Episode Transcript
[00:00:00] Good morning everyone. I'm podcasting today from Singapore, just spent a week over here in India and Singapore meeting with a lot of interesting companies.
[00:00:10] And there's a big topic I want to discuss this morning.
[00:00:14] Next week we will be introducing our largest ever study of the learning and development market and what's going on with AI and how AI is transforming L and D. Stay tuned for that. It's huge. It's a huge topic. It's a $400 billion industry.
[00:00:30] It affects the content industry, it affects the training industry, it affects your tech, your LMSs, the way you distribute HR into the, into the business, many, many other things. So that'll be next Wednesday, but today, based on the meetings I've been having, and one of them yesterday was with one of the world's largest banks, I want to talk about the confusion in AI architectures in the corporate space. And I'm going to talk about two major topics, all of which are covered in the imperatives research, which is all available in Galileo. The first is the architecture of agents versus the architecture of HR and business technology in the past. And the second is how you build stuff. So if you've been following the stock market, and I think a lot of you may have been, the software stocks are plummeting. Salesforce, Workday, Adobe, all of the application software companies stocks are down 20, 30, 35%. And the reason for that is the financial community knows that the IT technology world has shifted to agents, has shifted to token based business models. In other words, selling software by compute, not necessarily by license and redevelopment or re engineering or reimagining all of these corporate applications we have in, in every possible area around the agent workflows. And that means that the erp, hcm, CRM systems that we've been building and buying for the last three decades, it's been, it's been more than three decades really are going to get unraveled fairly quickly, not overnight. And this idea of a large system of record and paying tens to twenties to thirties of millions of dollars per year for that large system of record is quickly going away. Now it'll happen faster in small companies than in big companies. And you know, it may take decades for this to transform, just like it took decades for mainframes to move to client server and for client server to move to the web. But it did happen. And so the, these big companies that we rely on in hr, the, you know, I'm thinking about workday, Oracle, SAP, edp, Dayforce, et cetera are going to be, I wouldn't say upended but eaten away little by little by new tools and new applications which are sold by AI centric vendors but also built by you. Because it is very easy to build applications with AI compared to what it used to take to build applications before now. What does this mean in terms of architecture? The traditional approach to software in HR and everywhere else is a sort of monolithic process based design where if you look at something like onboarding and global onboarding is sort of a good example because it's an example of a process that's very multifaceted and very unique to each company. You can't really buy an off the shelf onboarding system because, you know, they change depending on the company and the business you're in. The other company I left out Yesterday earlier was ServiceNow. Even ServiceNow is getting hit by this, although they're in a different situation. And so if you wanted to build an onboarding application, you would have to build something that talked to the IT department for provisioning and setting up the person's tech, laptop, phone, et cetera. You'd have to have connections to the card readers in the various buildings which by the way are separate from each other, so that the person can get into the facilities they're allowed to get into. Then you would have to give them access to payroll and set up their payroll, their benefits, decide what their tax withholding is. They'd have to put all that in there. Then they would get their initial training or onboarding for business processes and compliance to make sure that they're ready for the job and ready for the industry and ready for the company. Then there would be onboarding of the work from the manager, perhaps an onboarding, introduction to the corporate culture, meeting people, connecting to people in the organization, understanding how the organization works. Then there would be onboarding to the actual work experience, the tools in the part of the company that the person works, setting their goals, et cetera. And of course that workflow continues because then there's the first three months check in, then the first six months check in, then the first performance review and all those other things. So this, this thing is a complex multi process, multi system workflow. And so what we would do is we'd go to IT and we'd say, here's the specs of what we want. We'd go back and forth with all of the, all of the iterations of this. We'd have to go do a lot of, you know, discovery to make sure we don't miss anything. They would start building it and there'd be a prototype, and maybe after, you know, three to six months, they'd roll out an early version, we'd start using it, we'd see what works, we'd see what doesn't work and so forth. And this monolithic application would take on a life of its own. It would need maintenance, it would need updating, because the company's reorganizing all the time. And there's new business processes and new facilities and new tools in IT and new tools in HR and new payroll things, new benefits, all that. And we now have a big piece of code to maintain. And we own it because we built it, and if we bought it from a third party, we customize it, we configured it, we added a bunch of stuff to it, and hopefully it works well. And then new scenarios come along, like we, you know, we have contractors that need to have facilities access, but they don't want all the other stuff, or we have special guests that come from the political environment that need safety precautions or privacy that other people don't need. So we need sort of a version for that. And we need a version for executives that come in and have different privileges. So this thing that sounded kind of easy in the beginning turns into an app or a system and, you know, it's a lot of work to keep it up to date. And as much as the vendor you bought this thing from might try, they can't possibly, you know, accommodate all these other features. So how do you deal with that in the world of agents? In the world of agents, each of those processes are owned by an agent. In each of those, the agent that worries about your IT provisioning, the agent that worries about your IT security, the agent that worries about your badge reader, the agent that worries about your payroll and benefits administration, the agent that worries about your culture and early introduction to the company and initial training, et cetera. Each of those agents is intelligent, just like all the other agents, and is probably owned by a subject matter expert in that area whose job it is to keep up on all of the topics and business changes and processes in that area. And then we have a super agent, which we describe in the imperative research, that strings these agents together into a workflow so that the user experiences the functionality of all of these agents in an up to date way. And the super agent knows who this person is and what their characteristics are so they get the right experience. This architecture is so much more flexible, so much more dynamic, so much more approachable than the traditional monolithic software architecture. This is where the world is going we have a client already in the defense sector that has 60 agents. They've already built 60, and they are now using tools from companies like ServiceNow, and there's others to manage the agents and keep track of what they're doing so they can coordinate the agents with each other. And that kind of an architecture begs the question of what is the role of the erp? Now, the ERP or ACM platform in our case, does have agents within it, but what those agents are mostly doing is automating parts of the workflow that have already been embedded in the erp. And the benefit of AI is not just to make your current workflows a little bit faster, but to move to a more automated workflow and a more autonomous workflow and to more interconnect these applications between the various different applications or systems in the company to create a more systemic solution. And the analogy that I discuss in all of these meetings is the driverless car. We have lots and lots of improvements to the driver experience. Power steering, power brakes, automatic lane control, collision, detective detection, automatic parking, et cetera. Those are the equivalent of assistants or agents to automate the work that we do today. But an autonomous car automates the passenger experience, not the driver experience, and focuses on the outcome, not the work that is currently being done. And so we really have to think beyond AI automating the work we're doing now and using AI to automate the process or the outcome we're trying to achieve and see what parts of the work we're doing now we don't need to do anymore. And in the case of the driverless car, the autonomous vehicle is 100 times safer than a driver led vehicle, regardless of how many assistants the driver has, because the autonomous vehicle has access to information and predictable autonomy in a way that the driver does not have and will never have. So as we built these new architectures, we have the luxury of saying to ourselves, why are we even doing this process? X maybe we, if we had it automated, we wouldn't need it then we wouldn't need these people doing this step and handling these kinds of tasks, because the system would do that for us. So this new architecture, built around agents, which is very new, of course, and a lot of people are still trying to figure out how they want to do it, is very focused on outcomes, not existing process or job automation. And that's the reason companies have been so frustrated by implementing assistance and not seeing radical improvements in productivity. Because if we don't shift the way and the things that we do and More integrate and automate these business processes in a horizontal way across the company, not vertical within one functional domain. We won't see the massive ROI of AI. And that, to me, is the potential value of AI. Now, the second part of this confusion we talk about, and we went through this with this bank, is the fact that you're not going to buy all of this. You could, but you're probably not going to, because AI is something that you can build and you don't have to be a software engineer or a computer scientist or have a PhD in AI to build stuff. We built Galileo with the help of our friends at sana, but for the most part ourselves with a very small amount of engineering support, because it's all about understanding the work, the workflows, the data, the information, and then tuning and continuously labeling and iterating to improve this system. You can literally describe a workflow or a process that you want an agent to do in English or any whatever language you speak, and the system will take that explanation and it will implement it either in software or in its own code. I've given this example a bunch of times, but I'll repeat it again. Last January, yes, last month I was out of town on vacation and I sat down with Claude code embedded within Galileo and I gave it a very simple prompt to build me a performance management application to allow people to check in every week, check in with their managers, evaluate their goals on a 1 to 5 scale, compare their goals against other people, and so forth. I just described it in English. It built the application for me in about five minutes. I ran it, there was a glitch. I took a screenshot of the glitch, I sent it back to the AI, the AI fixed it, and we built an application. Now, it's not perfect, it's not going to work in an enterprise setting. There's all sorts of if and buts and thens and thens and so forth that still have to be worked out, but that can be done in English. So these agents are not necessarily going to be things you have to go buy from vendors, although you might, or put them into the IT queue and wait for IT to build them for you. Because it, you can't build them without knowing everything that you know about how you want the process to work. So a lot of this new architecture of software is going to be built by you, by teams in your function, by technical people that, you know, sort of build apps within HR or within IT dedicated to hr. And you'll have autonomy over managing and controlling These apps over time, these agents rather. Now the question you sort of ask yourself if you sort of think this through is not only what do we build and how do we build it and where do we start and what are the most important super agents? And the dialogue there should not be around, you know, what will be the most interesting one to build, but rather where can we add the most value to the business? Where are things that can improve outcomes, faster time to hire, greater leadership pipeline, greater productivity in the sales organization, better experience for customers and so forth, as opposed to how can we clean up or act in HR? We know from our AI blueprint that about 30 to 40% of the current work you do in HR is going to go away. But it's up to you to decide where you focus that on and where you see the biggest business benefit. And then you're going to scratch your head and you're going to say, well, what about SAP or Work or Oracle or Workday or whatever the big, you know, vendor we have? What role does it play? And now you get into the vexing problem that those vendors are facing, which is where will they fit in this agentic architecture? Now each of them have different strategies to some degree. SAP has been focusing on Joule and componentized agents within the SAP suite. Workday acquired Sana, the company that we work with, building a front end to Workday or a front door to Workday, which you'll hear a lot more about from them in the coming months to build, and a set of application development tools to build apps around Workday that they're offering or will offer, including all sorts of intricate connectivity stuff that's pretty cool. Oracle is doing the same thing you know all about if you follow Salesforce, the Salesforce agent force, they call it Agent force, they're doing. ServiceNow has built a whole workflow based set of tools for managing agents and agent control tower. And we actually now have Galileo working within ServiceNow so you can easily use the ServiceNow layer as a digital HR business partner. They're doing that inside of ServiceNow and you can do that. Get access to all of our intelligence within your agent applications that you either build or offer directly. And I'm sure all of the other major. I talked a lot about UKG a couple of weeks ago. They've created all sorts of amazing agents actually for frontline work that are very unique and very interesting. So, you know, once you figure out where you want to focus your attention, then what you're going to want to do is look at the vendors you do business with and say, given what kind of super agent experience we want to deliver, what can you do to help? Are you going to build something in a component wise format that we can use? Do we register the code that we built with your platform to get access to the data? How do we get access to the data? Do we need the data? Some of these agents will use data in the core HR system, some of them will not. Some of them will need new data sources. And how do we leverage the security architecture privileges and the workflows that we have without being dependent on the workflows we have? Because the agents are going to operate in a much more autonomous, integrated workflow fashion than the ERP HCM products could in the past. It's, it sounds very confusing, but it's really not. Because when we went through this with this bank and we actually had two banks we met with over here, what we essentially concluded, you know, and this is going to end up being a workshop type experience, is let's spend some time prioritizing what are the super agent workflows or experiences that are the most valuable to the company, not to hr, based on what the company is trying to do, maybe it's opening a new line of business or moving into a new country or something like that. And then who are the domain experts in these areas that can work with us to specify in English language what are the things we need to accommodate in these agents and then finally, what toolset are we going to use? Now the way we see the super agent architecture is there will be multiple platforms involved. There will be the agent platform that all the employees use which will be, you know, something that it will select, it's kind of a course standard. And then there will be the agent platform or technologies that we use to build and deploy applications that talk to the employee facing agent. In the case of Galileo, for example, we have customers that have the copilot or other front end agents and they commit, they can talk to Galileo or they can talk to Galileo, learn to get information about HR or to ask questions about benchmarks, or to get coaching on leadership or to take a course or get a video or a tutoring session on something from our platform through their platform. So it's going to be a multi tiered architecture, but it won't, you don't have to buy one thing now, you know, several of the people we talked to and this is, you know, very common, said, well, what about all these vendors that are selling great stuff? I think you have to sort of take that with a grain of salt that there's going to be hundreds of agent vendors already are. You know, they're, they're so easy to build that there's open source agents out on, you know, the ChatGPT agent area already. I don't know if any of them are any good, but they call them GPTs.
[00:18:32] So what you want to do is it's okay to talk to all those guys because you'll get ideas and some of them have really creative ideas, but you want to start thinking about what you need first and then show them the super agent architecture that you're trying to build and let them fit into your needs instead of you fitting into their needs. We are happy to show you how to do this. The AI blueprint, what we've been developing is available to our clients. We're probably not going to put it into the public domain for some time because we're working on it with a bunch of companies now. But this is a very fascinating, empowering, important initiative in the world of AI. We'll keep you up to date on everything that's going on over the next six months or a year. Of course, we're doing a lot more work directly with Workday because of our SANA relationship with Workday. And so we'll be able to tell you more about what workday is doing and SAP and all the others.
[00:19:22] And it's a very exciting period of time. Stay tuned next week for the launch of our sixth massive Learning and Development Definitive Guide. It will be available to clients, it'll be available in Galileo, and it will show you one of your potential transformation projects in your company is to really reinvent learning and development in an extremely dynamic and positive way. So stay tuned for that. That's it for now, everybody. Have a great weekend and we'll talk more next week. Bye for now.
