AI Agent Builder Summit, Encore Presentation | Michael Garas, IBM & Stuart Frost, Geminos Software

Clips
More from AI Agent Builder Summit, Encore Presentation

Michael Garas

AI Partnerships, IBM Ecosystem

IBM

Stuart Frost

Founder & CEO

Geminos Software

play_circle_outline AI Agent Builder Summit objectives: future of AI, democratizing AI agents

play_circle_outline Partnership between IBM and Geminos for next-generation AI capabilities

play_circle_outline Limitations of LLMs for complex decisions

play_circle_outline The Critical Role of Causal AI and Knowledge Graphs in Data-Driven Decision Making

play_circle_outline Importance of focusing on ROI and business value in AI projects

Info
Transcript

Michael Garas, IBM & Stuart Frost, Geminos Software

Michael Garas

AI Partnerships, IBM Ecosystem IBM

Stuart Frost

Founder & CEO Geminos Software

Stuart Frost, founder and chief executive officer of Geminos Software, and Michael Garas, AI partnerships leader at IBM Corp., join theCUBE’s Scott Hebner at the AI Agent Builder Summit to examine the rise of decision intelligence in enterprise AI. Their conversation explores how agentic systems are evolving beyond task execution to support strategic, real-time decision-making at scale.

Frost shares how integrating causal AI with large language models unlocks deeper reasoning and context-aware responses, marking a critical leap in enterprise AI maturit... Read more

explore Keep Exploring

What is the objective of partnering with industry pioneers in exploring the latest developments shaping the future of AI? add

What are the challenges inherent in today's AI, how will the future be different, and why does AI decision intelligence promise to become a total game changer? add

What are the limitations of Language Model Models (LLMs) in decision-making processes and how can causal AI be used to address those limitations? add

What is the significance of the breakthrough platform that can combine the dynamics of a situation with a static underpinning, built in a reasonable timeframe at scale using LLMs and agents, and utilizing causal AI to understand why something is happening and what to do about it? add

What is the best recommendation for startups and large organizations when it comes to implementing AI technology? add

bolt Powered by CUBE AI

Michael Garas, IBM & Stuart Frost, Geminos Software

search

Scott Hebner

>> Hello. Scott Hebner here. I'm the principal analyst for AI at SiliconANGLE Media and theCUBE Research. I appreciate you tuning into this session. It's one of 12 that make up theCUBE Research's AI Agent Builder Summit. Our objective is to partner with industry pioneers to explore the latest developments shaping the future of AI. In this session, we'll hear from IBM and Geminos Software about how they are democratizing the ability to build a new generation of AI agents that are capable of helping us make better decisions and solve more complex problems. These next generation agents promise to help us understand not only what to do, but also how to do it and why certain actions are better than others, and to do it with full trust and transparency. This is not a trivial task. Why? Think about it. Simply put, enabling AI with new decision-making capabilities is super complex. It requires AI to apply knowledge to make judgments. And as we all know, knowledge is gained by comprehending truths and consequences obtained through experience and reasoning about changing conditions and how they may change outcomes. We then apply knowledge to make judgments by evaluating the causal relationships among actions, conditions, events, and consequences. And it is these judgments that enable us and AI agents, for that matter, to make actual decisions. Due to their correlational designs, today's generative AI and LLM solutions are simply insufficient alone to make this happen. Data is not knowledge. Predictions are not judgments. And correlation does not imply causation. To enable AI decision intelligence, an extended ecosystem of specialized AI models are necessary, architected to dynamically collaborate with generative AI services and LLMs. The good news? Industry pioneers are solving these difficult changes, providing us with a vision into the next frontier of AI. To tell us more about the challenges inherent in today's AI, how the future is going to be very different, and why AI decision intelligence promises to become a total game changer, I'm thrilled to have Stuart Frost, the founder and CEO of Geminos Software, and the IBM partnership leader at IBM, Michael Garas, join us today. We'll discuss all of this and the partnership that IBM and Geminos have built to scale next-generation AI capabilities across the enterprise. Welcome, Stuart, Michael. Thanks so much for being here today. This ought to be a really fascinating discussion.

Stuart Frost

>> Absolutely, Scott. Thanks for having us.

Michael Garas

>> Thanks for having us, Scott. Looking forward to the discussion.

Scott Hebner

>> Yeah. It's really getting interesting because all the conversations that I'm seeing out there talk about agentic AI and agents and how they wire them together. And there's always that subtext about, well, they can help you solve goals, they can help you make decisions, and they can help you problem solve, right? But no one's really going into the details, at least from what I've seen, of how you actually do that. And so that's why I think this is going to be a fascinating discussion because clearly, we want to move from task-oriented assistance into decision-making agents. And I still think there's a lot of work to be done under the waterline of the iceberg, if you will, right? Everyone sees the agents, but not all the guts and intelligence underneath the covers. So should be interesting. So Michael, let's start with you. From a enterprise perspective, what are you hearing from customers on why agentic AI requires agents that can actually reason and make decisions?

Michael Garas

>> Yeah. Great question, Scott. So what I'd say here is it all comes down to business value. Really, the goals of agents or any other technology is how can we drive ROI for an organization? How can we drive ROI for individuals? And what does that impact specifically look like? In terms of agentic AI, I think it's been a buzzword now that's been thrown around quite a bit over the past 12 to 18 months. From an enterprise customer perspective, we've seen individuals believe that agentic AI can do everything under the sun, all the way to the complete skeptics that say agentic AI just isn't where it needs to be today. I think concerns that are top of mind for enterprises, which I know Stu has a lot of expertise under, is how do we make sure AI agents have the context that they need? How do we make sure that AI agents can be trusted? And third, how do we make sure that AI agents can operate at enterprise scale? So these are the types of questions that are very top of mind for business leaders today. And Stu, I know you've got a lot of context here, so I'd love to pass it over to you to talk a little bit more about some of those topics.

Stuart Frost

>> Yeah. Thanks, Michael. This is very much top of mind for us as well right now. We're engaging with large enterprises, solving exactly these kinds of complex problems at enterprise scale. And that's also why we're partnering with IBM to help us fill that out. I think we've hit on a couple of key issues here. Scott, you mentioned the complexity of this challenge. And the reality is, in many ways, you can look at LLMs and agents and so on raw as kind of an easy, but, sorry, flawed and wrong solution to this problem. It's very rare in life that truly complex problems have an easy solution, right? And agents are not there yet. And what we're finding is that we need to underpin all of this with causal AI, which really gives you, for the first time, the ability to make truly data-driven decisions and put them in the context of the overall enterprise challenge. And what we're doing is underpinning then the agents and the LLMs with that causal information as well so that they can help us make some of the easier decisions, right? Not all decisions are really complex, but we do need that foundation. So that's where we're coming from.

Scott Hebner

>> Yeah. My view of this is from hundred, not hundreds, but many, many dozen, approaching a hundred companies I've talked to over the last six or seven months, is that there is a real desire to move AI into the realm of actually making decisions, helping people make decisions, give them superpowers. They can't deal with the data as it is, but decision-making is more than just getting information, right? And I think they've come accustomed to the power, let's just say the ROI, which is moderate, of generative AI automating tasks, being able to analyze information, identifying patterns, anomalies, all that kind of stuff, which is good. It's a good step forward. But there seems to be a desire now for actual decision intelligence to move their business forward. And I think what you guys are bringing forward here is that the LLMs today just are insufficient, right? And so why don't you touch on that a little bit on why is that the case? Michael, maybe you go first from an IBM perspective.

Michael Garas

>> Absolutely. So I'll preface and say it depends on use cases. So for us at IBM, for the use cases that are mission-critical and absolutely depend on someone's potential health or when their lives are at stake, from an LLM perspective, we all know that these things hallucinate. And so in those particular areas, we guide our particular clients to say, "How can we ensure that you leverage large language models to potentially speed up the process, augment areas where humans potentially don't need to be in the loop, but make sure ultimately there are proper checks and balances to ensure success?" We've actually built a product called watsonx.governance that helps from an LLM perspective to make sure that all stakeholders are involved in the loop from pre-production, testing, post-production, and monitoring to ensure that you can leverage the benefits of artificial intelligence without dealing with the drawbacks from an AI hallucination perspective. So there very much can be times where this is a data issue on a use case perspective. It also could just be a complexity of question type of issue as well. And so again, at IBM, we always look at what's the use case? What's the industry? And how can we best work together to develop a solution where humans plus AI are maximizing success together?

Scott Hebner

>> Yeah. And you're bringing the enterprise strength capabilities that ... Enterprises have a different set of criteria than a lot of , right?

Michael Garas

>> That's right.

Scott Hebner

>> And so between the guardrails and the domain expertise being put into the Granite, IBM Granite models, which are really impressive. The speed, the quality, right? And then you've got, like you said, the watsonx platform that kind of brings that all together. It makes a ton of sense to kind of build that enterprise strength capability. Not to mention that you can use pretty much any model you want, right? Because you guys are .

Michael Garas

>> Absolutely. Yes.

Scott Hebner

>> But I think the challenge for the industry is that the LLMs have inherent restraints to them, right? They can only go so far. And maybe, Stuart, you can kind of illuminate a little bit on that.

Stuart Frost

>> Yeah. Actually, Scott, this might be a good point to bring up that decisions fast and slow slide. And we can talk about that, right? So the way we're looking at this is that LLMs are obviously great tools, and an agentic AI based on those LLMs. Incredible tools. But to Michael's point, they're not good for every decision. They're good to spread knowledge around the organization. Might be good when you don't need absolute precision. And great. That's a very useful tool. But for those bigger, more complex decisions that are perhaps mission-critical or life-threatening, in some cases, health and safety things and that kind of stuff, they're not the right tool. We have to use something like causal AI. And I don't think there is anything else like causal AI, so I guess we have to use causal AI, in my humble opinion. Where we need that precision, we need the math behind it, we need the scientific approach. So what we've been doing over this last year, working very closely with IBM, is to really try to get the best of both worlds and build these causal knowledge graphs that actually also underpin the LLM. We use the LLM and agents to build them in the first place. So there's kind of a really nice feedback loop there. We can use the LLMs and agents to extract knowledge from the organization, might be documentation, service logs, inspection reports, whatever. We can combine that with time series data from sensors then and do the causal AI. And then we can feed all that knowledge back into that knowledge graph that then gives us better answers from the LLM and better results out of agents. So really closing that loop is incredibly powerful. And it doesn't require the old chore of building ontologies for every industry. We can actually do it much quicker and more effectively using agentic AI to bring in subject matter experts and do curation and so on. So that's an area that we're working very closely with IBM, as I mentioned, using their orchestration tools, agentic orchestration tools, and also their LLMs like Granite. We're very open on that. We do use other models, but we're finding that the IBM ones are very effective. And it's great for us because it does give us that kind of enterprise class credentials, if you like, with tools like .governance, .data, et cetera, and the watsonx frameworks. So it's really accelerating what we can do at enterprise scale.

Scott Hebner

>> Yeah. It's fascinating. I think that's how we start to progress the whole trade, right? I've come to think of generative AI and LLMs as sort of the equivalent to the browsers back in the early internet days, right? It was the most important decision you can ever make in the world, right? And what ended up happening over time is it faded into the infrastructure. It provided the foundational services, right? It is now, at best, a preference. And it provided a gateway into, in this case with AI, a gateway into the world of AI. And it can automate tasks and get information and it can do amazing things, but the value is going to be how you build around and on top of that. And if we can just bring up that chart on the brain, on the slow and fast one more time. That one. I think that's exactly what we're saying here, right? Is you're extending the LLMs and the generative AI by complementing it with sophisticated knowledge graphs that understand causality in addition to just understanding context, meaning, and relationships among the entities, right? And you're creating a feedback loop where it continuously learns. I just think it's really, really interesting because that is what's going to take it to the next level. Michael, what do you think? Is that stated reasonably?

Michael Garas

>> Yeah. I agree with that, Scott. And the other thing I'll say too is, Stu, I know one of your favorite phrases is, "AI that knows why." I think when we look at this, even pre-AI, software has done a really good job of sharing what the current state looks like within any business process. So it's very simple to understand what's happening today. It's very hard to understand without a scalable architecture, like the one Geminos and IBM have built together, to understand why a situation is happening and what you should do about it. And those deeper level questions are ultimately what are going to drive the next iteration of value that both startups and organizations are seeking for when it comes to agentic AI.

Scott Hebner

>> Yeah. It's a great combination. You've got the LLMs that are out there. Then you have the Granite models that IBM's putting out, which I think is a combination of a LLM, but it's enterprise-optimized, domain-specific, right? It's a different kind of an animal than the public, right? It's open source, all that kind of stuff. And then you infuse some of the stuff that's in Causeway, the Geminos platform, including the causal knowledge graphs. I think people are familiar with knowledge graphs for the most part, contacts, relationships, my entities. Causal knowledge graphs take it to a whole new level. And maybe, Stuart, you can just explain a little bit more about what a causal knowledge graph is and what it can do for an enterprise. And perhaps we bring up that chart for you to talk to.

Stuart Frost

>> Yeah. Absolutely. So I think most people probably listening to this are familiar with knowledge graphs in general, right? And in my view, what they do is capture what I would say is the static knowledge of around an organization, the entities, the relationships, the kind of hierarchical structures. This company has these subsidiaries, these plants, these technologies, et cetera. That's obviously important knowledge, but what it lacks is the dynamics of what's actually happening in the organization. And to my mind, as a long-time kind of modeling guy, my first company was a vendor back in the day. This is key. You need to understand the dynamics. And causal AI and the diagrams that come with it and the math that comes with it allows us to model that knowledge and capture it and express that in the knowledge graph. So this is a big breakthrough, to my mind, right? Because it really is the first time we've been able to get that combination of the dynamics of a situation and also this static underpinning of it. And for the first time with LLMs and agents, we can build all that in a reasonable timeframe at scale. So it's really coming together to create this breakthrough platform. And going back to Michael's point there, this is the critical benefit of causal AI here. When we do it right, we do understand why something's happening and, very importantly, what to do about it. There is no other form of AI that can currently do that. And that's what decisions are all about, right? Something's changing. The mean's shifting. Variance is changing. I'm getting outliers, whatever it might be. Why is that happening in my organization? What can I change to fix it? That's pretty much every decision that you need to make. And what we're seeing is that it is the right way to go. It's complex and you've got to do it at scale and so on. So what we're doing, in conjunction with IBM, is building this whole product platform that can be ingested by large organizations. And we're really lowering the barrier to entry on that, and using agents and LLMs to help us do that. I think that's a really key message that I want to get across. Because it's like we're the fork in the road in this, right? And we've seen this over and over again in technology markets. There's a seemingly easy route to a solution, but it's wrong. Sorry. It's not going to really work for you. But a lot of people are going to take that path, right? And they're going to struggle with it. We've seen this over and over again with LLMs, et cetera, in the last couple of years. There is a much better way, more effective way of doing it, but it is complex. And the vendors that really succeed then in these kind of markets are the ones that make it digestible, reduce those barriers to entry. And what we recognized early on is that we understand that we've got that vision, but we need help to do it at enterprise scale. And that's why we've partnered with IBM now for a number of years to build this out. And it's really coming together. So for me, that's incredibly exciting. And I do think it represents a big breakthrough in using AI to make these kind of decisions.

Scott Hebner

>> Yeah. You've said before, your experience with real-world customers, it just works when you start doing the prototyping and all that. Just before we move on here, just briefly, Stuart, you've referenced the math, right? And I think it's important to touch on the fact that we're talking about a new kind of math that is underpinning AI than what is underpinning generative AI and LLMs based on Judea Pearl. Could you just briefly explain that for a second? Because it's not like you guys are just making this stuff up. This is real math that you're essentially creating algorithmic capabilities with, correct?

Stuart Frost

>> Yeah. Absolutely. I think it will be looked back on as one of the great breakthroughs in math, right? He doesn't fully get the benefit of that yet, Judea Pearl, but he really tackled this problem. He's been involved in AI and stats for many years, right? He was the main guy behind Bayes and bringing that to bear. And he did get some recognition for his work in causality with the Turing Award about ... I think it was 2012. But it's been very academic. There's tons of papers on it. Very complex. But he did get some really key breakthroughs, and it's around interventions and counter-factuals, which classic AI just can't do, right? They don't understand. So he really invented this new ... He calls it the do-calculus, right? What if I do something, not just what happened, what the correlations are, what are the patterns in the data, but what if I actually intervene? And this is so powerful. And as it's progressed, then also there are now techniques for saying where are the sources of bias? How do I eliminate them? And all with transparency because it's a graphical model. Business users can easily look at these and go, "That doesn't make a lot of sense to me. How's the math working? Help me," without getting too involved, right? They don't have to go deep into data science for this. But a data scientist working with them can say, "Hey, this is what it's looking like. This may be some noise in this part of the model where it's not explaining it properly. That indicates perhaps that we've got an unobserved variable, so maybe we need to synthesize that to test that. We can maybe add a sensor at some point. That's something we should invest in." There's a whole kind of methodology behind it that we've built up here that applies that math at every point. And I honestly do believe it will be viewed as right up there with all the fundamental breakthroughs in math over the years. And what we're doing is really taking it out of that academic realm, if you like, and a very, very high barrier to entry on those papers and so on that you see, and bringing it to bear for regular large scale enterprises, right? The good news is it's taught in a lot of schools now. All the top schools are teaching it. So there are people coming out with at least some familiarity with it. What we're really doing is targeting those chief digital officers, chief AI officers who are tasked with transforming the way their business makes decisions, and then helping them to apply the math and the science behind it all in a way that they can actually leverage that and make a difference.

Scott Hebner

>> Yeah. You're democratizing it. Yeah. I actually talked to a guy at BMW over in Germany that has a PhD in causal AI. And he thinks he's probably one of the first, but it is being taught pretty much everywhere now. And for those of you that aren't familiar with Judea Pearl, you should really read the book. It's called The Book of Why. And actually, we'll move on in a second here, but I love the quote that he says that probabilities, LLMs, encode our beliefs about a static world, where causality tells us how those probabilities change when the world around us changes. Who doesn't live in a world that's not always changing? So good stuff. All right. So let's ... Michael, I'm going to put you a little bit on here to kind of describe in the essence of what you can actually do with causal AI. And we'll bring up that chart. Just talk to us a little bit about what you see the power of this being and how people actually start to implement. What can you actually do with it?

Michael Garas

>> Absolutely. So I think there's three key areas here as this chart really talks about, cause and effect, learn-cause structure, and predict and decide. You can see they're sequential in their steps. But what I'll tell you, Scott, is ... And this is something that really struck my mind during the early discussions with Stu, is I'm a big fan of the Gartner Hype Cycle and I'm also a fan of Geoffrey Moore's Crossing the Chasm. When you look at causal AI on the hype cycle, it still says it's anywhere from about three to five years away. And so naturally, when you think about where we are now, in my early discussions with Stu, it became very apparent that Geminos, from a cutting edge and leading perspective, not only had thought about this concept, not only had recognized that there was value from cause and effect, learn-cause, predict and decide, but they had actually implemented a structure to be able to identify value there. And so in our earliest discussions with Stu, I said, "We've got to find a way to work together." And Stu hit the nail on the head when eventually it came to how do we roll this out at enterprise scale? That's where leveraging tools like watsonx Orchestrate, which also underpinned Salesforce's Agentforce and also watsonx.ai's Granite models, which are very small purpose-fit models to help decrease latency and help enterprises move at scale, those were the components where we said, "This makes sense because Geminos is really leading from a causal architecture and data perspective, and IBM is leading when it comes to agent orchestration and making this scale in a cost-efficient way." And so very excited about the potential of this partnership and really what Stu's been doing here.

Scott Hebner

>> Yeah. So let's dive a little bit deeper into the actual partnership. And maybe, Stuart, you can give a quick summary of how you see it.

Stuart Frost

>> Yeah. No, it's been interesting. As you know, Scott, I've been a startup founder, CEO for a long time, few decades now. And I've done a lot of these kind of partnerships in the past. And they're always very challenging, right? It's elephant and mouse stuff, so you've got to be very careful. I've got to say IBM is doing it better than anything I've experienced before. Very responsive, very conscious of the different scales of the organization, how fast each one can move, and the strengths and weaknesses of all that. Very open technically as well. We have access to technical resources anytime we want them. They've been incredibly helpful in understanding how to get good use out of Granite, for example, et cetera. So it's really been eye-opening for me, quite honestly. I've worked with IBM on and off for a long time now, but this is a different IBM. It really is. Yeah. And obviously, it's a great fit for us because as a startup, you have to play to your own strengths. And building a full price scale platform at every level, it's too big an ask, right? So yeah, partnering the right way with the right people who get it and want to leverage our strengths. That's what it's all about, from my perspective. I think as well, there's a lot of potential on the go-to-market side to help us scale this up on that side too, while recognizing that we've got to be patient and get the right timing on that. But it's definitely part of the partnership. And we're very excited about that too. And frankly, just having day-to-day access to people like Michael and so on, who he's been a fantastic advocate for us within IBM, and Parker. I know I can get hold of them very quickly. And pretty much anyone in the IBM organization that I want to speak to, it's usually just a day or two to get access to and get their insights and help with whatever I need. It's a revelation.

Scott Hebner

>> That was the key to growth back in the E-business days, was opening up the ecosystem and partnerships. And so it's great to see that happening again. And the way ... my 2 cents on all this is when we were talking about ... One of the concepts we came up with at theCUBE Research is this notion of the ladder to agentic AI, right? You start with your gen AI, LLM foundation, you have to infuse domain knowledge, Granite, right? You need explainability, decision intelligence into the agents, which is the causality and the knowledge graphs and the causal knowledge graphs. And that's the Causeway from Geminos, plus what you guys are providing with watsonx to build and govern all that, right? And then there's the ability to wire all that together into agentic workflows that are adaptable, and that's watson Orchestra, right? And that's just to name a small number of products, but it really comes together nicely to kind of bring it all together, right? And then you have that continuous learning loop, which is the slow and fast brain. And I think it's a compelling story that sort of does recognize what's going to happen in the future, and is a step ahead there. So before we wrap up here and just ... We got to do this briefly because I think we're running out of time here. But before we go, what would be your one piece of advice to people that are watching today that are just getting started with their agentic AI journey? Stuart, you want to go first? And Michael, then we'll get your thoughts on it.

Stuart Frost

>> For me, if you are trying to really help your business make these kind of complex decisions at scale, you've got to recognize that that is not easy, right? And there is no out-of-the-box, plug in an LLM agent framework, and it's going to do it for you. That's never been the case. It never will be, right? My humble opinion. Complex problems typically require complex solutions. So take a close look at this causal AI stuff because it is a big breakthrough. We are effectively deploying it and generating really significant ROIs from it. And our mission is very much to lower the barriers to entry. And Michael mentioned the Geoffrey Moore books. I've known Geoffrey for 30 years now, I think. And a big fan of what he talks about. And our mission is to be that company that takes it across into the mainstream, right? And crosses the chasm, to use his phrase. So very much on track to do that. That's what we're doing strategically. We're only targeting those large enterprises that have these complex problems and bringing to their partners like IBM to make it all real and make it all possible.

Scott Hebner

>> Awesome. Michael?

Michael Garas

>> Absolutely. And mine goes back to a comment from earlier on in the session, which is really AI is exciting, agentic AI is exciting, but my best recommendation for startups and large organizations as a whole is really focus on what is the ROI opportunity you're looking to drive. What's the problem you're looking to solve? And ensure you've got the right technology, architecture, and data to be able to support it. So starting with the business value first and keeping the focus there throughout the cycle will help individuals ultimately be able to realize the solutions that they're looking for. And as we've said, Stu and Geminos are really at the cutting edge of causal AI to be able to solve those high-value decisions versus ones like answering what questions are in a frequently asked questions document. So very excited about that opportunity.

Scott Hebner

>> Yeah. And given that agents are about decisions, or they will as the years go by, one of the themes of our Summit is trust, right? Trust has become the currency of innovation. The more people trust it culturally, the more they're going to buy into the new innovations and start using it. And it's just going to create that cycle. So no trust, no ROI, right? And I think the combination of what you're doing at IBM and what Geminos is doing around more trustworthy decision-making, good combination. Well, look, I wish we can keep going, but we have to wrap up. Stuart, Michael, thanks so much for joining us today. This has been a really inspiring and informative discussion, so please keep us updated on progress. And for all of you out there, thank you for tuning in. Really appreciate it. You can access the entire AI Agent Builder Summit set of sessions out at theCUBE.net or on our YouTube channel. Please share the videos, articles, clips from all the sessions, including this one, with your colleagues and on social media. The portal makes it very easy to do that. So we'll see you again soon. We are the leader in enterprise tech news and analysis. Bye for now.