PegaWorld iNspire 2024: How AI and Automation will change the way we work – featuring Forrester VP & Principal Analyst Mike Gualtieri

Hello everyone. Welcome to this session. Thank you all for coming. I'll tell you a little bit about myself. I am Mike Gualtieri I'm an industry analyst at Forrester. So that means I do market research on various types of platforms in the market. So my particular coverage is on focuses on AI. And I've been covering that for quite some time at Forrester. And of course it's a very hot topic now.

Um, and the best way to describe how I research the market is by the Forrester waves that I do and Forrester wave is a is a is a deep evaluation of vendors in a particular marketplace. So I do AI infrastructure. AI infrastructure is the hardware. So you can imagine companies like Nvidia, the hyperscalers, Dell's Hpe's, the hardware to run certain AI workloads. I do AI platforms. So these are the tools and platforms that data scientists might use to build models. I do AI foundation model waves. It's our inaugural wave. We just published it last week.

Your OpenAI's, your coherence, your anthropic's. Um, but my favorite wave and the one I'm here to talk to you about is what we call AI decisioning and AI decisioning automation. So I'm going to talk a little bit about that. I'm going to talk a lot about that wave. Um, and especially why it's my, my favorite. Um, but the backdrop of this is my research on enterprise AI, because my particular focus is on how AI helps helps enterprises. And the state of enterprise AI is very it's become strategic. So at Forrester we started tracking about 2016, we started asking companies, to what extent are you using AI? And that's pre generative AI, right.

2016 and it was about 42%. And then fast forward uh now we've got about 88% of companies. World's largest companies saying they're doing some AI. It's going to be 100% and some may not do it themselves, but they're going to buy it in in platforms they use or in business software. Right. Because every ISV is trying to figure out how they can incorporate this into their platforms as well. Um, and we have a very simple definition, uh, that I've been using for many years. And AI is software because that's a big reminder here. It's not different from software.

It is software, it's a different type of software, but it's still a software, right? Um, and it strives to mimic and sometimes exceed in certain domains human intelligence. So yeah, I mean, maybe it can predict churn because it can look through a million records and look at all the different factors, right. So that in that particular case, it could exceed human intelligence. In other cases, it's complimentary. It's also and this is, uh, because of deep learning initially for computer vision. Now for GenAI. It's the fastest growing workload on the planet. So everyone knows, uh, about Nvidia GPUs.

And that's all got triggered in 2012. 2012 was the breakthrough where computer scientists figured out, you know what, I can use GPUs to run these workloads. Um, and so, um, it's it's the fastest growing workload on the planet. And we define those workloads as data prep training, workload training, the model and inferencing, the the workload. And we also think of AI in two ways. We divide it into two. Now I think I heard Kerim on the stage talk about Statistical AI. Um, now, some people would call that traditional or classical AI. I don't want to call it traditional or classical because it almost it almost makes it secondary.

No, we call it predictive AI. So some people call it statistical AI or predictive AI. And that's the more I don't want to say it, but well, I keep saying it. Traditional AI um, but both of these are very complimentary predictive and generative AI. Um, and you can see a simple example of why that's the case, right. So predictive AI, you would use that to predict, say uh, customers like whether or not a customer is likely to churn. And then you can do. Finally you can do the hyper personalization, the marketing of one right to use generative AI to automatically generate a treatment or a message that might be persuasive enough to prevent that customer from churning. Um, our perspective on AI is that it's going to assist workers, leading to soaring productivity.

The number one use case. I'm just giving you kind of a general overview of the trends right now. I'm going to drop right into AI decisioning pretty pretty soon. But the but the number one use case that we're hearing from enterprises is actually coding. Coding is using it for development. Coding assistant. You saw some of the tools for that uh, which actually extends and abstracts even above coding, uh, to actual business processes. Um, but it's influenced all of these roles and many, many more. Um, and it's also going to drive, uh, incredible amounts of process re-engineering, automation, I guess we can call it transformation as well.

Why? Well, the reason is, is because most of the use cases you hear about AI, the amazing use cases you hear about AI usually are only about one little step in a process, and it doesn't mean that that step isn't valuable. It doesn't mean it doesn't save $10 million a year or $100,000 a week. It doesn't mean that it's not consequential. It is. But business processes are designed to reduce friction. They're supposed to be and they're designed to reduce the number of bottlenecks and to mitigate the effects of bottlenecks. So what happens when you have a process and you have a bottleneck, and then you take AI and you squash it down? Well, that's no longer your bottleneck.

It's something else. So after you keep implementing these single use cases, these single step productivity, all of a sudden you step back and you say, wait a minute, we need to redesign that process now because we wouldn't have designed it that way if if these if the bottlenecks look different and now they look different. So this is that's the argument for why we think there's going to be a lot of transformation driven by AI. Now on to Onto AI Decisioning Lead. So I've been covering AI Decisioning Lead what we now call AI Decisioning Lead for 16 years at Forrester. It was the first Forrester wave I did, and at that time we called these business rules management systems. Right? So in the Pega Platform, I believe that's still called Pega rules, right? But it's a lot more than that.

Um, so in about 2018, we called, we started calling this digital decisioning platforms. Uh, and then finally in 2023, we started calling it AI decisioning platforms. And the reason for that is if we go back to 1985, the, um, the origin of these systems was something called expert systems, and that was the focus of AI research at the time. Uh, and expert systems would extract knowledge from experts and essentially put them in rules. And then an inferencing engine would use those rules to, to make decisions, come up with answers. Um, so that was sort of the AI of the time, right. Pre machine learning. Well full circle we have machine learning now. But we still have the fundamental technology of expert systems business rules.

So the platforms that I cover have all evolved Pega included to include not only. And I'm not going to call it rules anymore, I'm now going to call it human decision logic to include human decision logic combined with machine learning. Okay, this is why it's my favorite thing to research and why it's my favorite market platform because it is human controlled AI. The human control is the human decision logic, combined with one or more machine learning models. And the human decision logic is any arbitrary set of rules and processes, right? So a platform that includes that allows you to automate full business processes include human decision logic, which could be policy rules, governance with the goodness of machine learning models, whether they be predictive models or generative models. So, um, when at Forrester, when we do an evaluation like this of the market, um, how many are familiar with the Forrester wave just in general? Okay. Um, I'm going to talk about it a little bit more, but but I'm glad many of you are already familiar with it, but we have to put a business lens on it because I don't come up with this stuff myself.

I'm talking to our clients, so I am trying to channel what's important to them with these platforms. Um, and AI decisioning. And so AI decisioning is about those decisions and the enterprises, um, uh, the success of an enterprise is based upon those decisions, and especially when they're automated at scale. Right. And who makes those decisions? Leaders make them. They make strategic decisions. We're going to open up in this market. We're going to sell off this division.

We're going to buy this software. So there's leaders and employees that make human decisions. But increasingly there's decision intelligence that are embedded in applications. And that is our focus. Our focus is on the automated decisions which will help automate those processes. So the automation side of that is to gain scale, efficiency, agility. And when I think efficiency, I think reducing bottlenecks, when I think scale, I think of, um, uh, any number of decisions that can be made and automated automatically and the agility is the ability to and I think this is sort of, uh, someone can correct me. Pega's origin statement is the ability to change or design for change or some phrase like that. And that's what agility means to me.

And that comes from those well-designed processes. And right now with AI, people aren't changing those processes per se. For the most part, like I said, they're plugging in AI to make one step in the process more efficient or more amazing. But they're going to start, uh, to, to to redesign those processes to mitigate what now has become the bottlenecks. And those processes are now going to have the benefit or now have the benefit of AI and data. So that is the key focus of the AI decisioning wave. And why do we call it AI decisioning wave. For everything I just said. But it's still the decisioning wave because it also has the human controlled decision logic.

And when I say human controlled, it's because we are putting in the rules that can govern. So I've got companies. Look, I cover AI ML platforms too, right? So think of all the companies that data scientists use to create a model. Um, but companies ask me, oh, what platform should we use? I said, you know what, you should look at an AI decisioning platform because it's going to give you the capability to bring in outside models, create your own models, but also govern it with the policies of your company and the rules of your company. Um, but to be successful with AI, um, you do need data. You need to customize it to your to your organization, to your customers, and even the foundation models. Right.

There's various techniques for that. And you know what? It was unheard of. You would never say, oh, I'm going to train my own. Uh LLM. Uh, companies are training their own llms now. Uh, but the good thing is you don't have to. There's other techniques that allow you to use your data like rag, if anyone's heard of that. Um.

Stands for retrieval, augmented generation. Um, but the problem with most of the organizations is that they have a portfolio of hundreds, if not thousands of Applications. In past life, I've worked at banks and insurance companies that had a portfolio of thousands of applications. Acquisitions, you know, Bank of America, Liberty Mutual, companies like that. Um, and so you've got a data estate. That's just crazy, right? So you also have to get that you have to have a platform that gets that data together, and then something that can, uh, fit over it, um, for the AI. Um, and the other problem here is that time these systems weren't designed in a time when business was more real time, and all data originates in real time. Companies tell me, oh, I don't have real time data.

Yes you do. Every single click happened in an instant. Happened in real time. And the problem with that is the time to decide is shrinking. And many of the platforms people have, uh, lead to an improper action because they don't have the right data. Um, and it's taking too long to decide. The nice thing about AI is that you can embed that decision logic in your process and decide instantly, right? So we designed this, uh uh um, we have this concept called perishable Insights and decisions, right? Where there's a time to act and there's the perishability of those insights.

So we're not saying that data. Uh, uh, has a shelf life because it doesn't. That's how you do machine learning. You train it on older data to find patterns. But but insights often do. And you can see the example here, right, of you're shopping for an automobile in real time online. You want to recommend a loan. Um, there's an operational decision. Well, we're not making the profit we want.

We're going to optimize our rate or change our rate. Um, we're going to make a performance decision. Demand forecasts are strong. increase the lending pool and then a strategic decision. So most companies have invested in this, uh, I'll say the squishy middle right where we're going to make decisions later on with operational decisions, performance decisions with BI tools, you still need them. You still need to make those decisions. But increasingly, we have to make those real time decisions to where do you make a real time decision? You make it in a real time process. Um, and this is just another example, uh, for, uh, a Healthcare example, you all have dozens of processes that that could fit along this line.

So companies compete on fast decisions. So when we look at the AI decisioning wave that I promise I'm about to talk about, um, uh, the focus is on how fast those decisions can be made and how fast those the decision logic can be changed, uh, whether it's, uh, uh, Retraining a machine learning model on newer data, or changing the human decision logic. So every company is going to do this. And the question is how are they going to do it? Well, they're going to build and buy, because I keep reminding myself and everyone else that AI is software. And what do you do with software you build or you buy it? It's kind of the same decision point, right? And you've all been through it before. It's like, oh should we.

Oh yeah. We're going to build this. And then six months later, you know, a vendor comes up with that feature. Oh, we shouldn't have done that. So we encourage all of our clients look at your vendor roadmaps too, because you may not want to customize this because it may be coming. Um, but there is a, uh, a need to build and to build quickly. And so we define an AI decisioning platform more succinctly is software that provides enterprise, business and technology teams with the tools to author, automate, and and make business decisions in a wide variety of applications, leveraging many decision intelligence technologies. And that's the key many decision intelligence technologies. So again, in my AI ML platform wave, the only thing we look at is machine learning models, right.

But in this wave we look at how can we create human decision logic rules? How can we create. How can we use optimization math? How can we use machine learning models both predictive and AI? So an AI decisioning platform is kind of an overarching way to do AI for an organization. Um, and so this is just sort of a diagram that kind of succinctly puts that. So the decision, uh, use cases exist across, you know, any different operational business process, uh, you know, all the business processes you have. Um, and in fact, this used to be, um, the key question that I used to get asked by clients, what are the use cases and especially, you know, with, with, uh, with predictive AI and now with gen AI. But, you know, when you're an analyst, you see these waves of questions where people are asking, what are the use cases?

And then it switches to how do I do it? Because it falls like dominoes. You see, one vendor in a certain, uh, industry say, I don't know, banking. And they start doing a use case, and then the other one say, wait a minute. If they're doing it, I can do it. And so we've seen that now with, um, uh, with AI use cases. And it's very common for me to hear about a single use cases. And you've already heard many of those at the conference already. So the the, the the essential capabilities of an AI decisioning platform is it keeps humans, human experts in control.

And it does that with having the decision logic. And it can combine a broad number of intelligence technologies. I'm so desperately want to mention a vendor name in my AIML platform, wave, so that I can tell you a vendor that doesn't do this. Um, but I'm hesitating to do that. But think of, uh, machine learning. Well, I can say a hyperscaler. I can say like something like SageMaker at AWS, which is a great tool for creating machine learning models, but it doesn't offer a broad number of decision intelligence technology. It doesn't let you use human decision logic. Right.

Um, and it has to support rapid learning loops, um, as well, not just for the business process, but for the machine learning models. And we also look for industry specific solution accelerators. Um. Again, this is because we hear we hear this. So the way we think of this Platform and analogously is to um, well, this is a so I guess there's competitions among corn farmers. Um, where how fast can they hand-pick corn? And the champion, um, at least the, the most recent champion that I found last year, uh, could pick 480 ears of corn per hour. So is that a lot? I guess that's, like, more than one per second.

How do you do that and where do you put it? But anyway, that's it's 480 ears of corn per hour. But the combine right. This this piece of equipment, uh, can do 800,000, uh, per hour. So that's a huge difference, right. And that's what we think of AI decisioning platforms Enabling, uh, for enterprise decisions and automating those decisions. So when I do the wave, what I do is I come up with criteria. For those of you who don't know what a wave is, it's like Consumer Reports, except it's, you know, it's way more sophisticated. Um, because, uh, we come up with a lot of criteria.

We may have 30 or 40 criteria that we use to evaluate this particular product. So you don't have a few hours. So I've just zeroed this down to seven of those criteria. Uh, so the first one is connecting to data sources. Um, obviously you have to be able to connect to data sources. So this is the one that customers ask about, uh, a lot. But, you know, frankly, it's the easiest one. And most vendors have that. Um, but what many vendors don't have is a way to track the features or the signals, right?

Because if are there any data scientists in the house? Okay. There's a couple. So all of the data scientists, when I say a feature, they know what I'm talking about. But many of you probably don't know what I mean by a feature. It's a variable. It's just a it's a it's a it's a variable. It's a piece of data that's used in a machine learning model. So what we also look for is can they store and track those signals in that catalog.

Um, which is very important. So many people think about the algorithms, but the algorithms are mostly open source, mostly commoditized, widely available. Right. So it that's why we call out the data here. Because that's that's the critical piece here. Um, so we also look at what intelligence capabilities. So in the AI platform wave We don't really look at this because we're only looking at one thing which is which is the ability to create a machine learning model. Um, but in the decisioning platforms, we need statistical and queryable analytics because you're not just making a decision with a machine learning model. You might be using a query, you might be using some combination of more complex combination.

We look, does it have some way of plugging in? Um, what you may know as operations research or um, also known as constraint based optimization, also known as mixed integer program. Anyone know mixed integer programing? Probably the data scientists do. Okay. So the machine learning models can be trained and also imported. Right. So can the Platform train models. Yes it can, but can it also import models.

And especially in the world of gen AI, can it use models, gen AI models? And of course it can because the GenAI models are are available through APIs. But that's kind of messy, right? Do they have interfaces that let you bring those models in easily? Do they have prompt management studios? So we look at all of that stuff. Um, and then finally, what was the core of AI decisioning when we called it business rules management decision and digital decisioning. Is that human decision logic? And if you use Pega rules or any of the other vendors, you know exactly what I'm talking about there.

These also have to be no code okay. They have to have code, but they also have to have no code. Um, and I and I don't like the term low code, so I don't use it. Um, but I just did. So decision logic tools are designed for business experts who understand, uh, the processes and can change them, but they also have to be designed for technical people, too, because you have to plug into those hundreds of different applications. We look for productivity tools, so your data scientist may want to do something in a Jupyter notebook. Write Python code, but there may be some productivity tools that help them evaluate that model. So. So we look at what productivity tools exist for these other roles as well.

Um, and of course, uh, orchestration is a key point. See, when you take any of the AI, ML platforms, the Datarobot or Dataiku, a SageMaker, uh, and any of the others that cover there, the output is a model, and a model takes an input and it does an output. That's simple. That's too simple. It can be very valuable, but it's too simple. So how do people use that? Oh, you know how they use it because they take that model and they write all the Java or Python around it to do all the orchestration stuff. What a mess. Um, AI decisioning has the orchestration built in?

Dead. And that is the key thing. It allows enterprise to compete on this decision. Um. Uh, agility. Um, so trust is an important aspect of this to many of the clients, maybe most of them. Yeah, I'd say most of them who, uh, consult with me, uh, do inquiries on AI decisioning are in highly regulated industries. So. They need tools within the platform that can explain these complex decisions and not just the model.

Right. But explain the entire process in ways that they can understand and and explain the rules that are used on top of the the multiple models that they might use. So we look for that as well. And if you are building for change, which you are, um, you need to be able to test out the impacts of that potential change, too. So we look for, um, I guess we're calling it business simulation. But here's the thing about, well, you all know this about models, right? They're the results. Are often probabilistic or unexpected, right. So a best practice in using models is sometimes doing a B testing.

So you deploy multiple models out there at once. And then you, you you use them uh, like you use model A 10% of the traffic and model B 90% of the traffic or use CHAMPVA challenger. So these are critical operational techniques, um, uh, to uh, to actually put stuff out in production, uh, quicker than you otherwise would. Um, governance, uh, in the deployment of these models, which includes monitoring the AI and also guard models. So a guard model is another technique. It's like, well, we've got AI in here. We're not quite sure. So let's have a guard model. It's kind of like an oversight.

It's like it looks at the model. And did it just say that. Did it just do that. Did it just recommend that. So when a company wants to take advantage of AI, but they're also concerned with some of the uncertainty that that may occur. They may create guard models on top of that. So so we look for that as well. Um, and then on all of my waves, I look for, um, the architectural qualities, uh, some of the more technical qualities of scalability is this fault tolerant, uh, when it goes down, how will it come back up? Um, is it low latency?

High throughput? I mean, you've all used ChatGPT, probably. And you know how the output is very jittery coming out. Uh, that's too slow for many applications. Right? So, um, and then when you start combining. Other decision logic that needs variables from the mainframe, uh, you know, this becomes. Critical factor as well. So, um, all of these are used to automate a decision.

So these platforms are used to automate a decision in a platform. And some people say, well, okay, uh, a lot of platforms have decisioning built in. Why why would you need that. Why would you need that separately? Why would you need that centralized? Well, there's many reasons for that. I'm just going to give you one. Um, one is because sometimes that decision get made multiple times. So if you take the in different systems.

So if you take the case of, uh, KYC, know your customer, uh, which is something, uh, banks have to look up in a transaction. You may have to make that in seven wealth management systems and a retail bank system all over the place, right? So if you have one decision service that can make that decision centrally, that's another reason to to do that. So this is why AI decisioning is my favorite overarching platform, because it lets you combine AI with human experts, the human experts, in the form of human decision logic, which in 1985 people called expert systems. So, um, our recommendations, uh, focus automation on the decisions. Right? So people say, oh, how can I find AI use cases? Well, look at your business processes and look where decisions are already automated, perhaps in a more brittle way. Um, and then focus on how you can automate those decisions, uh, in that process and that and, and, and maybe, maybe you're already doing that, but not with AI.

So could I use AI in there to maybe make this an even better decision? Um, implement this with a decision, uh, an AI decisioning platform. Uh, because this is not a well-known category. So if your AI people are a team of data scientists who are super expert and they're using Jupyter notebooks and they're doing stuff with TensorFlow and PyTorch and scikit learn, and they're doing all this stuff, and they know a lot about creating models. Um, they may not know a lot about implementing software for an enterprise. So what's really a good thing is that my conversations have shifted in the last five years from chief data scientist. I still have chief data scientist who talked to me, but now it's a lot of enterprise architects, solution architects, uh, CTOs, CIOs was because now that AI is strategic, now they want to be involved in the Platform and many of those people have more of a software delivery and application delivery background. So they're able to recognize the value of this. And once you do that, then you can start competing on decisions, because now you have AI powered decisions controlled by human decision logic.

So with that I will take some questions. Thank you. Thank you. Thanks. Sir, I think you had a oh. Does anyone have some questions? This gentleman has one. Okay. I'm bringing it over anyway.

CDH capabilities. The difference between. Decisioning versus CDH capabilities. I didn't hear the second. The CDH. CDH Customer Decision Hub. O. Customer Decision Hub. O.

Well, I think that's a a a uh, well, let a Pega person, but I believe that that is, that uses the AI decisioning technology, but it but has additional things focusing on customer. Is that true? That's true. Yeah. Yep. Anyone else have a question? So you're getting all of that that stuff. It's just specifically focused on on customer decisions. Hello.

Hi. Firstly thanks for explaining everything to everyone. And it's really helpful. Uh, during your session, you were explaining that we need to do it overall a strategic way of implementing AI across processes. But there are multiple factors in a lot of organizations Organization budget. Uh, business outcomes. That is priority and stuff. So usually doing it like boiling the ocean is not an answer. And that's the reason they start small.

If someone wants to start small what is your recommendation how they should go about it. Yeah. So I think I think I'm observing that people are starting small because the scope of AI is very narrow, right? It's like, oh, I can predict that this customer churned or I can output an email, right. So the scope is very narrow. So, so people are starting small and that they're finding one step in a process and they're applying it to that one step. So what I was saying about the overall is that at some point after you do those, those baby steps, then you're going to see that all of these little baby steps have actually made all of these things more efficient. Maybe I can now change my business processes so it's not even using AI. It's just doing.

It's just it's just re-engineering your business processes. So I agree that you would you would start with these small steps. And the way people find these use cases is they walk through their business processes and they just say, is there something I can predict here? Is there something I am predicting? But maybe I'm not using machine learning model, maybe I could use a better one, or maybe I could add some decision logic. So. So that's what I was saying. Thanks for your question. Any other questions?

There's one over there. Oh, one over here, one over there. All right, you're closer. You get to go first. Yeah. So on the previous slide, you said, um, you need to first look at the, Um, yeah, the focus on decisions. But on the other hand, uh, you also mentioned that there's huge productivity gains on the, the development side and that that's. Yeah, maybe that's deciding what type of code you do. But how do you see that.

Yeah. Yeah. So, so on the on the development side, I was speaking specifically to um, well at least here at PegaWorld, it's Blueprint, but there's also and I think Kerim was talking about the code that he was making a comment about the code assistance. You know, how you can generate more code. He said, well, we don't think you should generate more code because you're just going to proliferate more code, which but that's what I am actually seeing in the AI decisioning platform vendors. I've seen it in Pega and I've seen it in the other vendors I've, I cover that everyone's trying to figure out how can I use AI to, uh, generate these higher level artifacts, right? Not generate millions of lines of more code, Having said that, there are coders, right? And they're getting organizations who are initially reticent to let them use things like Copilot or star Coder or any of the other is a bunch of open source code assistance. They were initially reticent to use that for any number of reasons.

Now it's the number one use case, and they're claiming 30% productivity gains on average for the coding task. A developer does more than that. Um, I think that the value of something like a Blueprint, which is much higher order use of this, um, is that your. If you're so the code assist is generating a snippet of code in a very narrow scope where something like Blueprint is at a much higher level, uh, designing a process. So I believe that's where it's all going to go as well. We're not going to be just generating more code. We're going to be generating the art of the higher order artifacts that abstract the complexity of the code. And I think that's what Pega is doing in, uh Blueprint. Thank you.

And now I'll make my way over here. And also, if anyone has a question too, you can go up to one of the mics as well. But, sir, I'm coming all the way over to you, so don't worry. Alright. Can I just squeeze in here? Excuse me. There you go. All right. So you mentioned that you were studying, uh, AI for quite a while already, and this is mostly about decisioning, uh, AI.

But what, uh, did you find as recent developments that are really promising new developments? I'm thinking of thinking like multi-agent, uh, AI or first principles AI, if that means anything to you. Yeah. Well. So we're actually developing this concept at Forrester. We call AGI for business. So we're we don't have to report out, but we're weighing in on AGI because people are saying, what about AGI because OpenAI's mission, like if you read their mission statement, it's like our mission is to create AGI, artificial general intelligence. The problem with that is that people can't agree on actually what that means. Um, and so it goes everywhere from superintelligence where, you know, it just rules us like some Star Trek episode somewhere, uh, down to, uh, um, AGI where it can, it can do everything.

But we're defining AGI business as very task. AGI for business is very task specific. So I'm, I'm interested in what are the barriers to achieving AGI. And I think there's a lot because If you look at the the neural networks that power these transformer networks, the connections are like, if you think of the brain, there's chemical and electrical electrical connections in many different types. And this is just one simple weight. Maybe it's not simple. It looks simple to me. Um, so can we actually scale llms to the point where they'll achieve this? I don't know, without some entirely new model architecture.

So that's one. And then the second thing is the concept of agentic AI. Maybe that's what you were talking about as well, which is an agent, but it's one that isn't reactive or responsive. It's one that follows a goal like, oh, uh, I've got this agent. Can it be a product marketing manager? Okay, here's your goals to increase revenue or market share or to be more competitive in this product. Can it carry on those tasks and interact digitally with the systems and colleagues, uh, to achieve that? Um, the way I think about that is I think about the self-driving car, because that's agentic if it works, right, because you're just giving it a destination and it's finding all the ways to do it. So I think a milestone would be if that actually can work on a large scale.

Um, but if you think about that, it has all these different technologies inside. It's not just one model, one AI. So I don't know, I guess we could talk for hours on this, so I'll stop there. But thank you for the question. Yeah. We have time for about one more question if anyone has one. I saw a hand right away. Oh. Thank you.

Hi. Thanks for sharing your views. Uh, how about security? You talked about the security. Security and data privacy, right? You talked about the seven aspects, uh, trust, uh, governance, etc., right? Yeah. But you didn't say anything about security or data privacy, which in a regulated environment, say healthcare or financial services. What would be your recommendation?

And how should that option be in those environments? Looking at data privacy and security. So there's many different ways to answer that. The one way to answer it is, is to remind that AI is software, right. So as a baseline, all of the data security practices that you would do still apply, right. But the second way to answer that is, well, what are the unique threats posed by a when you when you when you when you're using a machine learning model in this. And I guess the one that, that I think about a lot is data poisoning. Right? Data poisoning is where you can because models get retrained.

So you can start giving the model some data to make it think like there's a pattern so that when they retrain it, it's making a decision the way you want it, the way you want it made. So There's threats like that that are unique to machine learning models, um, that definitely are have to be addressed. Um, we don't look at those in the AI decisioning because it's a developing area. Um, and there's a bunch of startups that are popping up to address those. Uh, so for the purposes of our wave, we look at how the model is trained and the data security leading up to production of the model. But we rely we evaluate the security principles built into the platform to apply to that model, too. So that's how we look at it currently. Awesome. Well, thank you so much for taking the time today.

Thank you everyone. Appreciate it. Thank you.