Episode Transcript
[00:00:00] Today I want to preview some really in depth information we're going to be publishing in January in our predictions report about AI in the enterprise. And I want to keep this fairly high level. But the concept is really important because it really establishes a framework for your personal activities around AI and also your company's investments. When ChatGPT first hit the market in November of 22, I believe we looked at it as an assistant. In other words, it's a personal chatbot with a lot of information behind it, which can do a lot of interesting things. Looking up data, creating documents, analyzing information, talking to you, and later creating videos, graphics, sounds, et cetera. And the concept and utility of the product was your assistant, and we called them assistants for about a year. The word assistant refers to you as a user and how it assists you. So of course we were all pretty excited about this because we're all doing something, work, recreation, design, analysis, whatever, and we would like personal assistance and to some degree, the co pilot. The word co pilot also is a word that has an assistant concept to it. You're the pilot, it's the co pilot. And in the context of business, where we have hundreds and thousands of people doing different things, each of those individual people could get an assistant to make their individual work better. We didn't think at the time about job redesign, org redesign, cross functional assistants, et cetera at all. And then around a year or so into the topic, we started to call them agents. And the idea of an agent was that it could do things on your behalf, not just assist you. A copilot is subservient to you. An agent could actually be your boss, or it could work independently from you, even though you would be the one that trained it or directed it. And the word agent became meaningful as the AI systems became more transactional. Because if you remember in the beginning of OpenAI, it couldn't really do anything other than process text, later graphics and stuff, but it wasn't a transactional system. But very quickly, within a year or so, the idea of agentic AI, which is what we called it in the beginning, or transactional applications as part of the AI infrastructure, became more popular. And even though the framework for this is relatively weak today, it's not clear how that works. And it's going to be very individual to each system. Right now we now think of these things as tools that can do business transactions, go off and do research, look things up, book flights, et cetera. There will eventually have to be some infrastructure behind that because there's a lot of glitches and gotchas when the AI is doing transactions. And the latest manifestation of that is that some of the agents have their own web browsers, both Perplexity and chatgpt to go out and do things on the web for you using the infrastructure that exists on the web to order food or book a flight, which creates all sorts of risk because that means the AI knows your password, knows your website history, et cetera. Anyway, that's an agent. And so inside of a business, if you're building a recruiting tool as an example, the the agent could have a conversation with the candidate and then book an interview, schedule it. In fact, it could accept the candidate's resume and stick it in the applicant tracking system, or score it, or tell the candidate that the job they're applying for isn't really a good fit, but why don't you apply for this other job, et cetera. So the agent manifestation of AI opened up a lot of opportunities for AI to do other things. Certainly created a new language for the sales and marketing of this stuff too. But as we talked about in our 2025 predictions, there's actually more than that, because an agent or an assistant that supports an individual and maybe is automated in some way. By the way, if you look at our four stage model, it's becoming more and more useful over time. Stage two of an assistant is it's an automated assistant. An automated assistant or agent does things to automate your work, like a macro in Excel, writes code, writes copy, analyzes data, et cetera. And lots of companies have built automations that basically consists of prompts. One of the consulting firms apparently built a prompt that's like 400 pages long to analyze the financial statement of its corporate clients. And we have lots of that in Galileo. By the way, the 400 or so prompts that we built are basically automations in a way to help you do things that we know HR people do, so you don't have to dream it up yourself. But stage three, or what's next? Is when the agent does more than one thing and it does lots of things, and it crosses the boundaries between you and the job you have and other people's jobs to become a multifunctional or multiprocess automated system. And that's where AI is going for sure. In other words, putting together horizontal workflows and applications that can do many things. For example, Beauty Genius, which is the AI system that l' Oreal uses to help you buy makeup and other beauty products, brings together dozens of applications and tools for selecting Beauty products, including information about your skin, your hair, your desired look and so forth to recommend things to you. It certainly is not an assistant. It's way beyond that, as is the automated claims agent the Travelers uses when a claims officer comes to look at your roof and determines how much to pay you for hail damage and so forth. And these stage three agents that we call them are going to be massively more useful than assistants because they can automate and integrate businessware workflows to stage four, where we have autonomy. Because if you have multiple steps of a process integrated into AI, you can start to monitor the process and using data, optimize the process as a whole. So when we recruit people from this school or this location or this background, they don't work out. But when we recruit people from this background and this school, they do work out would change the recruiting process in a way that an individual recruiter would never perhaps even know. And there's thousands of those applications. So in the four stage model, we went from individual assistance to automated assistance, to multiprocess agents, to autonomy. Now, as I've been explaining this to people over the last few weeks and writing about it, I thought of an example that I just want to run through with you that is the perfect example to understand this. And it's the autonomous vehicle. You know, I'm in my 70s, so I've been around since before power steering. And in the early days of my life driving my parents car car, we had a, believe it or not, we have. We had a Chevrolet Corvair, which was a really fun car. It did not have power steering or power brakes, but it was very fun to drive. So it was hard to steer, it was hard to park, and if you wanted to stop, you had to jam your foot on the brakes pretty hard. And if you only needed to stop fast, you had to jam your foot on the brakes with two feet. And what happened that was in the 60s and 70s and what happened was car companies built assistance for power steering and power brakes. The power steering assistance were miraculous at the time. And then they became smarter and we had automatic lane control where the power steering could use the cameras to keep you in the lane, eventually leading to automatic parking, which I'm not sure I ever trusted, but a lot of my cars have it. Then the brake, automatic. The power brake was connected to the cameras so that it would do collision detection and jam on the brakes on your behalf if you were about to hit something. And I actually had that happen to me once and it probably saved my wife and My life and et cetera. Those systems became more automated, more intelligent as independent assistants. And they were assistants. They didn't really work as multifunctional agents. But along comes Waymo, Google and later Elon Musk. I think Elon Musk was late to the party and he still is, and others. Zook is the new one. And they said, look, the purpose of a car is not to make a safe and comfortable experience for a driver. The purpose of a car, if we fall in love with the problem, is to get you from point A to point B safely, comfortably and quickly. So the fact that we have drivers is an artifact of the history of the car where we had human assistants or human beings motoring the car. If we assume the goal is to get there, maybe we take the steering wheel assistant, power steering, the collision detection assistant, the lane change assistant, the power brakes, and many, many more assistants, and we integrate them into an autonomous vehicle. And then what happens of course is we eliminate the job of the driver. Because we don't need a driver. In fact, we don't need a steering wheel, we don't need a driver's seat. And if you look at the zooc, we don't even need a front row. We could just create a living room with wheels that's comfortable, that has WI fi, good lighting, food, whatever else you want to put in there and get you from place to place. And that is a stage three, stage four agent. And as I talked about this more and more, and I've thought about it for now a year, this is a huge, groundbreaking idea that we have to think through when we implement AI because these integrated, what I call super agents, basically an autonomous vehicle is a super agent, not only a stage three and stage four agent. I would call them super agents. They are agents that bring together agents. They don't just coordinate agents like the MCP protocol. That's actually fairly primitive compared to this. The super agent of the self driving car is much more comprehensive in its understanding of what each of the agents is doing because it needs their information in real time. The super agent is a different type of agent than a collection of agents sharing information with each other. By the way, we're working through this right now with Galileo. We have Galileo running in the Microsoft Copilot, in NOW Assist and in SAP Joule. And in all of those three cases, the prototypes and the early versions of it are really great. But the experience that we're trying to make sure is really good is that while you're trying to do something in one of these Other systems. Galileo contributes and manages and supports what you're trying to do. And you don't have to mention it in a separate way to get the information that Galileo brings to the market. So this integrated super agent is not as easy to engineer as connecting agents together. Which is why the Tesla is not working out to be a very good self driving car. Because they didn't design it as a self driving car. They designed it as a car with a bunch of assistants that talk to each other. And I know that because I have friends that have Teslas and the self driving features are getting better, but they're really not the same as a Waymo or a zoo.
[00:10:55] So you might even argue that a super agent by design is a completely different agent than a collection of agents with MCP or A to A. And I think that's where this is going to go. Now the big thing to think about in this super agent era is what is the problem? In the case of self driving car, the problem is getting from point A to point B. In the case of a business it's going from developing a product to selling it, to collecting cash, to supporting the product to renewing it. There's all sorts of business workflows and we're going to have to decide each one of us how big of a workflow we're going to automate. For example, there are now agents in the market for sales development reps. They're terrible. What they're trying to do is automate the process of a salesperson who reaches out to try to get hold of you to talk to them. And what they do is they spew out random emails which are very irritating. I get a lot of them to try to get you interested in somebody's product. I think these vendors have missed the boat here because what they're doing is trying to automate a process that should not exist. There's no reason to send me a cold email and clutter up my mailbox, ruining the brand of your company to get my attention. It would be much more useful if you fell in love with the problem and analyze who I am, what I do, what my business is and what is my relationship to your company and sent me or delivered to me even by mail or another way. A very personalized solution to help me. I don't want to talk to a sales development rep. I don't have time. But these vendors aren't thinking that. So the sooner you get to thinking about this as a super agent problem, the faster you're going to break through and Create something that's really, really valuable. Now in HR we have hundreds of things that when you collect them together, don't need to exist.
[00:12:39] Let me tell you what we're doing here. So we've been studying HR for a long time and our capability model, which by the way, you can use it, look at it, get assessed on it in our website or through our membership or through Galileo, by the way, Galileo has the capability model in it has 94 or 95 capabilities of HR. And when you read through the 95 things, you're going to look at them and say, yep, I got to know that, yes, that is part of hr. Yes, I agree. So it sounds odd that there are so many things, but these are 94 capabilities. And the way we've designed HR over the many, many years is we've created job titles and specializations for each of the 94 things. DEI specialists, people analytics specialists, Sourcing specialists, Recruiting specialists, Assessment specialists, Leadership Development specialists on and each of those specialists were collected into centers of excellence to make them come together and serve the needs of the company, the needs of a manager, the needs of an employee, the needs of an executive, or the needs of the business as a whole. And so we ended up with service delivery designed HR teams that take orders in a sense and deliver on things. And as we've talked about for years and discussed in the systemic HR research, that model model has fallen by the wayside and HR is now turning into a problem solving, consulting, product oriented organization. A product orientation to HR means that you're essentially selling, quote unquote, or delivering integrated solutions with customers and understanding what customer needs are. But it's beyond that. It's actually a problem solving organization, not just a product organization. Product organizations, by the way, are not problem solving organizations. They sell a product, they leave it up to you or someone else to decide what problem to solve. Solve with the product. I know this, I've worked at a lot of product companies. They try to solve problems with their product, but they don't always know the problem that you're going to use the product for. Sometimes they do, sometimes they don't. That's the sort of the challenge of being a product organization anyway. So you take a look at the 94 things which we're doing and you say to yourself, what would be a bottoms up, start from scratch first principles way to deliver these 94 capabilities to a company using super agents. And that is the blueprint we're on. We will introduce it in early next year. But what it forces you to do is rethink why we do things the way we do them. And let me give you a simple example. One of the integrated, quote unquote agents that the vendors are all building is a performance management and goals agent. And what they're doing is they're saying, we have a process that includes annual goal setting or maybe quarterly, whatever it may be. We take the goals, we write them down, we give them some sort of a quantifiable metric. We integrate the goals into some measurement system so we know how we're going to keep track of people's progress against those goals. And we periodically have check ins and meetings and status reports on those goals. We document those check ins and statuses and then at the end of some period of time, we do a performance review, either aided by AI or aided by other information. We collect a lot of data about the employee's work, the meetings, the status, feedback from others. We give the employee feedback on their performance. We write a formal review in some documentary form. And then we give them a rating and a development plan and perhaps a succession plan or perhaps an exit plan. And later we use that process to determine at the end of the period or year their raise, their bonus or other benefits that we're giving out to employees. And we use that agent to compare people to each other and to, to find high performers. Now that's a very, very, very old fashioned simplistic way of thinking about that problem. That actually is an example of a car with a whole bunch of steering wheel assistance added to it. Why would you do it that way? If you had an automated super agent, I certainly wouldn't. We don't do it that way in our company at all. We don't have any of those steps and we're doing great.
[00:16:50] Maybe what you would do, and I'm just saying maybe, I think everybody will think about this differently. You would do it a little bit differently. You would think about the individual's job and role, which is written down by the way, and you would determine based on their job function. Sales. Let's suppose we look at sales, engineering, customer support, those three areas. And you would say in sales, we're going to measure your performance by the revenue you produce, the number of new deals you produce, and the renewal rate of your customers. For engineering, we're going to measure performance by the number of lines of code you write, or the number of widgets you design, or whatever the engineering tool may be, and the quality of your end result designs measured by some quality metric. And in customer service, we're going to measure your performance by the speed and volume of your customer cases and the quality of the reviews of your case management from customers or users. Okay, easy enough. Those are pieces of information you have in your company. They're sitting around somewhere a little bit hard to find, but they're there. Well, if you built a performance and goals and succession and development agent, you would probably do what Repling did is you would say let's look at that as an integrated super agent. Let's evaluate, let's take all that data, collect it, which is a little bit of a project in and of itself by the way, to determine where this data is going to come from and let's evaluate during the process of the year who's becoming a high performing salesperson and who's not. It's pretty easy to do that. You can look at the metrics coming out of Salesforce or HubSpot and then let's look at the behaviors, activities, even the recorded calls of those sales reps, salespeople and evaluate based on their work what they are doing that seems to be contributing statistically to those results. And let's compare that to the others and let's give the folks that are maybe slightly underperforming tips and advice on what to do based on what the high performers are doing. That's a great application of AI that is essentially the self driving car saying to the steering wheel or the steering system, there's a bunch of potholes in this road. It turns out if we steer to the left we can find an area where we won't be so bumpy or the road is slippery. We're going to slow down. Those signals come in to companies all the time, but we use them. Ditto in engineering. What are the engineers doing that are the high performing ones versus others? Ditto in customer service. Now this is going to take some time and these autonomous self driving little motors here are not going to work out of the box. Just like the Waymo didn't work perfectly for a while either, it took a while to train it because this closed loop super agent system is going to need data. It's going to need lots of data from lots of use cases to get better and better at what it's doing. But if you don't design it that way, all your performance and goal agent is is a paper pushing simplification workflow engine that pushes paper faster. I don't think most of us really like writing performance appraisals. Most of us don't like reading them either. And the only reason we do it is for record keeping purposes in case we need to let somebody go. And it's very problematic and very time consuming and very difficult. And by the way, it's not really that valuable. The most valuable part of the performance process is feedback and development support and performance performance enablement. And by the way, let's add to that L and D which should be part of that agent, that when we determine that the individual in any of those three areas is underperforming, maybe we could give them some tips from the SMEs authored by L and D or authored by the L and D agent on what they could do to be better at their jobs. And that information is not in L and D, it's in the high performers heads. So the agent, super agent could send an email to the top 10% salespeople and say, I would like you to record for 10 minutes how you closed the largest deal we've ever done in the company and what you learned from that and do the same thing for engineering, the same thing for customer support. And those little conversations could be aggregated together into developmental tips for the people that are performing at a lower rate. You see where my head is going. I don't think this is very hard to understand when you think about the concept of a self driving car. Now it's not as easy to do of course in a company as you think because companies are not set up this way. In some ways these super agents break every historic organization structure and SAP oracle workday business process that we have because they work across business functions. They don't just automate one business function. And the job families and functional areas we have in companies are very historic going back to the industrial age. But I'm telling you, the companies that do this are going to have groundbreaking results. The claims agents that is being used by the insurance industry now, which is able to pull together massive amounts of information about the claims process. And the claims process is extremely important because there's fraud in the claims process, there's estimates in the claims process, there's inflation of various products and services in the claims process. And all that has to be accommodated into an insurance company's business. These are multifunctional super agents. You have these processes in your company, I guarantee you do you have to break your head out of the current mode of thinking in HR to develop them. And so when we publish the framework or the blueprint, you'll be able to see at least our best efforts at understanding what these things look like. And we're not going to get them all right? But we're going to give you some really good ideas now. We're working with a large airline in the Middle east on this. We've been working with a large pharma on this. We're working with a couple of insurance companies on this. We're pretty good about that. We're pretty good at this because we've been thinking about it for a long time. We've been writing about systemic HR for three years. We understand the systemic nature of human capital. If you'd like to do this on your own, you certainly can, but please feel free to call us and we will help you do this. Galileo, by the way, is really, really good at this, believe it or not. You can ask Galileo questions like how would I better integrate my recruiting function based on the following org structure? And it will break the org down into roles and tasks and it will show you you what can be automated and integrated. So it is, as one of our clients said, one of the best thought partners you have. You can do that with us directly or you can do that with Galileo. So that's where we're going from assistants to agents to super agents. And this will all be documented in the predictions report coming out in about a month. And please call us if you'd like to understand more. Thanks. That's it for now.
