Revolutionizing Business Orchestration with Agentic AI: Insights from Pega and Accenture

- Hello, everyone. My name is David Steuer and I'm a managing director in Accenture's Software and Platform Engineering group. And today I am with Don Schuerman, who's the CTO of Pegasystems, and Nick Kolegraff, who is an AI executive from Accenture. And today we are going to dive into the world of a agentic AI. We're gonna talk about what it is, its relevance in today's business landscape, and the role that Pega will play in this space, and really why it's top of mind for many IT executives. And Nick, maybe I'll start with you. Last year, at this time, discussions were all centered around gen AI and large language models. And now the buzz is all around agentic AI, and maybe you can start off with just explaining for folks, what is age agentic AI?

- Yeah, definitely. Thanks, David. Agentic AI, I think everyone can think of it as a way to solve incredibly complex business automation tasks where when we looked maybe a year ago when kind of the rise of LLMs and these technologies were kind of in the fruition, everybody was kind of looking at the individual task level and kind of just automating single tasks at hand. But what we've found with these technologies and kind of the rise of agent architectures is you could actually solve more complex business workflows where there might be a collection of complex tasks that need to happen in different orders in order to solve a complete business problem, and that's where kind of agentic AI and that kind of architecture has come into play to kinda create technologies around solving more complex, more challenging business automation tasks.

- Yeah, it makes a lot of sense and it really is something that has spawned from the advanced technologies that the large language models and the advanced technologies that the companies building them are bringing to us, and just the roles that these agents can play. And I've actually heard some companies talk about... When they talk about their workforce, they'll talk about their agent workforce working together with their human workforce. And Don, so from your perspective then, what's Pega's point of view around agent AI and what role is Pega going to play in this space?

- Well, I really like the definition and the description that Nick gave, about moving from sort of single tasks to actually being able to automate and orchestrate what are really almost entire workflows and in many cases completely from end to end agenetically. And so one of the ways that we've been thinking about this at Pega is agents need to be able to design workflows. That means they need to be able to figure out what tasks they do, plan and design the workflow. They then need to be able to automate the workflow. They actually need to be able to take in the tools to get all of the work done, and they wanna be able to optimize it. They need to be able to go back and look at how things were done in the past and see if there are opportunities to do them more effectively or more efficiently in the future. So when we think about the future of agentic AI, we look at through the lens of how do we help agents design, automate, and optimize the workflows that actually drive all these results for the business. I also think it's really important that we think about how agents work in a landscape where there will be some workflows that are already predefined for them. I think, for example, a bank doesn't want an agent necessarily figuring out every time how it might handle a fraud interaction. We probably want the agent to follow some pretty established and deterministic rules about how it does it. So there are some workflows that the agent is just going to automate based on a pre definition, but there are also times either inside of an existing workflow or when the work is outside of the scope of existing deterministic workflows, the agent needs to design a workflow on the fly, figure out what it could do, and put a series of tools and tasks together that actually drive to the outcome.

- And, Don, you're hitting on a pretty important point there. I think, a lot of times, when people think of agentic AI and the connection with LLMs, they think agent AI is all about using large language models, and you're actually talking about some deterministic work that occurs in between that as well. So it's really using lots of different technologies together. Is that right?

- Yeah, I mean, I think... Look, I think that you could pretty safely say that any kind of modern agent is gonna have an LLM inside of it and use LLMs to do a bunch of different stuff. But I also think even in our own internal experiments with agents, you wanna balance the LLM with rules, guidelines. And in some cases you're giving the agent a predefined workflow and steps of reasoning that you want it to follow. So it's that mix of using the LLM to have that freedom, but using orchestration to ensure that the agent is following best practices and adhering to rules and regulations that you want inside the business, frankly, the same way we do with human employees.

- Makes sense. And, Nick, Don talked a lot about design, automate, and optimize, and I like the way he kinda described those steps involved in the agentic workforce. And Accenture has recently come out and built a concept of a refinery, and I think that enables that. And maybe can you talk a little bit about what the refinery is and what it does?

- Yeah, definitely. So I think when we kind of look at enabling kind of agentic architectures or workflows or kind of using them to solve problems, we've found that you kind of need some building blocks and components of technologies in order to kind of enable that to happen. So AI refinery is really a framework that essentially utilizes to build on top of, and provide a foundation for these agentic architecture. So that includes kind of having a data foundation and a cognitive brain aspect that allows companies to put a semantic layer on top of their data to give it access to different LLMs and creates a framework for building their own customized LLMs if need be. A lot of clients have different use cases and may need to kinda do their own customizations and fine tuning for their strategic purposes. So we have a component that enables that within the AI refinery as well, along with kind of our agentic framework that sits on top of those capabilities that enable different levels of prebuilt agents, as we have found that there's different tasks and different consistencies across clients that may need different types of utility agents to solve various different tasks. For example, if they needed to do some research-oriented tasks, there's kind of research capabilities where they're reaching out, they're researching different aspects across the internet, and kinda pulling that and summarizing that information, as well as kind of those summarization aspects of the work that needs to be done. Kind of all of that plays into the AI refinery framework to enable companies to really take advantage of these agent architectures.

- Well, and you hit upon something, and Dom was hitting this too, that there's different agents that are doing different things within agent architecture almost the way humans work. You can have an agent that might be reading a document and creating the first pass of the solution, or reading a transcript, creating the first pass. You can have agents that are providing input. You can have agents providing overall knowledge based upon the company's policies. That's essentially what you're enabling as well. Maybe talk a a little bit as well about a lot of companies when they're building out these solutions, there's always a first step in incubating and kind of getting these agents to work together. But the challenge really comes into play when you need to scale it into an enterprise IT environment. Maybe talk about some of the things security observability that need to go along there.

- Yeah, and I think... And the scaling aspects, I think we're definitely finding that as the market of AI right now is evolving, we're kind of moving from people that we're doing kind of a lot of experimentation and a lot of kinda figuring out. Where can I use this? Where can I adopt this technology? Now, we're kind of finding that use cases across these companies are kinda stabilizing. People are finding that they have found some use cases that they can actually use company-wide. So scaling becomes incredibly important, and that's where we find that the framework that we kinda have for our agentic architecture around... We have three kinda core agents, the orchestrator agents, the super agents, and the utility agents, that kinda create a foundation. We found that when you're kinda doing complex tasks that are not kinda single task-minded, but it is the entire kind of workflow that needs to be curated and created, the orchestration piece becomes incredibly critical around that whole architecture. And that's everything from kind of the observability and the monitoring, because you're moving out of the POC world where we're kinda doing an experiment, we're trying to find value, and we're looking for that value. Now, you kinda have a core business function running on top of this system. So when things go wrong, that could mean revenue loss, that could mean additional expenses, that could be more time that somebody has to spend on some aspects. So that's where that... In the orchestration piece... And on top of that, in terms of just risk management, overall operational risk management, because now we've got... You've got compliance in the mix, and these agents also have to adhere to whatever compliance, or business rule, or logic that needs to be in place to make sure that they're operating in an ethical and compliant manner. So that orchestration piece is incredibly critical because that's the agent that kind of is surrounding all the other utility agents and the task agents that are performing these tasks, but it's the one that's also kind of aware of the business context, aware of the compliance, and the rules, and the logic that would go into understanding how to orchestrate these other agents so that they're utilizing and adhering to kinda those business rules that would need to be in place, but also notifying the humans that are kind of also monitoring this system to say, "Okay, hey, there's an issue over here. We noticed that there's some drift happening on this agent." And so those monitoring pieces and operations become super critical on those scale out architectures and most of it kind of sits in that orchestration agent's purview.

- Yeah, and that's where, Don, really, I think, as this evolves, it's going to be... We're gonna find more and more opportunities to help organizations with the combination of what we're building from a refinery perspective to manage agents and orchestrate those agents, but then also in interfacing with Pega to help manage that work and the ways of working. And maybe, Don, you could talk a little bit about that as well.

- Yeah, I mean, it's been really interesting to watch this space a little bit. And as we've seen this growth in agentic AI, we've also seen the word orchestration bubble more and more up to the top of a lot of these architectural considerations. The cleanest example is Gartner has started talking about this concept of business orchestration and automation technologies. And Gartner has very much built this up in the context of a world of AI agents where I can foresee a future where it's very possible that a company will have as many AI agents as it does employees, if not more, given the fact that the cost profiles are so different. So in that world, orchestration of those agents becomes absolutely essential for all the reasons that Nick kind of referred to. So what we're thinking about in Pega is, how do I help do that? And we talked about sort of this idea of being able to design the workflows built, in some cases, for agents, which is why we're continuing to invest in things like our GenAI Blueprint product, which is really a workflow designer that we've used historically traditionally more and a traditional SDLC where it goes through a design time to runtime. But increasingly we're seeing the Blueprint capabilities get pulled into runtime in real time as an agent might need to design a workflow out on the fly, but still actually follow and adhere to the same kind of best practices an organization would use if it were designing a more predictable and standardized workflow. So being able to make sure that orchestration exists and then plugs in. The other thing that we think is really interesting is, how do we make the universe of workflows that live inside an organization available to an agent? So imagine that you have an agent that really it's job is to interact with customers and provide customer servicing capability. So, again, I don't want that agent figuring out how to handle every request on its own. If existing workflows are there, I want that agent to be able to dynamically pull in those existing workflows, guide the customer through it, and get them to a rapid response that already adheres with the customer's best practice. So making sure that the workflows you design and build in Pega are then available via agent API. So all of these different agentic front ends that are gonna be emerging still have that universe of workflows. We start talking about this as being an agent fabric. How do I start connecting all these agent pieces together but ensure that whenever possible they're operating from the same best practices, the same set of rules, the same view of information about a customer?

- Yeah, that makes sense. And I like the connection of Blueprint, of what Blueprint is doing and what Blueprint essentially is going to evolve to. Makes perfect sense to be able to build those workflows that the agents can take advantage of. And Nick, maybe over to you. Don had spoke a little bit about the impact this is having on businesses. You made the statement that actually may come to a point where you have as many agents as you do, AI-enabled agents as you do people in the workforce, maybe more, maybe less, but we're definitely moving in that direction. What other impacts do you see happening with business based upon this architecture and these solutions?

- Yeah, definitely. Obviously, the biggest impact I think is just in terms of inefficiency play. So looking at kind of reduction in terms of capital and operational expenditures as a result of implementing different agentic AI strategies. And I don't necessarily see it as... I think folks might be a little bit concerned in terms of, oh, well, it's reducing workforce and now AI agents, and other things are kind of driving that narrative. And I don't necessarily see it as that. I see it as kind of just a repurpose or enabling companies to find some efficiencies and find those operational savings in either capital or operational costs, but then kind reallocate and reinvent themselves in other areas that maybe they wouldn't have otherwise thought of pursuing some new capital investment, taking on either new markets or expanding their business and kind of shifting that energy transfer. I think agentic AI is really... Energy has only changed and shifted. And so I think agentic AI is kind of enabling that energy transfer, that energy shift into different areas where kind of the remedial tasks that computers typically are better at are kind of compounding and moving that direction. The agent architectures can take on those tasks, but the more human-focused tasks are really just gonna open up for the humans to kinda go focus and take on those tasks more than kind of the repetitive type tasks. So that's kind of how I see the agentic AI landscape kind of enabling business as a whole.

- Yeah. As a you former physics major, I love the conservation of energy as sort of a metaphorical principle here. One of the things that we've been talking a lot about at Pega is this idea of the emergence of the autonomous enterprise. So through agentic automation, through large language models, through more statistical AI, more and more of what we do in the business becomes autonomous, becomes self-learning. And I don't think this is like, to your point, that we're moving to a world where every business is a driverless car. What I think we're moving towards is every driver and employee in the business has the autonomous features surrounding them to ensure that they can operate safely, they have a better experience, they're focused in driving on the right things. And so, as we see this shift to the autonomous enterprise, one of the things that we're really focused on at Pega and frankly inside of our own transformation as we start to use AI agents, as we start to develop more automated and autonomous portions of our business, is, in addition to the technology, the sort of fundamental ways in how this will change how we work, the different sets of skills. The ability of being able to interoperate with an AI agent is itself a skill. The ability to prompt a large language model effectively is itself a skill. So I think the other thing that's gonna be really interesting for organizations as they seek to reap what I think will be both efficiency benefits, but in some cases maybe profound sort of disruptive and transformational benefits, thanks to the freeing of that energy. What are the skills that they need to be developing inside their organization? And how do they need to be thinking slightly differently about how people are going to work a year from now, two years from now, three years from now?

- Yeah, no, there's always work that needs to get done in also your discussion around enhancing the work that folks are doing. It's similar airplanes, right? We have airplanes that could practically fire themselves, but we still want a pilot in that airplane at all times. As we kinda shift into the end of our discussion, and what I wanna do is maybe, Don, to go back to you and then over to Nick on any... We talked about a lot of different things, very interesting around agentic architecture, and this is something that is evolving but rapidly evolving. Any takeaways that you to highlight before we finish?

- Look, I think the importance of orchestration, and we hear agents referred to as the digital workforce. We know how to apply orchestration now to help orchestrate the human work. We're going to want to naturally feed agents into the same kind of model. And that's why I think watching what Gartner is evolving around both. The call to action that I'd offer for folks who may be listening is check out what we've done with Blueprint. You can go to pega.com.blueprint to check it out, 'cause it gets you into this idea of AI-driven workflow design. And we are adding some capabilities to Blueprint where you can actually see workflows being executed by agents and experience that in real time. So you begin to get a sense of what this future is gonna look like and begin to think about how that maybe changes the way you look at things like how I service customers or how I intake work from employees, because I think it is gonna profoundly change some of these things and profoundly change the way we think about the types of work we orchestrate in the business.

- Yeah, no, that's great. And being able to actually see that in action by just taking a look at the demo, and I'm sure that'll evolve too. And Nick, from your perspective, any final takeaways for folks?

- Yeah, I think one of the significant values I see for our clients and Pega's is in their service orchestration. So if you think about agentic architectures, there's kind of orchestration of the agents themselves, but then there's also kind of orchestration of the people around the agentic architecture, and as Don was kind of alluding to is how people work and how people operate will change with agentic architectures kind of on the rise. But with Pega service orchestration around the age agentic AI, it really enables folks to solve that crucial component of how do we not only kind of orchestrate the agents themselves, but how do we create new processes and new business operations around this agentic architecture, 'cause those things will likely change. And that's where Pega's core strength in kind of its ability to orchestrate those services from both aspects, from the technology side and the business operations side. It is a differentiating package of technology.

- That's fantastic. I'll just add to that. I think the other takeaway I'll throw out there is, it's one thing to build these type of agentic solutions in an incubation environment, another thing to take these solutions and to scale 'em into IT, to scale 'em into the business, so there's change, not only change that needs to occur within the business when you implement these things, but changes that need to occur within IT, and things you need to do to, even across the whole agentic architecture, to make it work within the ecosystem. And so I thank you both for this discussion. Fantastic, very timely, very interesting discussion. And I'm sure there'll be more, that evolves as we continue to advance both in the work that we're doing for our clients. Thank you very much.