PegaWorld 2026: Beyond Buy vs. Build: Making Agentic AI Work at Scale

We're going to we're going to get started today. Thank you so much for, for for coming here. I think we're going to have a very interesting session. It's definitely a topic that's very hot. We're going to be talking about Agentic AI and the discussion about when do you when do you use a product? When do you build, when do you use something like a cursor or a cloud or open AI to actually build something using AI native? And how do you bring that all together to scale? And I'm telling you, AI is moving so fast that probably by the time we finish this conversation, anything that you hear is going to be obsolete within minutes, right? So maybe you don't need to be here, but okay, well, let's, let's jump in. Um, so I'm going to be facilitating the discussion. I lead Accenture's Pega practice.

I also, I'm also very involved with many of the AI hyperscalers as well. And I have with me Mart Gombert from Rabobank and Marte has been building agentic and agents for for really the KYC and, and the fraud process over at Rabobank. And I think Rabobank is really one of the most advanced banks when it comes to what they're doing with Pega not only. And within Marte's area, March focus is around the the case management around that and the agents that are being built around case management. Many of you also probably know Rabobank has done a lot, a lot of work around NBA as well. Um, although that's not where Marte is specifically focused and, but I do have Ryan here and Ryan is part of Rob Walker's team senior product strategy around customer decisioning. And so I think we're going to have all our bases covered as we have this conversation around both, really both how you're building agents, not only for, for cases, but also for, for CDH and I know Mart, we were going to wanted to start you off and have you set some context, but I think you've got a video and maybe should I.

Let's first do a very quick introduction about how we use Pega at Rabobank because we started already 26 years ago with Pega with smart investigate in the retail domain. And currently we have a community of Pega community of 600 people or more than 600 people within the bank, and we use it across six different domains. And one of the domains that is effect financial, economic crime. And we had the discussion about the word effect yesterday. And I see already one laughing and but it's financial, economic crime. Not a Dutch curse, just to just so folks are clear indeed. And so there we use Pega. And I want to show a short video because then everybody knows the context about fact and how we use also. AI within financial economic crime. When in fact, analyst investigates an alert. Understanding past activity is critical, but that history is scattered across systems and alerts requiring manual reconstruction.

This makes the process time consuming, repetitive, and inconsistent. Our solution changes that generative AI is embedded directly in the Pega workflow, where analysts already work. It securely uses approved case data to generate clear, structured summaries of previous alerts. No searching, no copy pasting. Analysts instantly see key patterns, past outcomes, and risk signals. If they need more, they simply ask and the chatbot responds in real time. This enables faster, more consistent investigations, helping analysts focus on higher value work with less manual effort at scale. This delivers a significant value when implemented across Fech while improving consistency, completeness and strengthening control. So financial economic crime is all about investigations from, let's say, CVD risk, but also helping with offline onboarding and online onboarding and also fraud investigations and anti-money laundering. And we started around five years ago within this department. We had all kinds of different systems, all scattered, all workflow management or case management included in different systems.

And we thought we should go to one single unified view for all the analysts, although the analysts are always different analysts and doing different investigations. We want to route the right case to the right analyst at the right moment. And that's where Pega came in and we are now. We are almost finished with migrating all the workflow and case management to Pega for the financial, economic crime domain. And how as you've gone through this process, how is it? How has it evolved? Like how did you evolve into Agentic as you, as you went through this? Yeah, we started two years ago with with GenAI and Ronald de Ligny helped us with that, with building the first, let's say the chatbot. And we started with doing it with, um, without PII data. So we thought that will be the easiest way to do it, because if we start all over completely into the system, into the case data, there's a lot of PII data involved.

And then it also you see a lot of risk. And also, Show up at a AI. Act that we have in the. In the Netherlands, in Europe, it will be hard to implement that in, let's say, a short amount of time. So we thought we start with work instructions. That's maybe easy. We have. We have a system with all the work instructions. So we integrate it into Pega, into Knowledge Buddy and the analyst can start asking questions. So we did it. Uh, yeah. Quite fast. In three months we were live with this system and we use Knowledge Buddy from Pega for that. That's that's fantastic. And a great, great result very, very quickly. Um, I do want to maybe pivot over to you, Ariane, and maybe I. So I spent, I was over at, uh, anthropic last week. And so this is kind of top of mind for me, right?

Where there was a recent, uh, quote from, from anthropic CEO, I think he was in a meeting with with Jamie Dimon and he said he said if any SaaS because there was a question, of course, he's always getting the questions right about about SaaS and the SaaS moats. And he said, well, if any company is coming in and saying our software is complex and hard to write, that that's not going to work, right? That that's going away. He also said something which is interesting. He said, we are going to see very clearly the moats here are going to go away, and we're really going to pivot, and we're going to find what we're going to find are companies that are companies that do well. The SaaS companies that do well are going to be the ones that that use AI differently, that are taking AI and using that to evolve their product.

And maybe you can comment on that on where does, how does Pega see that? Yeah, I think that Dario is not wrong. Complexity becomes a lot less of a burden with new AI tools such as Cloud or Copilot or Gemini. And there's a lot we can do. We see it as an amplification of what we already did. If you think about the product that I'm responsible for, Customer Decision Hub, that is an authority when it comes to decisioning across channel in real time at hundreds of billions of data points. And that logic, that infrastructure you cannot replace. Yes, you can replace a lot of complex stuff like we saw this morning when Kerim did this massive demo and beautiful demo about predictable AI and how things become flexible. But again, I think it's an amplification of what we already do. And that moat, to use that word that we have of real time decisioning at scale is something that cannot be replaced.

So we put agents and agentic next to that to amplify what we already have. And remember that an AI does not necessarily replicate security and governance and auditability. Something that we've been building for 15 years inside our product. And it's funny you mentioned anthropic because I also was invited to a dinner last week at anthropic, and this, of course, was top of mind for a lot of customers in the room, like RBC and BMO and CIBC and Porter Airlines. I was in Toronto. I live in Toronto. The same topic comes up and everybody feels the same way. Actually, the hype that you see in the market is a lot less than what you hear from professionals dealing with this day in, day out, just like Mart was explaining now as well. Yeah, I mean, even even anthropic themselves, they're not necessarily coming out and saying all SaaS is going to go away.

In fact, many are using. In fact, we saw today Pega we're going to be able to use anthropic models and OpenAI models and other models to build better with with Pega going forward. So it's kind of all, all, all coming together. I do want to go back to you, Mark, because you mentioned mentioned sorry, back to you, Aria in for a moment and I'll go back to you, Martin, because, Arjun, you mentioned customer decisioning and maybe double click into that a little bit because there are some customers looking at NBA and saying hold on a second. Um, I'm not saying I can use I'm going to take gen AI and have Gen AI make all the decisions for VA because that there are some that might be looking at that, but that's not necessarily realistic because you do need rules and you need things to be deterministic.

But there are some companies looking and saying, well, but do I, do I really need to use Pega NBA or can I, can I using a tool, an AI enabled tool? Can I build my own NBA. Right. What are maybe double click on that a little bit on what are some of the core capabilities, right that Pega is bringing to the table around that? It's a topic that that is top of mind for many of our customers, many of our prospects, people that are dealing with next best action, real time decisioning, day in, day out. It is a very common topic that we need to address and that we have to look at how can I help us? In my mind, it's like a 28th division. The 20% is what an AI can easily do. It is not hard for an LLM these days to create a recommendation for this person.

It should be this message at this time. But again, what I said before, try and do that at scale across all your channels in real time with a sub 200 millisecond real time interaction to that customer, that is difficult. Next to that, what we've built over these past 15 to 20 years is also that security and governance. I keep emphasizing that because that is, especially in industries like Mars, incredibly important regulatory, you have to be able to define and defend why you promoted something to a particular person at a particular moment in time. You always have to come back to it. Auditability and regulatory compliance are top of mind, and that is something that an LLM cannot address today. Yeah. Be a challenge to and and you brought you brought up you brought up a good point. And by the way, there's a there's a barcode up there that actually addresses.

It's a, it's a double click on a white paper that Rob Walker and I co-wrote together that really dives in, gets into the technology of, of, of this decision specifically for an NBA of what are some of the core capabilities that, that you need and when do you, when do you need to make a buy versus build? Because there are some situations, right? If you have kind of a more simple, simple decisioning process that it might not make sense, Pega might be overkill. It might be might be okay to do that AI enabled development, right? But when you have, when you really want to be able to do personalization, as you were saying, and if you have, let's say you have to run a thousand, a thousand models that you're personalizing on because of different, different offers, different actions, different things that you're taking. That's, that's when, uh, having that out of the box in Pega, it's a, it's a huge, huge asset company.

Absolutely. Yeah. Um, and within the, within the bank and I'm sure your bank very advanced as you're going through these sorts of discussions and these sorts of decisions of when do you, you mentioned and you showed the video of the agents that you're building in Pega, you know, and at the same time, how how does the bank look at when to build an agent in Pega and when to build an agent in something else? I showed we have, of course, the end user is within Pega. So we think it's great to have, let's say the agents into Pega. If it's real time, it's also getting real time data, and the data is in Pega that we built at least all the agents within Pega. But if the data is outside Pega, for example, we are now building a transaction analysis agent. So the data is not in Pega.

So we will not bring all the data into Pega. So this agent, we will run outside Pega with Pega Process Fabric identity Process Fabric. And we connect both systems also together for for example, for the auditability that we have at least all the logging in a central place in Pega and see what is. What did an analyst say about this case or about this investigation? What did an agent with impact say or do, and what did an external agent also do for this investigation, for this client? And in the in the demo you showed is that that's using all agents in Pega or are you are you actually are some of them using agents outside. Of what the assistant. That's our philosophy that we that we built up on top of Knowledge Buddy. So Knowledge Buddy is only, let's say static data that we have within the bank.

For now, it's work instructions, but also policies will be included. And later on we also include the case data. So we have currently an AI use case running about case data to interact with previous case data, but also from also different systems because we migrated two years ago to Pega. And so all the old cases are still in an old system. So we include them also in the summaries and brought them into Pegasystems or currently it's still an, let's say, separated AI use case. And then we include also the current case data about the current case where the analyst is working in. And later on we combine all three together to the Pegasystems. And then it doesn't matter anymore where the data is coming from. And also in the future, there will also be also external data sources or agents included in that system. I see.

Okay. And that's kind of where it's going to evolve to. And, and Aria and by the way, I will, um, I'm not going to necessarily wait to the end. I'll pause maybe in a few minutes for questions. So if you've got something percolating in your, in your head, get ready to come up. And we do want people going to the into the microphone because they're recording this. And that way we can. We can hear your question, but but maybe you can share. And it was it was a fantastic presentation. I really, I was really impressed with what I saw that this this morning. Um, what types of agents I know you're building product within CDH. So what types of agents are you um, building or planning to build. Like what is being built now and what is being planned to build in CDH. Yeah. So I'd like to take a little step back.

I mean, I think most of you in the audience, if not all, have been building your own gen AI stuff. And it starts with creating a three headed cat, playing the drums, but then you become more serious and then you think, well, what, what can I do to make something that makes our work more seamless, maybe faster? Building your own agents and rolling them out to your to your organization, that typically leads to a bottleneck. And it's also something that came up in my dinner last week where somebody from, from a company called Clio Legal Software, and they use a lot of AI that we can we can develop like ten times faster, 20 times faster, but deploying. We cannot do faster. That was one thing. The second thing they mentioned is they needed legal experts. It was, as he said, a bunch of nerds in a room building this software, which is now highly popular for, for, for lawyers across the world.

But he said we didn't have the legal expertise, so that human expertise was still very much needed. Now, looking at what we're doing specifically for Customer Decision Hub, we built what's called Customer Engagement Studio and urge all of you to look at the innovation hub and see what that really is, because that is that amplification that I'm talking about. For CDH itself, it operates at runtime. It makes sure that the user can actually much faster create strategies to roll out to their user base, to their consumers. Remember, we're actually developing not necessarily for our customers, for their customers. We want to make sure they have the best customer experience and they get relevant actions fed to them. It could be sales, it could be information, it could be service, it could be anything. So the agents that we're building inside customer engagement studio are the ones that that make that strategy creation much faster and more seamless.

So you can roll out your engagement policies. Who should this action go to? When should it go to this person? Do we have certain gaps way more faster than in the past? That is one thing. The other thing is what we call a conversation agent, which is basically that initial touch base that we have on Customer Decision Hub where you upload your campaign brief. Now, I used to work in telco myself many years ago, and the campaign brief process is a long process. It's a nasty process in many ways. You have many, many different stakeholders from many organizational business units. So you have to look at pricing. You have to look at your creatives. You have to look at the rollout. You have to look at the audience that you target. We can do that now in a matter of an hour. If you will upload that campaign brief, we extract all that data and we create your strategies for you.

So the conversation agent is one of them. We're thinking of a research agent that can go out and do more research for that content that you're generating in your strategies. We're thinking of a creative agent, of course, that we can use our own internal creative creativity to, to generate your content and the treatments, as we call them. You can call out to an external agent. But again, all predictable AI, I think we need to emphasize that message, not roll out just numerous agents operating at random, but have them operate according to the workflow in CDH. Yeah, that's that's really exciting, right? The development of those agents. I have a question for you on that. So at Accenture, we one of the things we do is part of song. We have marketing agencies. And so the agents that you mentioned, an agent that creates the brief.

An agent, an agent that creates a. That does research. We're actually building similar agents for for a broad Marketing process for an agency process, right? What would those. Would we be able to have those agents communicate with the, with the Pega agents. We'd be able to integrate that. Is that. Yeah. As you saw this morning when Karim gave his demo and also when when Alan basically showed the the high level architecture of our entire AI stack, you could see that we have the Process Fabric for the orchestration. We have MCP servers support, we have a to a support agent to agent. So yes, we can communicate with external agents as well. Feeding that into the predictable AI in Customer engagement studio, which still operates according to a workflow underneath. Now that workflow is fully customizable to your needs, but it still follows those steps to make sure that you don't end up with with random actions that could go out.

And we've seen numerous examples of that happening. So yes is the answer. Conversely, you could, if somebody built a research agent, a broader research agent, they wouldn't maybe again, part of the buy versus build discussion, you may say, oh, you know what? I'm not going to use the Pega research agent. I'm going to, but I am going to use the Pega Constellation agent. And you can still mix and match kind of the agents to, to work together. Absolutely. Our organizations need to. Absolutely. We see that the model is basically recommending. But CDH is executing. And that is again, that real time component at scale across all channels. So they work together. They amplify each other. Fantastic. Let me let me pause to anybody have any questions for anyone. Please step up to the mic. Yeah, I have several questions, but I'm going to start with three.

So the first question is, um, yeah, you already built your um, process with a genetic AI. So, uh, what's the output now? The percentage that everything is automated. Is that a question for Bart? Um, who can answer that? I think it's more for you. Yeah, yeah, yeah. Can you repeat the last part? The the percentage from 1 to 100. Um, how much automation now is happening using your genetic AI. I'm sure you build a lot of genetic AI. Yeah. Yeah. So at the moment we don't have yet a lot of genetic AI. So we built a lot of GenAI agents and in my opinion, so we have a lot of humans in the loop because we are in a regulatory environment within the financial economic crime. And so a lot of humans in the loop control. Of course, what the agents are doing, but there will be a lot of automation in the background to get documents, for example, already provided to the analyst or to the agent, and then the outcome will always be checked by a human, and I believe in maybe a couple of years it will be maybe more humans.

We will look out of the system and then it becomes more authentic. And my second question is, um, did you, uh, how is your, um, database architecture? Looks like, um, did you expose everything to being used by the users and at the same time, your system automation or did you build several, you know, database or access to or that Agentic AI? Yeah, yeah. We use several database. Of course, we have Pega as a system, but also several external databases. And we build agents in between. And the user is interacting in this situation, for example, in the system with the chatbot. And that is going to the database or to agents in between. And later on, Maybe. Also, we removed the chatbot and then put it directly into the screen in Pega for the analyst that they don't have to ask questions anymore, but can directly get the data from the database.

But then a lot of agents in between get the answers already to their workflow. Yeah. And my last question is, are you in PegaCloud? Yes. Okay. That's good. Thank you. I always at Rabobank, we try to always run at the latest version minus one. And that gives us also all the flexibility that Pega provides. Thank you. One thing I want to jump on is your question around um architecture. I think that's an important one because with Pega, we do not pretend to own the data warehouse or the data lake, we integrate with that. In my previous role, I was responsible for the ecosystem for Customer Decision Hub. We're very open. And now with authentic, of course, we're making that more open with MCP and A to A and all these protocols to make sure to to be able to communicate to the outside, but specifically to databases, data warehouses, we integrate with the likes of Databricks, Databricks, snowflake, Google, BigQuery, you name it, because we realize that's also the reality that many people here face.

They're building upon legacies and legacies of, of architecture that is still around that we need to cater for. And we can't replace that all of a sudden. Now we do have the capability to move Legacy transformation to new Pega, as we saw this morning as well. But the reality is that many customers still have those stacks that are decades old. We saw that from as well when they were presenting this morning. And, you know, it's interesting, I think one of one of the bigger challenges in building Agentic solutions is actually figuring out the security around it, right? Because and we do, we do a lot of work around security. And a lot of that work is actually helping our clients security teams to get more comfortable with what they can turn on in an MCP gateway, because it's not just turn it on for everything, because once you turn it on, you can have access, access to things.

And so there's, there's both kind of an art and a science to how you enable that. But that's key in making the, the agents work. Um, any other questions? Otherwise we'll keep going. But there's one, there's one. Oh thank. You. Hi. My name is Arnaldo Garcia. I'm with Accenture. I work out of our federal practice in DC. My question is simple, but maybe numerical. So percentage wise, how much more efficient would you say you are through the use of AI through your agentic workflows? Yeah. At the moment we are because there's a lot of measuring going behind and it's sometimes difficult to measure. So what we saw is we do, of course, we do a lot of automation in parallel of GenAI. And the question is always, have we see a big reduction in our domain, at least if the number of analysts that are working on CBD, for example, or transaction monitoring.

But the question is always that that this was that because of we moved to Pega, for example, because there is already a big reduction. And also the automation that we are implementing every, let's say, Sprint. And what we are doing, we are measuring the success of GenAI. Also, for example, the knowledge assistant, we know exactly how much time people spend in the current work instruction system because we measure almost everything what we can, we measure it. So we see, for example, they are spending one hour in the work instruction system and we are measuring also the screen time that they are spending now on the knowledge assistant chatbot. And, and we can compare, of course, and we see there a drop in the usage of the work instruction system. But there again, you don't know because we let go a lot of junior analysts because of all the automations and also because the policies are maybe more clear or the work instructions are better.

So people are not using the work instructions anymore. So the question is, yeah. Did that come to GenAI or because we don't have that many junior analysts anymore, but more media and senior analysts. So it's very difficult to compare. Thank you. Just a quick follow up. So would you say that you now need more senior technical resources rather than junior legal resources. Because citizen developer it's typically somewhere in between the two. So would you say out of those three profiles, which one would you need the most? Yeah. So I was talking about the junior analysts that are working in our system as the end user and the senior analyst. And if you talk about the citizen developer. We also think with the with all the GenAI on the software delivery lifecycle, we don't need that much engineers anymore that really are building, for example, components or modules, but more the citizen developers that really understands the business needs and can.

With App Studio or Infinity studio can really translate that into a workflow that the business can use. And of course, we focus still on with a lot of engineers that are focusing on building that modules and the integrations. But before when we used, for example, cosmos, because we are almost completely over the Constellation right now, and then we had a lot of engineers that can build everything from scratch. You're welcome. The workforce question is an interesting one, of course, because there's a lot of angst, if you will, that that people are being laid off. And we do see those numbers. But at the same time, just last month, I think it was there was a report that there have not been as many developers employed in the US as that month. So that has only grown. I thought the number was like 2.3 million developers, which is interesting.

But at the same time you also see that, for example, the Forward Deployment Engineer PhD is also a new role that's popping up in my mind. It's kind of like a combination of a business technical consultant, but that knows AI and goes to customers to explain what to do with it. So it's a little bit of both. You see the large numbers, Oracle letting 30,000 people go, meta letting thousands of people go. And at the same time you then hear people are being hired back because people realize that the AI cannot replace. And same example that I gave before about Clio having to hire legal experts because the AI doesn't necessarily understand everything. You need that human nuance on top of it. Other other questions? If not, yeah. Go ahead. Hi, my name is Tarun Maheshwari. I'm one of the solution leads in New York Life Insurance.

So I wanted to understand. Can you double click on the the Knowledge Buddy thing that you talked about? Right. So what was your experience or challenges that you faced when you were building that rack model or the operational data structure? Is it possible to throw some light there on that one? How was the experience? How did you train the model? Did you use the, uh, data store inside Pega or did you use some external vector database? Yeah, we used the vector database within the Knowledge Buddy, and we get all the data refreshed every day. We retrieve it by an API. And the most difficult part was because we had an old, let's say, work instruction system, but there was no API available. So first we had to create an API on that system and to get all the data. So first we started with screen scraping with all the work instructions.

And then when we start using it, we saw, hey, the work instructions are not written for GenAI, so if you. For example, an example. What we had is. If you ask the question what to do when you. When a client deposit more than 300 €0 and in the work instructions, you saw a table that. It's more than 200 €0, then do this. Is it more than €50,000? It's this. We always get the first answer because the first row was already applicable in this situation. We had to do it all over again. And we said, you first have to read the whole page and then give the answer, etc. that kind of things. And we see it still. We had a user acceptance test, for example, last week, and we get a lot of negative feedback about knowledge assistant or Knowledge Buddy. We call it knowledge assistant within the bank.

And because they now compare it with copilot work, for example, copilot work has access to all the policy data, and the Knowledge Buddy has access to the work instructions. But that you can't compare because if you have the policies, yeah, human derive, let's say the work instructions out of it and they put it in the system. And that's maybe harder to recognize for GenAI than the policies and create the work instructions out of it. So it can also make a shift now that we maybe stop with work instructions. If we just load in all the policy data, we can get work instructions out of it automatically because the policy is what matters. Thank you. Thanks. You're welcome. Yeah. If you excuse me, if you think about it, it's we now have there's a lot of different tools now emerging in the marketplace that that you'd want to figure out.

How do you get these tools to work together to really drive value and drive, drive down your opex costs, but also drive, improve your net promoter scores and overall value for your business? And it's a combination of having Pega almost as a process harness right in controlling the workflow. And then you have to determine when do you have an agent that is going to run somewhat independently. Right. And then but there's also tools that are productivity tools for an individual like, like Microsoft copilot, right? Like Claude Cowork, right. Where essentially you shut down your laptop and Cowork shuts down, right. Agents aren't supposed to shut down. So it's, you know, you could say, is that an agent? Is that not an agent? I don't know. We could define the definition of it, but the point is there's lots of different tools, some tools that are going to go across processes, some tools that are going to make the worker the individual more productive at their desk.

And then combining that now with what, what Pega has to wrap, wrap that and manage an end to end process, especially in a, in a complex environment, I think is going to be really interesting and not to mention making sure that employees know how that all is all going to work, right? When do they when do they use which tool and how do they develop their own automations where it makes sense. So there's a lot, a lot now in play. Um, and it just keeps changing by, by the day. And I think also it's our role to decide what they know, not what they use because the most, the best thing should be that an analyst should not think about, oh, I have to go to this chatbot or this chatbot or this agent, etc., but that we decide based on the question, where do we need to route the question to and get the answer from?

Because what you now see is people are using copilot work and it's getting the answer from SharePoint, but they don't know if it's maybe the answer is getting from an archive folder because they can't see. Of course, there will be sources mentioned, but nobody will check the sources anymore. So we should take care of. We have regulated sources. So really, the policies that are current and and the question should be routed to that system. We don't have to just provide the analyst with 3 or 4 different chatbots, because then they don't know anymore where to find which answer. AM I curious? The the session this morning and all the advances that that Pega is making with Pega Infinity around the agents, what what did what did you think? What excites you the most? What do you want to double click on a bit more when you when you get back, what benefit the bank?

Yeah. One of the things, of course, what Karim was showing is about Blueprint. We also use Blueprint within the bank for at least every new application. It's very easy to use. You just type in something in Blueprint and you get all the application details already. But yeah, we are not building every day a new application within the bank. So then what you see in our department and Blueprint is used once and then we are start building it just on the normal way. So I think what Karim also showed, it's now integrated into Infinity studio. So it should be really nice if we can also use it for every new feature that we want to build on top of an existing application that we can use Blueprint or the AI assistant to really help us to build the applications faster. You're going to have your team start start working that.

Yeah, yeah yeah yeah. Other other questions from folks. Otherwise I have one for, for, for you, Ari. And when you look at, um, just the role that that Pega, maybe this is even more broader than CDH, but just the role that Pega is going to play in overall kind of orchestration. I know you touched upon that, but maybe talk because I think there's a lot there's a lot of confusion around that. Right? First of all, any every company you talk to, it says they do orchestration, right? I mean, it doesn't matter who. And so what? What do you see Pega.com role in the as as what is their role in orchestration? That's a good question, because to your point, every company wants to become that control tower and they want to own the, you know, the processing around the agents, making sure they don't just run around and do their stuff.

And Mart already highlighted a very good example. You know, if you tell the AI this is your source, but that source is not curated, if you will, it could come up with old stuff that shouldn't be there. The human has to be in the loop. So I do think that it's, it's something that every software vendor definitely has to cater for. We do that with Pega Agents. Process Fabric. Um, but the way we do that is still based on the workflow underneath. And it's not something new that we invented. This is what we've been doing for decades, right? With our, with our software. So it's there. And again, the AI amplifies what we're doing. So we play that role naturally, if you will. Um, and I think that's where we continue to go. Another point is the, you know, the insights that we generate. We didn't speak about this particular agent, but we we also roll out a data science agent, for example.

Now that is specific to Customer Decision Hub, but it will help those data science employees that are very, very specific in their, in their knowledge to get more understanding of what happens behind the scenes. In my mind, that's part of that control plane as well. You want to make sure that people understand things. And I always say it like this. I want to know what I don't know. That's where the AI can help me. And that is a good example with the data science agent that we're rolling out. So again, in my mind, we're playing that role naturally. We're amplifying that with the agents that we roll out and we play that role of control plane. We don't necessarily have to own that in the market as being the control tower. But for our software, I always compare it to like a beehive, like you need a beekeeper, you know, because you want to ensure that those bees, they do their work and they are controlled and still lead to the honey that you that you want.

Yeah, I know, and I think that's pragmatic, right? It's it's hey, where, where we are, where Pega is good. We want to, we want we'll be the control plane in that area, especially for, for companies that are investing in that space, but you're not out to be the control plane for absolutely everything that's out there. And I mean, the other thing that I, I don't know if folks probably know this, if you're, if you're developing agents and why you even need a control plane, right? But if you think about it, an agent, it's at its basics. It's a markdown file, right? Um, and if you don't have a control plane and there are some places that everybody's kind of going out and just coding, coding agents, building, building agents, right? Which then creates a bit of tech debt on steroids after a while. Right? So having a way to make sure that your, you're controlling that, that agents aren't drifting, right?

Because if you kind of leave them now, it's still early days, right? Because some folks are. What do you mean by drift? I've had an agent running for for a month already. It's not. It's not drifting. Well, actually, over over two months. Three months. It can drift and it'll start because it'll, it'll just remember things that are, that are, that are older, right. And it won't necessarily, it'll get less and less accurate in what it, what it needs to do. You also have a lot of IP within your agent. And how do you how do you how do you protect that? How do you make sure that the agents doing doing the right thing. So all of that is, is important. How do you make sure we didn't even talk about Tokenomics yet? But then we got we're going to finish up. But how do you make sure the agents using the right models at the right time?

You know, when opus 4.8 came out, it was like, oh, this is great. I could do wonderful things. Well, you don't you don't need it just to respond to an email and you're crunching tokens. If you do that versus using, you know, a more basic model. So I think we have we've got a couple of minutes left. I do want to I want to have one fun question that I'm going to I'm going to ask out there. It's more looking, looking into the future. And it's about a topic. It's related to identity, but it goes well beyond that. So if folks have heard of this concept of RSI, if you're playing in an AI space, what RSI is, is essentially it's recursive. It's called recursive self-improvement. So it's the idea that models will eventually could eventually create the next model, right? So every model to date has been built by human right.

Mythos built. Everything's been built. GPT five five built by a human right. The next release, six, is going to be built by a human. But what the AI companies are saying is that that we're we're evolving. We're getting to a point where a model can actually create the next model, right? Potentially. So just throwing that out there. How many how many think that will happen in the next 12 months? My hand went up immediately. Not a lot of people. So most people don't think that will happen in 12 months. How about 20? Will it? Will it ever happen? How many think it will ever. It will ever happen. Okay, so most people think it will ever happen. Just not not not in 12 months. Um, what do you think, Ari? And do you think it will? Yeah, it's a funny acronym. And it reminds me of what Mark was saying.

Of fact, we had one too. In in Pega, we have a strategy optimizer bundle of which the acronym was SOB. So that didn't really go. Well in the. Market either. Um, but RSI is in repetitive strain injury. That's the first thing I thought of. Um will it ever happen? No doubt. Definitely. But it depends on how we as humans feel about that. And there's so much exciting stuff happening around this. It's it's a philosophical debate. It's not a technological debate. And the fact that the Vatican is involved in this. The fact that that anthropic goes there, um, you know, it tells you how big the impact may be for humanity. I'm not a futurist, so I don't know if it will actually happen. I do think it will happen, but it will probably be contained because people, you know, in the end, there's also ego.

And I think the human wants to control that in the end. Yeah. Well, we, uh, maybe, maybe a year from now, none of us will will be here because it will just be agents attending instead on our behalf. Um, I think we're, we're pretty much at time. I'm going to wrap up. Like I said, if you're interested in getting a little deeper. And we did go pretty deep in the in the white paper, feel free to download it. Take a look and we can have discussion after. Thank you so much. You guys have been great.