more strange, emerging consumer market with glasses and things like that. And of course, the, the reason I mentioned the word promiscuous is that if you're, if you're wearing glasses, the AI is seeing everything you're seeing and it knows where you are. So it's basically. You've basically given up all of your personal privacy now to meta. I don't think most companies are going to like that. I do believe, though, if you look at our architecture, that we will have personal AI at work. No question about it. It'll be on your phone or your computer, and it'll be very good because it'll be watching your meetings and your schedules. It'll be a little bit like OpenClaw. Microsoft has a version of OpenClaw and they have their own thing called Scout that does this. So these personal AIs, which will probably be big next year, will be the way we interact with employees, and we'll be talking to the employee's AI, not to the employee directly in some cases. And that AI will then talk to the employee in the way they want here or decide what to do. So it's a very interesting world, and it's in some degree very dominated by the money that's been thrown at it and the excitement of people wanting to become billionaires overnight. And, you know, I think the capitalist system here in the United States is showing maybe its worst side at the moment, where everything that makes more money for a company is okay, as long as it makes money. And you know, when Apple decided to just raise prices by 20%, I guess my general attitude about Apple just changed a little bit. But I don't buy a lot of Apple stuff really. I've been a PC guy my whole life. Now let's talk about corporate stuff. So when we launched HR 2030, the purpose was to give you a direction and vision and architecture for all these new tools you're going to buy, not to tell you what to buy. And so we've been very meticulously working very hard actually on a framework which I call a reference architecture. And when you see it, those of you who do see it, you'll see that it is a very extensible reference architecture of agents and super agents that you could use to copy in your RFPs, to use for your own internal development, for your own planning, for your own budgeting and just for thinking through how you want to evolve your technology. I think it'll be very useful for vendors, I think it'll be very useful for consultants. We're not going to put it into the public domain. You're going to have to buy it from us or work with us to get it or to see it. And we're building a tool for it. So you'll be able to query it and use it to design different things for your HR technology architecture or your enterprise collaboration architecture, whatever you use it for. And as we get into it and we go through the functional requirements of all the different pieces of it, we're running into this enormous topic of the rules and the privacy settings and the security processes that take place in an agent architecture. And you know, I haven't been able to find much written on this. IBM's written a lot about it and there's a lot of blogs, but there seems to be very little complete solutions in the market yet. And I think there's a lot of that's because it's new. But there's also this issue that nobody's really thought through the intricacies of this idea of having a bunch of self defining technologies and tools and systems inside of the corporate walls. So we've been trying to figure out how to do it and we've made pretty, pretty significant progress. So I want to share a little bit of that with you. So the first thing that you have to kind of think about in an agent is an agent is sort of a little bit like a human being. It has skills. Skills are a technology, not a human skill. It can essentially incorporate a skill that you developed somewhere else. So you're going to have a shared library of corporate skills that the AI uses. Like there might be a skill for succession management, or a skill for goal setting, or a skill for wealth management, or portfolio analysis or risk analysis in an insurance company. And that skill would be reused or adopted by different agents. And the agents would acquire that skill with the click of a button, as opposed to humans take a long time to get a skill. And these agents, which are like people, have responsibilities. And what we do, what we don't want to do with an agent, is give it access to everything. Just like you wouldn't give a person access to every piece of data in your company, you would give them access to data that's relevant to their role and level. We need a way to manage access to data and then we need to monitor what this person is doing in some fashion. And then we need to decide from a workflow standpoint when this agent wants to do something, if it needs approval, who does the approval go to and what path does the approval take. And then we have to take the work or activity or intelligence or decision making of this agent and put it into a bigger frame, which we call a super agent. All of that is infrastructure that doesn't really exist. The big agent management platforms that are being developed, Agent365 from Microsoft, the agent control tower from ServiceNow, the agent system of record from Workday, I'm sure SAP has one of these, I know Google has one of these, are intended to be this control space or control plane, getting back to this weird language that's being created to make these decisions. But it can't, because that piece of software is new and it doesn't know what rules you've already built in all of your other systems. Every enterprise piece of software you have, from Salesforce to Cornerstone to Workday to Oracle to Expensify or whatever it may be, has rules in it also, because that's what these things do. So the implementation of an agent requires the agent control tool, which registers the agent and monitors it and might give it access to certain data, but it also has to talk to these other systems because they're not going away. You're not going to start from scratch unless you're a very small company. So you need someplace to put this. And back in the days of my IBM mainframe era, there was a tool called RACF R, a C F Resource Access Control Framework or something like that. And it was the ultimate end to end computing security thing. And every password and access and data. A decision that had to be made went through racf, but that was the days when everybody had mainframes and everything was on IBM that never would have worked in an open systems world with Unix and PCs and everything. So there's going to be a whole bunch of opportunities here for vendors to build this control and management framework. The more I've been thinking about it and looking at it and talking to everybody about it here, the more I realize this is not going to happen quickly or easily. Your IT teams are talking about this and working on this right now. And I think this is going to be a very customized situation.
So to avoid waiting for some vendor thing that comes along that may, you know, do this for you, the most important thing I think to do is to just think about it logically when you design an agent, and most agents, by the way, are being designed by individual people for work that they do. Someone, you or someone else needs to decide how's this agent going to be managed and what are the privileges and rules around this agent. That could end up being a job, that could end up being a lot of jobs. If it's an agent for pay equity analysis, that means that the agent, it might have very sophisticated access to third party data to evaluate pay across the market and by geography and by role and by job and by tenure and all these other things. But that means it's going to have to touch your compensation database which is in some HCM system which has the privileges around it. And so that pay equity analysis agent is going to have to use the security in that HCM platform, if you like that security system. So there's going to be a policy agent. We're going to call it a rules and policy agent I think is what we're going to call it, that handles these decisions for you. I think the agent system of record and the control tower and the agent365 and tools like that might end up being that agent. Hopefully they develop that kind of functionality. I think that's where they're going. But regardless of who develops it, you're going to have to do it. And the reason you're going to do it is because you don't want to code this into every agent. It would be as if every human being in your company had to know every rule for access to everything. Most of us have no idea what the rules are. We find out what the rules are when we run into them because the rules were set by somebody else. So, so that's a huge part of the architecture Another part of the architecture we talked a lot about in this webinar did last week is the orchestration layer. When an agent gets a query. There are different, by the way, there are different types of agents. There's agents that do things, there's agents that monitor things, there's agents that control things, and there's agents that collect data for other agents. So there's different kinds of flavors of these things. But the ones that are doing things need a framework for decision making. It's not always linear. If this, then that. If you're a recruiting agent and you're evaluating a series of AI digital interview scripts or data, the digital interviewing agent, or whatever you call it, is ranking candidates using some criteria. What is that criteria? Who sets that criteria? How do we evaluate that criteria? How do we bypass that criteria? There's a whole bunch of decision making that goes on in hr, maybe more than anything else. This is why HR is so more complicated than finance. In finance, there's a right and a wrong way to do a lot of things. Most things in hr, there's a lot of judgment everywhere. So we're going to need an intelligence layer that helps you make those decisions and also explains to the agent or the user what the options are. By the way, this is where Galileo is going. We now have Galileo running against enterprise data in SuccessFactors. We're doing some implementations of this right now and in Workday and in our own prototype database where you can talk to Galileo about issues at a high level, at sort of a business level. And it will give you and talk to you and give you options of what to do based on real data. The demo that I showed last week, for example, was, and I'd be happy to show it to anybody who wants to see it, is a real system, a real HCM system, where you go to Galileo and you say, I need a successor for our VP of AI engineering who's available and who do you recommend? And it goes into the system, it looks at people's job titles, levels, their career histories, their training histories, whatever Data is in SuccessFactors, it gives you a list of candidates ranked by what it thinks are the highest rankings, and then it shows you the characteristics and backgrounds of each one. It lets you select one and then tells you what is strong and weak about that candidate. And after you select a candidate, it creates a development plan for that candidate based on data and another part of SuccessFactors, and sends a message on teams to that person offering them to start their development plan because they've been Selected for a new job. That's a super agent that we built around Galileo. Galileo made a lot of decisions in concert with successfactors to do that. So that's another big layer of the architecture is where is this context going to come from for the decisions that you're making? You might argue that each of the context related decisions is in each agent itself. Like the paradox agent will decide who to hire and will tweak Paradox and play with paradox and somebody will manage paradox to do that. But there might be a global national decision where we say because of something that's happened in the company, we're not going to hire people from this experience or without this credential, or without this level of depth or whatever the decision might be made. And so this context layer that I talked a lot about at that webinar is really another level of the architecture we have to think of. The third one I want to just mention on this podcast is the data access layer layer or the data context layer. In all of these agents that we build, you've got to get data from different places. If you go build yourself then, you know, just go trying to build yourself a personal agent. You're going to experience all these things. Try to build a personal agent for, I don't know, skills analysis of some sales team. And you want to compare tenure to sales quota attainment or something. Or you know, maybe you want to compare tenure, job history and educational history and degree to their performance. Kind of a big spreadsheet shouldn't be that hard to do. You could kind of go to one of your favorite tools and sort of speak English to it and it might be able to build that. And then if it works, you could turn it into an agent. Well, where's it going to get the data from? You're going to have to connect it to Salesforce. You have to make sure it gets the right data out of Salesforce. Maybe you, maybe you don't have privileges
[00:19:30] Speaker B: to get that data.
[00:19:31] Speaker A: Then you're going to have the same problem with the skills data. Then you have the same problem with the history, job history data. And you as an analyst or a decision maker are now playing data scientists going out there trying to find this stuff. So we need a layer of software. Globe calls this Lumra and there's a fabric from ServiceNow that does this. Databricks is getting into this workday, is planning on building more tools for this that connect together heterogeneous systems to give you context around these agent related questions. So you don't have to personally hire a data analyst or a data architect to figure out where all the data is. Because the agents, again, you can't code everything into every agent. It's just like humans. Every human being in the company doesn't know everything about the company. They just don't. They never will. So we have to have specialization and we have to have these layers in this architecture. So I guess the purpose of this podcast is not to educate you on everything here, but to give you a sense of maybe respect and perspective on where this is going. Now, most of the vendors who build things are going to deal with all of these issues and try to build tools that make this easy. And that's coming, of course, and you need to evaluate vendors along these lines. And we're going to have a rules framework for you to look at to help. But every company is different. So in. Despite the fact that AI is very highly interoperable because it can generate API calls pretty easily, this is going to be a very tricky part of your solutions, and you have to work with it on this. If you're out there buying standalone AI tools today, you probably won't run into this stuff, but you will over time, and we just want to educate you on it. The final thing I'm writing a lot about, and I'm going to be talking more and more about in the big webinar that we're doing in a couple of weeks here, is why are we here? What are the problems we're trying to solve? I'm very convinced that the purpose of AI is not to eliminate jobs and it's not to cut costs.
I know that's been a huge theme for the last couple of years, but that's not where we're going. Even if you cut the whole HR department down to zero, you're not going to save that much money, because the work that HR does is going to get done by somebody. And if it gets done by the sales managers or the finance managers or the IT managers, that means they're not doing sales, finance and it.
So the cost of HR is not the issue. Yes, it's bureaucratic. Yes, it might not be perfect. Yes, some people in HR might not be the perfect roles, but this stuff we do is there to support the business. On the flip side, the reason we're here and the potential value is for what I call dynamic enablement for growth. If we do our jobs well and we use the right tools, the company will grow faster, we'll hire more people, we'll hire better people. We'll have better skills, we'll have better alignment, we'll be able to adjust faster, we'll be more dynamic, we'll have better leadership and so forth. You guys know this, but somehow the AI industry got a little bit sidelined into this let's cut costs thing. I'm not sure how that happened because that certainly wasn't the vision of all the AI architects in the beginning. But maybe it's because the price and the cost of AI is so high. So, you know, the reason I'm even bringing up all these architectural things, if. If all this technology did was cut down the size of your call center, fine, you know, buy it, you know, go out and do it. But it's just much more complex and interesting than that. Even if you do use AI to reduce the size of your call center, you now have an intelligence system about what employees issues are. And that's much more valuable than shrinking the size of your call center. I mean, yes, shrinking the size of your call center and answering calls more quickly is great. That's important. And that really makes a big difference in productivity and growth. But getting to know what the issues are that you're employees are facing in their jobs or roles or in the business, that's way more valuable to me than cutting the size of the call center. So the reason, just to remind you of why we're doing all this work is, and I'm pretty clear this is the way I see it, at least, that we're going to be in a new world where we're really business enablers. Now. We're not administrators, compliance officers, or data analysts.
[00:23:53] Speaker B: We are in the dynamic enablement for growth business.
Okay. All right, that's it for today.
Sign up for the HR 2030 blueprint. To learn more, check out our new HR 2030 course in Galileo. Learn.
And sign up for the Global HR Excellence certification, which starts in September.
And please call us if you'd like help applying any of these important principles to your company's personal situation. Bye for now. Have a great July 4th holiday and talk to you guys again next week. Bye.
