The Hybrid Orchestration Control Plane: Powering the Agentic Era

Hello everyone. Welcome to Stonebranch UserVerse Online. We're excited to have everyone here today joining us for our four-part series as we cover topics around the future of Enterprise Automation and how AI is transforming the way organizations orchestrate their business processes. Today's our second session, and we're going to be taking a deep dive into hybrid orchestration control plane, powering the agentic era. My name is Nicole Crawford. I'm with Stonebranch and I will be your moderator today. Before we get started, I do want to go over a few quick housekeeping items. We want this session to be interactive. So on the right-hand side of your dashboard, you'll see a little tab with Q& A. Please enter any questions that you have. We'll try and do our very best to get to those questions during the session, but if some reason we run out of time, we will definitely get back to you after the event. In addition, on the right-hand side, you'll see a resource tab. There'll be a link in there to click to get more information about the topics that we're discussing today. And so with that, I would like to do some introductions to our speaker and we'll go ahead and get started. So with us today we have Scott Davis. He's our chief marketing officer at Stonebranch. We also have Peter Baljet, who's our chief technology officer at Stonebranch. And we also have joining us today is Dan Twing, who's the president and COO of EMA. So with that, I'm going to hand it over to Scott and we'll go ahead and get started. Hey everybody, just a quick introduction for Dan. So I've worked with Dan for, I guess seven years now, Dan. It's something like that, right? Mm-hmm. So ever since I joined Stonebranch. And Dan's always had a very strong pulse or finger on the pulse of the workload automation market. As we moved into the AI and agentic market, this agentic era that we're talking about, I got to talking to Dan and I realized very quickly that he really gets this, he understands the agentic orchestration, the need for hybrid orchestration, and more importantly, the need for a control plane. So we invited him to jump on with us. Dan, he's obviously an analyst who has a lot of context from speaking with a number of different vendors like us, but also vendors outside of the traditional SOAP area. And part of that really comes back to talking to people about the AI world and all the different AI tools out there. So I think he's going to add a lot of good context. And with that said, I'm going to pass it over to Dan. Dan, over to you. Great. Thanks, Scott. Well, I'm happy to be here. Yeah, Scott was posting a lot of interesting things. We were both noodling as this was happening and I was watching and it's like we need to cross streams here. And you added a lot to what I was thinking for sure. Yeah, I'm anxious to dig in. Yeah, that's good. All right. So Dan, just signal you want me to move the slides. I'll start here and we'll go into your first slide. So I sort of start with the idea I hear a lot, right? AI is going to displace workload automation or SOAP, whatever you want to call it. I've certainly talked to some investors and so on, and it's just not how I see things shaping up. There's this sort of replacement assumption that AI arrives and the enterprise can just discard old technology and just start running on AI. And of course, if you've taken any time to look at this, there's a lot we can do better with it, but it's going to take some efforts and some best practices to be worked out. Because those old layers represent is systems that still run the business and the foundations newer technologies really depend on. I used this graphic in a presentation in Europe earlier this year and a number of people just went, "Yeah, that's my data center." They got it. We see all the shiny stuff up top, but we all know there's a whole lot of things we're standing on. So given that background, so does our trusty old workload automation survive AI? I've heard that question a few times. Let's do a quick history lesson first, Scott, if you can jump ahead here. So now I'm going to make a statement here, and of course it doesn't hold completely, but we don't really replace technology in the enterprise. I mean, eventually we do, ultimately do. But, more we collect it and we layer it one after the other because things are still useful. So every major technology wave that arrives in previous generations, it continues to produce value for a long period of time. I mean, we've heard of the death of the mainframe since I started my career, and certainly there was a big part of the market modernized and moved during the COVID era, but a lot of the market decided to still stay there and do other things. And so we have all these layers. In fact, there are a lot of web applications that are starting to show age as well as things have changed. So modern enterprises add technology a whole lot faster than they retire it. And modernization will make changes with the technology, but it doesn't eliminate this coordination function. A lot of what I see SOAPs or WLAs doing, in many ways, it's part of what makes the enterprise backwardly compatible. So you can't just walk away from it, certainly. And Scott, if you'll go from there. I'll change slides in just a moment, Dan. I just wanted to make a comment. When I first saw this slide, I was laughing at myself because our chief product officer, Gwyn, who a lot of people on the call saw on Tuesday, he's told me a number of times, every time we do a conversion with one of our new customers, it's like an archeological dig. A lot of times, even in those conversions, the people we're working with don't even know everything that their automation system is actually automating. They could be touching dozens of systems and they don't fully... The people that set it up are no longer there, those sorts of things. So there's nothing more true than what you're showing on this slide. And that's a big part of what the hybrid IT automation market was about, connecting the mainframe to the cloud. And obviously this agentic era is taking us the next step forward. Okay, let's go to the next slide here, hopefully. All right, there we go. I know this one can be a bit of an eye chart. You can focus mostly on the middle part here, the picture part, the words will back things up. But when I look at this market, I see something different. I've been doing IT a long, long time. I started my career on automated telemachine settlement reports and file transfers back then didn't work real well. So lots of 2: 00 AM calls to fix things from the folks that were watching the workload automation tools. And when alerts happened, I got the call. Many of us have sat in different roles like that, but over the years I started to envision, it's really more like this helix. I mean, we started in the '60s and '70s with coordinating batch work, and the premise a lot of times when I talk, especially with investors who don't really understand the market is that, well, that's old tech and it doesn't do anything, but with the modern stuff, but we wouldn't have the cloud era we have today without workload. I mean, it picked up so many waves along the way, and these are just three or four of the biggest waves, but it's constantly, it has adapted to new tech because what this function really is coordinating across many different domains, that whole end-to-end view. There's a lot of automation in an enterprise that happens locally in a domain. The stuff we tend to get involved with is the stuff that is either crossing a lot of parts of the organization or it's very high volume or very important to the business, and then it needs more centralized or enterprise class control. And so that has really been an enduring problem, and because of the overlap of how we update technology, that tends to... It takes us less time to absorb that tech into this class of software than it does to displace everything this software already does. And that pattern has repeated for 40 or 50 years, and I just don't see AI knocking it. I don't think this wave is going to be any different. It's a radical change to be sure, but the core function of this software remains important. And if you'll go forward, Scott. So if you think about is all automation equal, defining different types of automation, I'm using terms, and I only created these terms trying to think about what's going to happen as AI comes forward, because the domains are all going to get smarter. And by domains I mean service management or network management even within IT ops or sales or HR. I mean, all these different domains, the whole ERP systems, they all are going to get smarter with AI. And so we have to be aware that that's going to happen, but yet there's a different layer that when I use the term Enterprise Automation, I mean this is what has to be more coordinated and have some centralized view so that we have that end-to-end. So from there, Scott, we go ahead. Yeah. So when I started to think about how do you explain to folks why this is so sticky? And this coordination thing to me, the domains are almost like tiles in a floor. Think about IT as a factory floor and the tiles in that floor are things like the data or observability, cloud, ERP. We have all these areas and they're all getting smarter. But what WLA or SOAPs have been, in my opinion, is we are more like the grout. We connect and hold everything together and the system works because individual tiles are weak without the grout. So that's really the role that WLA has played, and that coordination problem just doesn't go away with new tech, and I don't believe it's going to go away with AI either. Scott, I know... Yeah, there you go. Thank you. So I did some research back, it's been since 2019 where I asked really think about all the automation in the enterprise, everything inside of apps, everything we have to do, and how much of that is passing through your SOAP or WLA? The answers varied from about 15% to upwards of 70% depending on industry, big banking, very high use of this sort of thing, some manufacturing, not quite as much. But on average, it's about 45% of the automation is going through these centrally controlled systems for the reasons of, again, cross-domain coordination or it's very critical to the business, it's very high volume, there's SLAs connected to it, or because of technology continuity, even. So yeah, if we move forward from here, Scott, you can see that this is why that crossing work just doesn't go away. So does AI change that? Well, again, I see the domains getting smarter. So our orchestrator in the middle here used to just orchestrate early on, it would be a lot of applications, it was sort of end of day work, and then as the cloud and more things came up that were more real time, we started to coordinate things intraday. Batch became a smaller thing. It might be in minutes instead of days or hours, and we start supporting online apps, and there's a whole lot of things that you're aware of that workload was doing. But now those things swirling around us are getting smarter too, and they have their own orchestration going on. So we have to be aware of that. And when I talk about a control plane, I don't believe a hundred percent of the work is ever going to flow through here. It never has, but it's that core group for most enterprises around 45% that needs that extra lift. So as every domain becomes smarter, I think it actually complicates the coordination problem. It doesn't go away, but it will complicate it. And that means that our enterprise level orchestrators have to get smarter too, so they can all work together. And I think that's really what's happening. And of course, Stonebranch is well on the path to make that happen for their product because they see the same kinds of things. Go ahead, Scott. Dan, I've- Oh, you want to comment? Go. Yeah, I read your research report on this topic, and I recall you talking about it, this orchestration becoming federated. I thought that was a good way to explain this. You sort of see federation in lots of technologies where you'd be able to centralize more of the automation in the past. Now you have these agents that are being built in other orchestration tools, and that's not going away is what you're saying. What makes this complicated is now we're going to have to figure out how to centralize orchestration of all these other orchestrators alongside other types of automation, right? Yeah. Okay. All right, let's go to the next One. Yeah, and the federated thing too, I mean, it roots back to the fact that we aren't displacing technology as fast as we add it, and all of those things mean that you need some consistency too. And so yeah, I think as AI came forward very early on, everyone started to go, "Oh, it's data. We have a data quality problem." And yes, you need good quality data. Currently, everybody's focused on context. When I think about that, data tells you what happened, and that's often sufficient if you're doing some basic reporting. If you're going to go beyond that, you do need context, which is more the meaning and understanding around it. And that's sufficient for a lot of the work that's going on right now, and even within workflows as if you bring reasoning into a workflow, context may be enough. But if and when we start bringing reasoning into the orchestration decisions themselves, possibly changing workflow orders or what have you, the order of workflow or which service might stand in if a service is down that's necessary for one step in a workflow, there's a lot of places where we could alter orchestration as well with reasoning. I don't know that we're pushing that envelope yet, but if we get there, you really have to have more than even context. You need to have state awareness. You really need to know what condition are all of the systems in and where is the automation? And really, if you think about SOAPs or workload, they are great at carrying the state and context of the workflows themselves. We've always done that, but it's internal. And part of what you're going to hear me talking about is to make this federated, we're going to have to have more sharing of this kind of information, and you just need to be aware that you have enough data context or state for whatever it is you're going to ask the reasoning to help you move forward. And you need more, the bigger the blast radius is, where if you're going to radically change a milestone in a workload, you need to have a better understanding than just context before you start doing it. So we can go forward. Yeah, so enterprise state really makes action contextual. Now I'm mixing terms again, I realize that, but is the action appropriate and authorized under the current conditions? Some of the workflow things will be reasoning to just interpret an alert or put it into natural language, and that's one level of using the reasoning. It's not really going to change or alter the outcomes. But the things that will radically change it is where we need a little more information so that our reasoning can be accurate. And so workflow state, dependencies, progress, exceptions, prerequisites, we've done a pretty good job of that over the years in SOAP, and it's really this longstanding understanding of how to manage and coordinate automation. Those skills, that muscle memory we all have, it remains important as we move forward. But if we're going to stir in changes to those workflows that are reasoning based, we may need to get more information from other domains. We can do a better job of getting infrastructure health, service conditions, any security signals, other kinds of things that are going on. We're going to have to consume more of that if we're going to be making more reasoning decisions about how a workload gets done. I think the control plane's going to have to combine this relevant state from multiple systems and observability platforms are going to be supplying a lot of that environment. We do some basic dynamic adjustments in workflows today, but we can get way smarter about, what do we do when a workload is going to miss its timeframe or we need more resource? We're going to get way more flexible there as we bring this forward. But a caution I would say is don't reason about a workflow decision that you don't have appropriate state for that. I think that's going to go a little bit beyond what we're currently talking about in just purely context. So Scott, if you'll move... So does smarter automation then replace deterministic automation? This is another one that I hear often. First and foremost, we all have heard about token costs and how do we manage cost at that level. If you start introducing reasoning into workflows, I mean, think about it. In the past, we tested things, they ran deterministically, and you pretty much had confidence that if it executed, you were going to get the right outcome. But if we bring reasoning into it, we're bringing change right into the workflow, which means we have to test what we're getting before we can trust it. So it's not just the token cost of the reasoning, there's going to be other costs to understand if I got the right outcome, and there's going to be additional logging around that and so forth. So reasoning where it makes things better is worth all of that extra cost, but not everything requires reasoning. I mean, look at all of what we've done for years without it. So the first thing AI can do is expand our deterministic automation. It can get more people to do the right kind of definitions and we can automate more things and AI can help us define it. But if we run it deterministically, it's still going to be less expensive. And if we don't need the reasoning, why bear the cost and the risk that a hallucination or something might enter and we don't have the right guardrail in place? So determinism doesn't go away. It's really where it's sufficient, it's also the most efficient. And where we bring reasoning, we need to understand what that means. So think about where you're applying it and if it adds value. But this then, if you'll go to the next slide, Scott, this is where we... And this is a term, I got to give Scott credit, this is not my term, this was your term, but all of my data was pointing here as well, and I think... Yeah, I'm going to use it. Thank you, Scott. Yeah, I'm glad you're using it. That's perfect. The hybrid future has really got to be deterministic and probabilistic. For a lot of practical reasons, for a lot of economic reasons, deterministic execution, it's dependable and you don't have to crosscheck. I mean, typically we have enough data masks and editing and all the things around it that if you have a very predictable thing that if you execute it, you got a high, high probability that the outcome's what you expect. And again, with reasoning and you bring judgment in, now we're going to have to look at that a little more. I mean, if you've even used any of the AI chatbots, what's the first thing on the bottom of the screen? It can make mistakes, which means that same thing happens in production. So where the reasoning benefits us, it just has to benefit us more than the expense to not just complete the reasoning, but complete all the control structure that's going to have to exist around it to trust it and let it run in production. So enterprise outcomes are going to depend on both forms of work, and there's plenty of things that don't need reasoning to get done, and why do you want to triple the cost of that just because you can apply some new tech? I mean, that's why we layer tech, because what's efficient is still useful. So AI is going to help us to find those workflows better, I think, and reasoning will come in only where it adds value and where we learn to use it effectively. Now, this may look like an eye chart. More of this is describing a world you already know, so let me walk you through here, but this is sort of my functional architecture for what we need. How this emerges could be an enterprise combining products. Scott and Peter might go chase parts of this. They probably already are, and I do know some of what they have planned, but if you look at the green part at the bottom, these are those operational domains. They're all getting smarter, remember... And today, they are doing things like creating outputs directly, but some things flow up. That 45% I talked about, here I'm reflecting 30 to 40%. That's not because I think that we're going to see volumes drop within SOAPs, quite the contrary. It's just that if everything else is growing faster as all of the automation comes up in all of these other things, we're going to be doing more, but it will still likely be probably a smaller percent of the overall enterprise automation, but it's still that part that needs that coordination. So some of that will flow this way. I think the new thing here is the part I have in purple, and I'm calling that outcome assurance. It's a combination of better observability, and we're going to have to observe the automation directly, and that means logging why a reasoning decision was made? What was the context and state in the moment that we decided to make it work this way or that way? So we're going to have to log more. We're going to have to crosscheck certain things. We're going to have to validate the outcome. So these are new functions that I see coming, and some of this we've had, we had predictive analytics, so a lot of the SOAPs already have many of these characteristics, which then leads me to why do I think these things not only survive, but become more important and become a major part of the control plane? It will still be federated. Those domains are going to do those things, but how do you now get a central view? I think this outcome assurance or whatever others want to call it, this new layer is where the federation comes together. And really the battle for the control plane I think is going to live more at this level because why change the execution if you don't have to? I do think that many of the WLA or SOAP players are likely going to be some of the dominant players that rise to that control plane level, but there's going to be either challenges or help depending on how we think about federation and how we integrate it coming from the likes of the observability market. And really, we've often done business process automation and though of it slightly differently, but I think we're going to see process and workload automation consolidate under this framework as well. So that's sort of my vision of what's going to happen based on my research, but certainly we're going to see a lot of how you put that into products will be interesting as vendors try to solve this problem themselves. And Peter and Scott, I know, are on the right path to walk this as the market figures out how to bring all this forward. So that's my part, Scott, but certainly interested in your thoughts on that. And you and Peter, of course, are going to show us how you are going to make some of this happen. Yeah, thanks, Dan. Thanks, Dan. Thanks for really setting this up. I couldn't agree more with everything you're saying, and I think it ties really nicely to what we're going to introduce now, part of which we introduced during our keynote a couple days ago. So this is Peter and myself, we're going to chat through a lot more detail on the architecture that you're going to need to be able to control agents and agentic systems from something like a Universal orchestration platform or what we're calling a hybrid orchestration control plane. So first things first, we introduced this slide on Tuesday. We have two sides of AI. The first side is AI-powered features that we've added to UAC. That's on the left-hand side. That's your copilot. It's Robi. We have AI-powered support, root cause analysis, a new AI Workflow Builder, which I know a lot of people are really eager to play with. What we're talking about today is everything on the right-hand side, so the AI orchestrated by UAC. So just to set the context of the part of AI that we're talking about. Now, before we get started, I want to call out something that I think is really interesting. So like Dan, I have come across a lot of people asking the question, "Well, are AI orchestration or agentic orchestration tools going to take over all automation?" When I talk about AI orchestration or agentic orchestration tools, this is where the agents are actually being orchestrated. This could be LangChain, LangFlow, sometimes n8n. Amazon has its own set of tools. Every hyperscaler has something, all the model builders have something where you can actually build and orchestrate agents. We call that the inner loop. So inside of the bigger circle, you have this inner loop, that's your agentic orchestration layer, and this is kind of set up to help you build a mental framework or a mental model around how this works. The outer loop is where UAC lives and works. So the outer loop is everything that's orchestrating both the agentic and the deterministic and the actual facilitating the connection to the enterprise systems that needs to happen in a governed way between the agents or the inner loop and those enterprise systems. So hopefully this helps with the mental model between those two types of orchestrators, because what I've noticed is everybody in AI right now is trying to call themselves an orchestrator. I wrote an interesting article about this over on my Substack channel, and there's no lack of, in air quotes, orchestrators in AI right now, and they're all orchestrating different things. So the two most relevant ones for us to discuss are just the inner loop, which is the agentic orchestration tools or platforms, and then the outer loop being UAC. Yeah. That's a good point, Scott, because we don't want people being confused when they hear the word orchestration, that it's the same thing. I think it's just, the same words used in a different domain or a different level when you're talking about just running agents and trying to coordinate agents. Exactly. They're two different motions. One's orchestrating the enterprise processes, the other is orchestrating straight up agents. And both are right. Both need to be there. Both need to exist. It's how you connect the dots. Right. Scott, if you don't mind, I'll just add a little something there. If you think about it, I mean, there are schedulers everywhere. If you go back, and yet the centralized scheduler had its role too. So whether you have cron or AWS Batch or every environment has that, right? And the coordination problem means crossing environments. So even when they have all these environment level orchestrators, you still have a coordination problem. Yeah. Precisely. This is just a repeat of multiple schedulers. Yeah. We talked about it. Scheduling also... This follows what happened in the cloud world really closely. There's a good parallel there where in the beginning when cloud started to pop up, there was all this shadow IT that existed, and I remember it very clearly, IT teams trying to battle with line of business teams because all it took was a credit card to go purchase a new set of cloud tools you could start using. So, IT lost control. It wasn't until they wrangled it in with centralized tooling and centralized processes that they were able to regain that sort of visibility and control that they needed. On this slide, just a point around deterministic automation. Dan touched on this, but we also strongly agree that AI is not going to replace deterministic automation. They're going to work together. And Peter, if you don't mind, maybe talk through a little bit about the cost here and why we believe this to be true. Yeah. Scott, I always think I spent a lot of time with manufacturing systems and I think the parallel that I like for people to see is if you ever go on a shop floor, you don't see people doing all the work. You see places that are highly automated because it's very repeatable, has to be done quickly. There's not a lot of decision that needs to be made. And that is a very cheap way to do it, just to automate it. There's other areas where you have people because it involves more complex movements that involve decision making. And I think that's the same that you see here is that there's a lot of automation that's been put into IT landscapes over the years that works well, works the same way every time, runs very efficiently, very cheaply. That's going to stay there. And I think what we see is agents are going to help probably generate that deterministic automation, but also it's additive to where they're going to be put into places where there does need to be more decision making within it. So we think there's going to be a lot more use cases, but they definitely are going to work side by side. And that's why we use the word hybrid. And it's something that we see being combined with all this historical automation that's in place. Yeah, exactly. And as we have in the slide here, the future requires both. Agents are going to be used where ambiguity exists and deterministic automation is going to create value where certainty is required. It's that operational mission critical automation that you don't want to turn over to an agent that really is not going to be repeatable or do the same thing every time. Yeah. And what's interesting, Scott, is what you showed on the earlier side, which is that left-hand side where we're incorporating AI into the using of the product, building automation actually can build that deterministic layer. So that can be done much more efficiently. So it's going to be a combination, not just on the execution, but on the definition slide. Yep, I agree. All right. Quick stat. In addition to the great information that Dan shared, there's a recent report that came out from Gartner that says by 2030, 90% of I& O heads will use hybrid automation the way that we're talking about it. That's up from 10% in 2026. So it's not if, it's when. I think everybody's predicting that this is going to come together. Everybody's predicting that agentic steps or agentic systems will be used in operations. It's going to take time. And what we're seeing, and we'll share a little bit towards the end of this presentation, are actual customers coming to us and doing this today, but it's the early adopters and that's exactly what we expect to see at this stage. So we'll talk through some of those use cases. To make it super clear what we're doing here and how this is progressing inside of Universal Automation Center and other SOAP applications as they mature too, if you look at this first layer, this is really where we sit today. This is your deterministic automation. There's no agentic steps. It's very repeatable and there's no probabilistic AI inside of it. The next stage and the one that we see the most of right now is where they're adding an LLM task, or better said, they're calling an LLM model to do some reasoning or thought and then passing the information back into the deterministic workflow. We'll talk a little bit more about these patterns in a moment. The next step, which is where we think everything is heading, is this agentic orchestration workflow where you have some deterministic steps. You may call an LLM model to make a decision or synthesize data or analyze text. It'll pass that unstructured data back as maybe structured data and send it to an agentic task, which splits off and runs its agentic agents in the agentic orchestration platform. Then it'll send that information back to the workflow and maybe call another agent. But the real future workflow or even some current workflows we're seeing are mixing this deterministic and AI or LLM calls along with agentic orchestration processes altogether. And so this is your purest form of hybrid orchestration as we talk about it. Went backwards. So how do the patterns wind up coming together? And I'm going to talk about the first one. I'm going to pass it over to Peter to talk about the second one. The first one is exactly what I just showed. You have this workflow as a step pattern. You might schedule something, this could be time-based, it could be event-based based off of another system. Ultimately, you'll have maybe an LLM task, but when this agentic system happens, it's going to spin off and run the agents. There could be three, there could be 20. We have one customer that has 15 different agents running in AWS that is doing a bunch of root cause analysis and they're working that into a workflow with us that will automate workflow analysis across the board. This is that sort of step, and I have a workflow that I'll share with you guys at the end of this that explains that one. Peter, the agentic task, kicking things off is really interesting and probably a future state for a lot of what's going on. Can you talk through this a little bit? Yeah. When we talk about having a lot of deterministic automation out there presently, what we see also is that one way to help make agents, as Dan said, just make sure that they do the thing right in a predictable way is for them to kick off a workflow that either exists or that you develop and they can sit there and determine which is the best workflow to run. A lot of these workflows have variables that affect the behavior of how those workflows work. So you can set the variables to make them behave differently when they actually run. And so the agent can be really good about picking out the workflow, setting any parameters that make it work a certain way. And then you get this very good deterministic outcome from automation, but you also have that intelligence upfront that can actually dynamically change it based on some of the information coming in. So this use case is one that we really see another way to do this hybrid world, but something that gives organizations the ability to do things repeatably, but also build intelligence upfront. Yeah. And we're referring to this as a governed workflow. So if I look at this next slide, it's exactly the same thing that we just looked at in the pattern that Peter was explaining, but just said a little bit different. In this case, an agent is being used, it's been given a task, and that task may be to go, let's say, spin up a data pipeline someplace and retrieve the data to continue the rest of its task. When it needs to trigger that data pipeline, it would make a tool call to UAC. And at that point, the governed workflow, which is no different than the workflow that Peter just shared here, kicks off in UAC. This governed workflow serves as a buffer between the agent and the actual enterprise systems, thus being governed. It keeps everything in check. It makes it so that the agent never actually touches the mission critical processes and systems that you have in place. This buffer is incredibly important. It's where all your policy, approvals, credentials, dependencies, validation, retries, and audit live. And the agent cannot be used in a malicious or bad way because this governed workflow will protect the actual enterprise systems. Governed workflows is going to be something you hear us talking a lot about through this series and going forward. We think it is absolutely the way of the future. All right, so what we have here is a blank marketecture. This blank marketecture is what we're calling the Enterprise Hybrid AI Architecture. What I'm going to do is I'm going to walk through every step of this. This is the mental model, the way that you can think about hooking all of this up and making this hybrid world a reality for you. So it's blank here, we're going to fill it up. So let's start with sources and triggers. Sources and triggers don't change much from what you're using today. Today you're using APIs, you're using events that may be system events. So you're also using time-based triggers. So at two o'clock in the morning, go ahead and run this workflow. And you also might have a human that triggers a workflow. The real add in the agentic world is that you're going to have agents being able to trigger workflows as well, govern workflows like we just talked about. And so you have agents that are initiating work, and then you'll also have agents that are performing work, and we'll show that in a moment. So any one of these four boxes can trigger a workflow now. The hybrid orchestration control plane is really where UAC comes in. This is where you can create the workflows, those governed workflows, the deterministic workflows, or even the hybrid workflows. This is where your workflow orchestration runtime lives. This goes back to state. Dan talked about state being important. I personally think nobody talks about state and everybody should be talking about state. This is sort of the invisible layer that has the potential to make AI work or not. So there is some state within the agents, there's state within UAC, and it's connecting the dots between those things that give the context, the business process context across the whole business is given to the agent through UAC. So it knows if there's a workflow down or if there's a service area that needs to be set up or a tool is down someplace because it's under service and that's why the workflow's not working. It provides that intelligence back to the agent. Next, you have guardrails and compliance. A lot of this is what people are building in agents today. They're trying to use prompts to do it. Prompts are just instructions. They're not really guardrails. They're not really going to stop the agent from doing or finding ways to execute things because you can't think of every possible scenario and put it in a guardrail. And the longer the guardrails or the prompts get, the system prompts, the more error prone they are. So you can add guardrails for sure. They're important, those prompts, those system prompts that people usually use. But the business rules, the policies and the governance controls that UAC adds is what really ties this and makes it safe and reliable. Next, of course, you'll have human in the loop ability through UAC like you do today. Everybody on the call has probably seen Universal Portal as a way that you can do that, but you can also do it through UAC, the Controller itself, or any of the integrations we have with Teams or ServiceNow or Slack, the ability to be alerted when something needs to be checked to make sure that it's not wrong before it moves on to that next step is incredibly important in an AI era. And of course, you get all your SLAs and reporting that is completely necessary when you're talking about audit or dashboards or getting that telemetry data out to an observability platform. It's really what helps you manage the entire business. Now, the hybrid orchestration control plan is also where that deterministic automation, the AI enhanced automation or the agentic delegation, i. e. Calling out to agentic systems is controlled and managed as well. I always hate taking our product and making it the Center big box, but it's really true in this case. There is a lot that happens here, but UAC alone doesn't do everything that's required within a hybrid AI architecture. UAC is this hybrid orchestration control plan that helps connect the dots between the sources, the triggers, and the next three stages. So in this AI and agent governance stage, this is where you are seeing AI model gateways or AI gateways. The one that we're running into the most right now is Kong, but there's lots of different ones out there. The API vendors, the Boomis of the world, the Mulesofts, they're trying to get into this market. It's just Kong is easy and reliable right now. In a current state, this is where people who are using agents or making agents available to their employees or even models available to their employees, everything passes through the model or AI gateway because they can control costs that way. They can cap how many times an agent runs or how many times the model is used, or it can help select which model to use. Maybe the process you're running needs to have a less expensive model than using the best of the best or high or strong or powerful models to do something simple. So your gateway is how people are controlling that right now. There are going to be capability and endpoint registries that are introduced. You're going to have catalog of agents that can be introduced. Some of that can be managed inside of UAC, but you're going to have agents all over the place. There's a lot of companies that are positioning themselves to be sort of a control tower, not a control plane, but a controlled tower where they are going to capture all of the agents that are available throughout your entire enterprise ecosystem. This is where you'll go to find out which MCP servers are available, APIs and different workflows across the business. Again, in some cases we'll take over some of this, but in other cases we'll just tap into those tools like we do today. It could be just as simple as a security key or something that we tap into like you would through UAC if you're tapping into any other system. Next you have FinOps. Now FinOps, I just looked at today actually this morning I received it, the hype cycle for AI from Gartner, and these FinOps tools are at the peak of hype right now. They're all the way at the tippity top of that hype cycle chart, if you guys are familiar with it. And it's because they're talking about it a lot. There just hasn't been a tool that everybody's surrounded themselves on. There hasn't been the tool that it's merged as the one master FinOps tool, but it will, and many of you will wind up wanting to connect into these FinOps applications. And we do that through UAC. You would pull in that data and share it back to the FinOps tool as an integration. And then of course you'll have evaluation and quality tooling. And this tool exists today. There's a lot of it out there, but it's sort of scattered and some of it's inside of other agentic platforms. You may have an agent that evaluates the quality of another agent. You may have tooling that does that. So this is coming as well, but this is really that whole layer that we would connect into to help manage the agentic pieces of the business. Next layer here is your interoperability and execution layer, and this is really important. So just like today, you're going to have RESTs, APIs, you're going to have web hooks, you're going to have events that trigger automation. But in addition to that, which didn't exist three years ago or even two years ago are the model context protocols. UAC in 8. 1 just released the latest update of our model context protocol, which allows you to actually connect to UAC from Claude and trigger automation inside of UAC from Claude or any other, your choice of model harnesses that exist out there, ChatGPT, those sorts of things. But MCP is also how we may integrate in some cases with other agents or other systems. There's even a future state where MCP or agent to agent protocols, A2A is another emerging piece that actually is how agents would communicate with each other and pass information back and forth. So you may have an agent that lives in say ServiceNow or SAP that wants to talk to the agent in UAC to perform a task and to do some back and forth agentic work. So we're not seeing a lot of that yet, but that's sort of the vision for where people are going. So clearly the agentic piece is where... The agent to agent is what's emerging as the sort of go-to there. But the world's changing fast. Today it's agent to agent. It could be something different later and we'll adapt that as well. The other place you're going to see agents, and this is where you're actually connecting to agents. Again, you can connect directly to agents, but you probably in most future states will go through an AI model gateway, again like Kong. And you're going to have native executors. Now we have the nice Stonebranch logo here because we'll next year you'll begin having the ability to build agents just like you do in any other agentic platform within UAC. But this could be any native agent. So you're going to have native agents and applications all over the place. Basically any application, enterprise application you have is busy adding agents to it. I use HubSpot and Salesforce a lot. Both have agents inside of them. SAP ServiceNow, they are busy bees building up agents all over the place. And this is your native agent that you would go and connect to as part of a process. The other place where you'll have agents is this external agent runtime. This is your inner loop. This is your agentic orchestration places and where you're actually building multiple agents to do processes. So UAC, again, would come in and integrate with each one of these things or integrate through an AI platform to go ahead and trigger agents or interact with agents or interact with models. So this is a very important layer that helps you connect. And then of course, this is the same in this final layer, the same enterprise landscape that you connect to today, whether it's mainframe, on-prem, cloud systems, I won't read them all, but this is what you're automating today. UAC still automates these. The agents may want to touch these, but our strong opinion, and I'd say the opinion of the growing masses at the moment is do not let agents touch your enterprise systems. It will go haywire at some point. So you use a governed workflow, you use an orchestration platform like UAC to help create that governed workflow and the buffer between the tools. So you never have to expose these enterprise systems directly to agents. So that is the enterprise architecture. The last thing that goes basically vertically all the way down all five of the layers are additional controls. These controls will live in GRC tools. They'll live in your identity and security tools. You'll have audit and compliance that you can pull from both UAC as well as observability tools that we're connected to. You'll have policy and guardrail services. Some of that will come out of UAC. Some of them may come from third party tools as well. And then obviously the resilience and reliability, the retries timeouts, controlling the blast radius of anything that goes wrong with this, and this will largely be inside of UAC. So the full picture looks like this. You have your sources and triggers, you have your hybrid orchestration control plane, which again is where Universal Automation and Stonebranch live. You have your AI agent governance, you have your interoperability and execution layers to connect to all of the AI tooling out there. And then you have your enterprise landscapes that are what you're actually automating. And then of course you have the layers across everything to help connect across all of these layers from a centralized control plan, which would be UAC. Peter, Dan, any comments on this before I move on? Think this will make sense for everybody? Things that people need to watch out for? Yeah, no, I think one of the key things for everybody to really understand in this model is that when you talk about agents running and the agentic loop, they need to take action with what they call tool calling. And those bottom two layers are really important, but those are things that you actually have to give agents that you have to provide them. And that's part of this overall hybrid IT AI architecture is there's a lot of things that are already out there and then you wrap agents with those tools, those APIs, those even CLIs. You see a lot of that evolving about how agents are able to call things. But that's really important that you have a system like UAC that has all those capabilities that agents can interact with. Great. Yeah. And Scott, I'm impressed with how much the similarity there are from my higher view diagram in this, but you broke it down into exactly how you implement this in your vision inside of UAC. And you've just described federation in a very functional way. Yeah. Yeah. Th's why I brought you back to the federated thing. We really believe it's going to be a federated world. So there's a lot of dots connected and it just becomes more complex. It's not insurmountable. You can connect them. And we've made this real- Scott, I think we have the muscle memory and the knowledge of how to do this because we've been doing it forever, this is just a new kind of thing that we have to bring into the fold. But we know how to- Another workflow or another workload that you need to control. That simplifies it a bit because of the other tooling, but the... Yeah, go ahead, Peter. No, I was going to say what you're seeing is everybody is putting agents within their software. And then, Scott, the slide you're about to show is there's pure play AI platforms, but it's the domain that you wrap them with that enables them to do a lot of things. And that's what I think when we talk about SOAP, that's what we're doing is wrapping them with all of this automation and knowledge about our domain. Precisely. So speaking of the slide that we're on now, this is for people using UAC. We have these integrations available. They're all on the integration hub. Minus Kong, it's going to be on the Integration Hub in a couple days, but it's available. If you need it right away, just reach out to us. But currently you have connections to where we started building the strongest integrations were with the hyperscalers. So whether it was Azure, Amazon or Gemini, we started there because those are the ones that have the most GDPR sort of built-in processes. Especially for Europeans, you can connect to these and they're more safe from a GDPR standpoint. But you can connect directly to the ChatGPT, the Clouds, the Geminis, and even local models like Ollama. And you can do that with integrations we've built. You can create your own integrations and you can create them very quickly now with the AI Agent Builder, which was announced on Tuesday. The other piece that, again, I'll just plug real quick, is that the MCP servers available on 8.1. So when you update, upgrade to that, there's a whole bunch of neat things you can do even from a self-service standpoint if you don't feel like opening up UAC. You can control it from the model builders or the ChatGPTs, the clouds of the world. The other thing we've added here in the integration hub is an n8n integration. And we did this because we spoke with some customers who were starting to build experiments or prototypes inside of n8n. n8n's a great developer tool. It's neat. It's just not very secure and it's not going to run reliably all the time. It tends to break after a certain period of time. So what we did, the approach we took here was much like the approach we took with Airflow or other orchestrators that exist in the world. You can connect those in Airflow DAGs or in n8n, the workflows that exist there and control them from Universal Automation Center. That adds event-based capabilities, that adds your observability, that adds auditability for the automation that happens instead of no auditability that might live inside of n8n as it runs a workflow. So this gives you the SLA compliance, all the things that you need, just still run the workflows in n8n if you want, but you can control it from UAC. It's a really cool integration for anybody using n8n right now. These are just the integrations. What I wanted to touch on though is we have some early adopters, some people who have already started building workflows. The three places that we're seeing this, the three sort of use cases are summarized here. The first is really anybody creating data pipelines. There's a real clean opportunity to analyze unstructured data, to synthesize, to add sentiment analysis to that data, to enrich other data with various data sources. And this one that I'll talk about in a moment there, enriching data that they have with publicly available data. But one of the cool things you can also do is mask data. So you don't want to expose people's personal information to cloud-based AI tools or public AI models. You probably want to strip that data away with a local model that doesn't give that data any training to anybody else. So we have tasks that we've already built and people are already using to go ahead and review files as they come in and then anonymize it before it moves on to another external third party model up in public where you may be concerned about GDPR or other regulatory problems. The second in the list here is orchestration of third-party agents. I think we talked about this at length. We have a few customers that are already doing this specifically around root cause remediation type stuff. I'll talk about that. And then of course, root cause and remediation specifically is where we see a lot of people going first. This is as simple as adding LLM tasks to a failure that populates a ServiceNow ticket with what actually happened. Instead of passing just raw log files in a ticket or asking a first line service person to go figure it out, take 20 minutes to review the logs, the root cause analysis and remediation kind of workflows we've seen being built are inserting an LLM task or an agent that analyzes the logs for you and then populates that data inside of ServiceNow so that you can really... I always screw this up, Peter, shift left. Am I right there? This is the shift left sort of approach? Yeah, shift left. That's it. Some part of my brain wants to say shift right and I always know what's wrong, but it's shift left. Okay, so that can happen there. Let me show you guys a few workflows. One customer's set of workflows, we've anonymized it obviously. But this is a pharmaceutical company. They're very data intensive. They are pulling pharmaceutical data from third party tools. They're combining third party data with public data, and they're even privatizing the data the way that they're doing these things. So I think the cool things about this workflow, if I can get my mouse to go over there, for some reason I can't. Okay, here we go. The cool thing about this workflow, in the first workflow, it's completely deterministic. So they have three different workflows they've built, but the deterministic workflow winds up passing into other workflows. So this one's totally deterministic, same kind of workflows you guys are mostly building today. This one actually reaches out to a public AI path. So there is a website that hosts a whole bunch of different pharmaceutical industry data, could be pharmacy news, could be data about medications, just all the latest stuff. It's actually pulling that data with an agent. It's doing a data integrity check with, in this case, ChatGPT, and then it's pushing it into their analytics tool, which in this case it's very pharmaceutical specific, it's called Revvity. So this one is pulling public data, analyzing it, turning it into structured data, and then dropping it into a structured database of sorts that they can analyze. This third path here is using a file transfer UDM. They're moving data from one server to another, but before they move this data from an on-prem server to a cloud server, they're running it through Llama. So they're doing this data integrity check, but they're also masking the data in this step like I talked about earlier. Then it gets pushed into a few other cloud places, but ultimately everything winds up in this analytics tool. It's synced via SharePoint. If you guys have ever used the cloud to cloud file transfer integration that we have, which is very powerful, you can just move from bucket to bucket very easily, whether it's SharePoint or any other cloud bucket. And then of course the process owner gets notified. And I know we're a couple minutes over. I'm going to keep going. For those of you that need to drop, please do, but this will be available on demand and you can come back and watch the last few use cases we talk about. Okay. The next use case is the root cause analysis one. So in this one, it's really cool. One of our customers... I don't know why that happened. One of our customers is using Dynatrace to monitor what's going on in their entire environment. One of the environments they're orchestrating or looking at is Universal Automation Center. When Dynatrace detects an incident, it triggers a workflow in UAC. That workflow starts. So this middle layer is your orchestration layer. It's the UAC layer. It starts the orchestration, it triggers a incident response workflow. Part of that incident response flow is an agentic call to an agent orchestrator. If memory serves, this one is the Amazon agentic, I think it's AKG or something like that, system. And this governed sequence inside of the agentic Orchestrator has multiple agents that are working in tandem to figure out what caused this incident. Once this agentic system figures out what happened, it traces it does everything it do, it actually passes it over and opens it up in the ServiceNow update. And that ServiceNow ticket includes all of the analysis. This ticket by the way is also... Or these agents by the way, are also looking at ServiceNow and pulling in other tickets that are similar. So hey, this is a similar problem to this ticket. This is how it was resolved to give them hints around how they can resolve it quickly. So not only is it given in the verbal English version of what happened and what they see from the log data and how to fix it with recommendations, they're also adding in that other tickets that have been similar so that you can follow those steps if it's an easy thing. So those are two really interesting case studies that have emerged from our installed base so far. Once these are up and running and going, we have enough of them, we'll start having customers come on. So if you guys are doing anything now that's really cool, reach out to me. I'd love to feature what you're doing or brag about you out on one of these kind of webcasts in the future. So we expect to see a lot more of this in the installed base. It's early days, but it's coming fast. And just about every RFP or new request we get coming in these days includes agentic orchestration as part of the set of questions that they're asking us. So this is going to be a thing. It is a thing. Peter, maybe you can close us off here and talk about why we're so important in this world using what's on this slide. Yeah, sure. Yeah, I think again, we use the word hybrid and we've been really working on this for going on about two years now trying to figure out how AI, and it's evolved a lot in the last two years, especially from an agent perspective and the agent frameworks and capabilities. And we've learned a lot about how they need to interact with what exists today. Organizations, especially IT organizations have a tremendous amount of existing automation. And we've talked to a lot of customers and the adoption is still early, but people are learning that, hey, there's the hybrid where there's a deterministic portion that will be there for the reasons that we've discussed. Agents will do one of two things. Agents will not replace it, but they'll be used where there's more intelligent decision making that needs to get done. Some of the examples that we showed in those workflows, they will be on the front end in creating automation. So we've seen a lot of capability where we can create automation a lot easier. So it's very much a hybrid world. And then the governance is really key. When we talk to customers, it's still the probabilistic nature of AI, putting that governance layer on it so that you can have control on the cost, have audit trails about what happened, especially if you get more into the autonomous nature of it. And then also even when you're executing the deterministic automation, you want to be able to tell what choices were done to actually execute that if it was in fact an agent that was used to kick it off. And so all of the governance and traceability around there has to be there. And that's something that we've done for a very long time. So we're taking that same reliability governance audit trail that we've done with enterprise automation and doing that same thing around all of these agents and AI functionality that we'll be able to incorporate. Yeah, exactly. So, very excited about where this is going. Our customers are getting more and more excited too because it is a learning path. So they're seeing what they can do. We're kind of listening and making sure that they've got the platform that enables that. Yeah. And I think the way to think about this from a learning path is the people on this call, they're in automation, they care about this world. You can start working on this now. You can start learning this. This isn't something that's going to displace you. This is something that gives you more power. It gives you the people that really get good at agent work and combine it with deterministic, you're in the pull position going forward. You're the ones that can run the whole business, especially if you're anywhere near believing that agents are going to replace people someday, which I'm not quite sold on. But think about if you're the guy running the whole thing, this is the reason to start getting on top of this. In addition to all of the productivity gains and everything else. So that's the end of the slides. I don't see a lot of questions. I was just flipping through. If anybody has questions, feel free to submit them now. Peter and I are willing to obviously stick around for a few minutes to- Hey Scott, there are a few questions in the chat, if you scroll down. Okay. I can read them to you if you want. Dan did have to jump, but Peter and Scott, if you have just a few seconds, I can read these questions. Yeah. I see one. I'll read them as I go here. Okay, great. Does UAC 8. 1 support multiple LLM providers such as Azure, OpenAI, Anthropic Cloud and other enterprise hosted models? So yes, we have those integrations, but an even more important thing to acknowledge here is that because we're vendor neutral, because we can access any of those tools, you're not locked into just one model. You're not locked into just one ecosystem. You're not stuck just in Claude or stuck just in OpenAI. You can do multi-model orchestration from UAC. So you might use Ollama, a local model to do one thing. You might then use, I don't know, OpenAI to perform another task, and then you might use Claude to perform another task all within the same workflow. So that is completely possible and you can do inside of our tooling. Let me go down. There's one that says, "Can we incorporate human approval through the Universal Portal before executing sensitive or business critical operations?" Yeah, human in the loop is a key capability that we're going to incorporate with AI so that you can actually have a human approval, whether it's through the Universal Portal or through UAC. We're looking at how that's going to get integrated in. So that is definitely part of the governance. Yeah. Yeah, exactly. And you said we will have it. We have it. Just to clarify for everybody. Well, it's not in the Universal Portal yet, but it is going to be incorporated in any place that you run an agent because there's the whole agentic functionality that has to be there. Okay, cool. Well, we're- There's another one that says," Can I use text to make workflows in AI Workflow Builder, including dependencies, logic, error handling?" Yeah, so you can make workflow, have text to workflows. I think Gwyn even did a demo where he took a picture of a workflow and had it create a workflow. You have to incorporate all the features into building a workflow. So in 8 .1 that just came out, you can build tasks, certain types of tasks and then workflows to a certain complexity, but that'll get more and more sophisticated over time, even just to the point where you can actually build new integrations through a workflow builder. So it's not just what you have defined in UAC, but also something that it can go off and build with AI. So there's a big roadmap that we have with increasing complexity and what you're going to be able to do. Wonderful. All right. I want to be sensitive to everybody's time. We have two more sessions coming up. I think Nicole's going to tell us about that next. So Peter, thanks so much for joining us today. I know Dan left, but silent applause for Dan for joining. I think that his session really helped clarify what's happening in the market. And let's go ahead and pass this over to Nicole, and we'll see you guys on the next one. Yes. Thank you, Scott. Yes, I'll be mindful of everyone's time. I did want to mention, so earlier I said that this was a four-part series. So next week we do have two more sessions on Tuesday, October 20th. We have what's new in UAC 8.1. And then on Thursday, October 22nd, we have the hands-on AI workflow training. So we hope to see everyone there next week. Lastly, you will receive an email from us over the next couple days. It will contain the link to this recording, so feel free to share it with your colleagues. But again, we just want to thank you for taking your time out of today. I appreciate you being part of Stonebranch UserVerse Online. And with that, we'll see you next week. Thank you so much.