Hello and welcome to a special edition of Analyst Angle, where we'll continue our discussion around what to know and how to prepare for the next frontier of AI and business. Our view here at The Cube Research is that 2025 will represent the year of the rise of

agentic AI, where networks of AI agents collaborate and help humans achieve goals, solve problems, make better decisions, and even act autonomously when human intervention is not feasible or warranted. We bring up a chart here just to kind of refresh, um, you know, our

viewpoint here that we have published through our research and in previous Analyst Angles. Um, obviously today most businesses rely on what we would consider to be AI assistants. These are task-oriented assistants that take a prompt and help you automate the achievement of a task, whether it be a

prediction or or generating some kind of content or a forecast. Um, very, very effective. Um, it's at the core of what most people use generative AI for today, and the implicit ROI is clearly there. We see AI agents taking that next major step forward that evolve from being

task-based to being goal-based and helping humans and sometimes autonomously, um, problem solve, make decisions, and actually achieve things that are more than just a task. So the big change here is from a task to a goal, right? And then agentic

AI is when you have multiple agents that are collaborating within a network to solve more complex goals or challenges, and in particular the ones that tend to be more organizational, um, in nature. And the progression here is from correlative, you know, predictions and statistical analysis into decision intelligence.

into reasoning and of course into actual problem solving and there's going to be a technology transformation that's happening under the cover of these agents right they're going to require semantic reasoning and they're going to require causal reasoning to understand cause and effect and they're going to

need to continuously learn from each other um the transition from generative AI to agentic AI is well underway with that I'm excited to be joined by Dr. Darko Matovic the co-founder and CEO of Calin Darko is a research scientist

turned technology entrepreneur he and his team are pioneering the world's first trusted AI agents for data science his vision is to empower everyone with their own data science team and to democratize AI for the masses under his leadership Calin has emerged as an impactful pioneer in

helping Enterprises transform how they make decisions solve complex problems and augment their talent pools with autonomous AI agents Darko welcome I'm so thankful for you taking time to talk with us today it's a pleasure Scott thank you so much for the invite yeah it

looks really nice behind you there it's you're over in London correct yeah it's just yeah it's it's one of the coldest days actually in London but it's it's nice and bright at least we saw we saw the sunshine today so after after a few weeks yeah well

good it was it's finally rained here in the Northeast I think we a couple weeks without any real rain so um but the sun's back out uh one how I've been uh following quite closely as you know as being part of the Leadership Council and

spending some time with you um and your team over the last year and the innovation that you guys are driving is really impressive and inspiring and in particular that you're doing it in partnership with real customers um you guys have definitely been keeping busy

um can you just talk a little bit about your mission your journey to where you are now and and where you see this all heading into the future yeah absolutely so the the real problem uh that that we all have on our hands is that we are generating more data than we can

ever be analyzed um there's an exponential growth in data just for uh for people to visualize this if you were going to take a terabyte discs and fill them with data you could actually build one and a half uh walls of the Great Walls of China with those one terabyte uh discs

right that's the amount of data uh that the world is currently generating now um as as you know the audience would know here we need you know data scientists to analyze their data the trouble is that human data scientists have only been growing linearly um and there is kind of

a capacity to how many new data scientists uh we can train at universities and things like that so we having this exponential growth in data only a linear growth in in human data scientists and this gap is only widening combine that um with the need or the

want of large Enterprises as well as small as well as governments in the last few years as AI has become you know a technology that everybody uh cares and talks about um there's now a 10x maybe 100x more demand

for um for doing data science so when you combine the exponential growth of data uh and just the linear growth of human data scientists and the need for the world to do more data analysis you come to a very obvious problem where we just simply do not have the humans to analyze

all this data in fact some of the research we've seen uh has shown that um as much as 88% of corporate data has never even been looked at right so that's just like a massive uh um lost opportunity for society for or uh companies for

governments to make better decisions right uh so the only way you solve this is uh by AI by creating AI data science agents that can kind of do the work um of human data scientists now we're not suggesting that uh we're going to come

up with AI data scientists to uh remove the need for humans in fact we need all the humans we can get and we need the AI we need basically new types of teams in which human data scientists and AI data scientists work together in new ways um and we can kind of you know get

uh go and and and and kind of achieve this uh this goal of analyzing this exponentially growing data so uh we we at Coal Lens uh we are on a mission to create the most reliable most trusted AI data scientist or you know uh as you

said 2025 is all about agents we can call it a data science agent if you if you like um same thing so that's really what we are on a mission uh to achieve and and we are already unlocking tremendous amount of value with um for our

500 companies. It's actually kind of fascinating to see, in fact, sometimes our own team gets surprised at the speed at which we can unlock value compared to kind of conventional ways of doing this. So I'll pause there. Um, super happy to dive deeper into any of

these aspects. You know, I think you're really on to something in, and just to talk a little bit about the talent, I think it's becoming a talent crisis. Um, we actually published a report on the AI Talent Marketplace, I think it was back in October, um, where we did some

analysis of the supply and demand dynamics, um, with around data scientists and AI engineers as well as AI executives. Um, and what we found when it came to the AI engineers, that there was a 2.3 times gap in the supply and demand of these types of

skills, and that was in, you know, obviously 2023 and into 2024. Um, so supply is not keeping up with demand. Um, in addition to that, the cost that you have to pay to get the best data scientists is even going higher, and

I remember in the report that almost half of the people that have a data science job at a particular company are actually looking to potentially go to another company to make more money. Then you add to that, if you look at the universities, there's a 4.4 times gap in

the demand for getting a data science or AI degree with the university's ability to actually keep up and take those people in. So it suggests that the supply and demand is only going to grow greater. So this notion of a Gen AI, as we talked about a little bit earlier, right, augment

Human talent can do more things autonomously, make people into superpowers, give them superpowers, right? And in the area of data science, I think that's just a great idea uh for you guys to target because um, you know, pretty much every company I talk to, they worry about the talent

pool, that's right. Yeah, and as as you know, Scott, when you that you're referring to the current demand, but um there's so many use cases uh in which if the cost of doing data science was lower, or it was uh or the marginal cost was zero, you

will actually end up uh, you know, doing uh uh data science. So the the potential demand for for data science is actually probably in the order of of of 10 or 100, maybe a thousand X from from what we can measure today, right? Like you know, I can

give you an example with one of our customers. Uh, they had a kind of medium um impact use case that um that would have never been justified to be undertaken by human data scientists because it would just be, you know, the cost of doing that data science, science

with humans would not be justifiable given the ROI you can unlock. But once you have an agent doing it, well now all of a sudden it becomes economically viable and you're kind of unlocking that that value and you you're effectively creating new demand that previously was

just not going to be there. Uh equally, there are some use cases. One of our uh um Fortune um 100 customers, they have enormously complex supply chains, 50,000 uh SKUs uh to model it and um they have amazing data science teams and you know

really great, but you know it's a team of 20 and they've been uh trying to model the supply chain, but the supply chain got so complex that it was very hard for them to model it at a very grounded level, at like a single uh

product, single skew right now. And those people with automation, with agents that can, can basically do the work of a data scientist, and all of a sudden this team of 20 now created 50,000 models that automatically get updated every month, and they unlocked tremendous uh um uh

value in in that process. So by making data science effectively um free, you're enabling uh some use cases that were previously just not possible. So there's all, all these kind of interesting things you can do when the marginal cost of doing data science goes to zero. Yeah.

it's amazing. So um we have a, a, a demo here, here I think it's just a short little demo so people can see and kind of bring to life what we're, we're talking about here. So if um we can bring that up, that would be

[Music] Great. Today we want to understand how we can retain as many customers as possible by uncovering the causal drivers of retention. We'll be working alongside our AI agent to develop insights and present them in an actionable application. Let's begin by asking our

agent to outline the steps needed to improve retention. As we can see, the agent has developed a plan to achieve our objective, including causal effect estimation and churn modeling. We want the agent to summarize the findings from exploratory data analysis, so it adapts

the plan to do just that once the plan is approved you can see the agent get to work the agent will go through all the suggested steps allowing you to edit and update wherever you deem necessary finally we'll engage our grounding agent a fundamental part of

the workflow that enables business stakeholders to easily engage with and draw value from data science outputs building an app that's usable by executives analysts and other data scientists is as simple as clicking a button let's ask the grounding agent to explain the model

parameters watch as it provides a detailed analysis of our model breaking down the details for each factor influencing churn we'll also task our agent with identifying customer segments most likely to churn it swiftly provides this information and potential targeted interventions for each high-risk

group lastly this can also be done for specific customers across our customer base with the grounding agent offering tailored recommendations with rationales for each suggestion this customer churn demo showcases the power of causal lens's AI agents that helps solve business problems using quantitative data enabling

them to become trusted decision-making partners across your business so Darko pretty pretty impressive um it's fair to state here that while that was a customer churn uh use case the data science agent can be applied to pretty much any use case right as a companion to absolutely yeah

absolutely and and there's a couple of modes in which you can use this agent right uh the the demo you just saw is uh is part of enabling uh non data scientists analysts and similar personas to be able to do data science and um and that's that's pretty

exciting because we have so many examples where uh the organization has hundreds of questions uh sometimes thousands of of questions that all go you know all get sent to to the data science team um and the data science team just simply taps out of capacity

just you know they they're like hey yeah we can do this but it will take six months right um now imagine if the data science team could actually prepare certain workflows um effectively um more of a tailored tailored agent um for specific use cases kind of like what you

saw there and enable the analysts um to be able to self-serve without having to uh you know get in the queue for the data science capacity so that's kind of what you saw there uh we of course have agents also that are specialized in

different parts of of uh of data science tasks things like you know autonomously cleaning data or autonomously building models or building apps things like that where you can empower directly the data science team themselves so that you can make each data scientist into 100 uh and the the

flow there is a little bit different uh and it's kind of like the the end stage as you had on your initial slide and I really like your slide by the way really great one really resonated with me uh once you we get to the agentic AI site

kind of the final phase we'll also have agents that are just fully autonomous that have been verified by the data scientist in the business but have been deployed in a way that can uh that can act autonomously so I guess what you saw in that demo is like that middle

piece where um it's about enabling um uh the the analyst we have you know first step is usually enabling the data scientist and then um the ultimate step is uh autonomous agentic AI that is just you know uh we're able to deploy in a trusted way and it's going to do a bunch

of stuff for us yeah this notion of supercharging your data science team is a great idea and um I still come across you know quite a few companies that actually don't even have a data science team exactly yeah yeah they got business analysts and they got people that can do

Tableau and things of that nature but you know they're going to need to make that transition this is great um and actually um you guys put out um on LinkedIn a great what I thought was a great ad and we'll put it up right here

that I think really nails it um hopefully you can all read that what if everyone could have their own data science team not in five years not next year this week um I think that kind of like nails the value proposition of where you're heading with all this and

uh impressive stuff so so Darko the uh you know we're all witnessing these innovation cycles right and it seems like every technological transformation that we've had over the last many decades then innovation cycles are speeding faster and faster right as is the promise of value

that that innovation brings um but another thing I hear quite consistently is can I trust these AI agents and you know with with the generative AI stuff you know they're familiar with hallucinations and bias and the concealment of confounding effects

things of that nature but can I really trust these AI agents um these task-oriented ones sort of have been getting mixed reviews so what's your approach to improving trust and transparency and explainability for the people that ultimately have to use this

and to make you know business critical decisions yeah so so Scott we we think about uh the value proposition in two parts and automation is is extremely important and we've been talking about that in the first part of the conversation but automation without trust is irrelevant

it's not useful so trust is is critical and and the way we think about trust uh comes from actually uh quite a few directions the very first thing is is can we enable this uh this agents to to reason in a causal way so causal

reasoning is a very very critical part um of of having trust into the uh into the agent and as you know we the pioneers of causal AI we created this category um and we you know we have pretty exciting IP in a way that we can

ground um this agents in causal reasoning so uh if you have causal reasoning you effectively have the ability to to kind of build a digital twin uh of an Enterprise or or government or or supply chain or whatever the case is um and you have a

ability for humans to to kind of verify that so so domain experts to be able to verify that so that's one layer of Trust on the modeling side so how do we know that these agents are operating and understanding the the world the real world uh in a way that the real world

actually behaves, and the only way to do that is by learning the cause and effect relationships in the real world. Having a digital twin of a manufacturing process, having a digital twin of a, um, of a supply chain, um, having a digital twin of how customers uh behave, um, that is

fundamental at the modeling level. Now, there's other, that's that's that's very important, but not sufficient. There's other areas in which uh you want to be able to bring trust. So, um, the way we've designed our Enterprise platform is that it actually learns um about your business

context, learns about your environment. Say, for example, you're a bank, well, you it's very unlikely that your compliance team will allow you uh to have uh deep neural networks deciding on whether you you provide a loan to someone, right? It's very unlikely. So, so in that business

context, uh, the AI agent, the data science agent, should know that, should maybe favor maybe a causal model or a linear, linear model or something that a compliance team can look at and actually approve, something that the regulators can look at and and can be happy with. So

that's what we call a long-term memory, where the agent actually learns from your, as this agent is is acting in your business, in your environment, it's learning from the human data scientist, how do we build models or how do we build applications here. Um, so that's

really, uh, you know, a critical part where where you can bring um trust. So, causal reasoning, learning uh from humans, um, uh and learning from the organizational context are just two of the ways in which we think about trust, but you know, it's a multi, you know

faceted uh approach um and there's many many other layers that that I'm happy to take you through yeah and you you are you you are someone who lives this as a uh a core uh priority right and I know you're involved with the World Economic

Forum for example and I believe you're involved in working with them on AI ethics is a little bit about what you're doing there yeah yeah it's it's called um the it's kind of like AI Governance Alliance um where some of the the the best minds

in AI um are there um and it it's about kind of crafting the way to how do we you know uh everybody knows that AI is transformative right everybody knows that this is the fastest um uh fastest growing technology the most impactful technology of our day um and you know everybody

would love to have it but we need to to roll it out in a way that is safe and trusted um and so we're very very passionate about this topic and that's why we you know we started the company uh with causal reasoning because we

we always felt that unless you have a causal understanding of the world it would be very hard to build uh agents that can decide on your behalf yeah well you know one thing I've definitely learned from talking to not only just your team but other vendors that are in

the space and you know even my youngest son's he's over near you going to Oxford University to get a PhD in probabilistic math and uh you know what they tell me is that there's some super complex and eloquent technology and math for that

matter that's under the covers that's all anchored as you mentioned in in CAO AI and emergent Trends and stochastic, you know, kind of probability things that I'm not going to pretend to try to understand. Um, but when I, well you know, when I experienced Decision OS which is

your your platform, um, when I went through the trial and I got the certification and all that I was just like wow, you know, first of all it brought it to life for me, but you know, I was thinking this is democratizing this stuff for like everyday businesses.

And so you start off with Decision OS, I think that's your underlying platform for the agents, can you just uh kind of talk a little bit about that? Yeah, and so in that ad that you saw earlier on the tube, um, it's uh it also

said that uh agents arriving December, so December is is around the corner and um we're going to have a completely different um, you know, look to Decision OS than the one you've seen, so if you thought that that was was powerful, uh, you know, I would like to invite you to

to try it again on the 11th of December. Um, I think, you know, it's going to be at least 10x what you experience, so the the way you will experience Decision OS on the 11th of December is essentially an environment, a a completely novel like

nothing that the world has ever seen, right? A novel environment where uh AI agents come together with the humans to take us from raw data to better decisions in a trusted way, so it's really like completely changing the way we build um AI, the way we we we build do

data science so normally data scientists are used to kind of like IDs of some sort like Jupyter for example and that's all great you know of course you know it's it's useful but but now imagine a completely different world in which uh it's designed agent first and imagine

that at some point 99% of all data science work from data cleaning to dashboards building to model building to um retrieving the right data from my SQL database all that being done by by agents and by the way we're talking here about multi-agentic systems right with

different specialities and ability to build your own like this really cool stuff um and then having the human data scientist acting as the 1% backstop validating and and uh and ensuring that whatever the agent builds can be trusted uh and can be can be put in production

uh to to to kind of change our world right so really decision becomes now agent first it's a multi-agent system that enables agents to collaborate with each other um and build you know take you from raw data to better decisions and having the right ways for humans to

come in and ensure that the results uh make sense and and validate it the most exciting part probably uh for for our customers is going to be the ability to build your own agents for a specific use case um we're actually pretty excited we we um uh this week we had our first

customer build um their own agent for a very business critical part you know they're um they're in the finance world and uh they wanted to kind of you know uh have a an agent that selects um uh certain channels and and the optimal allocation um and it was so exciting for

for them and for us to build the very first use case um agent that was able to do that and now you know they they feel like wow we have a digital AI digital um data scientist that can just do this for us now every day previously they they

had had to like wait for three months to to refresh this kind of analysis so it is really exciting and and you know I do invite you to come back on the 11th of December try try it again um I'm going to do that I think it was back

in March maybe that I went through the the certification the tutorial the experience and you know experiencing is believing and but I do remember every once in a while getting caught up in the Python and all that stuff so I'm assuming you know it's going to make

make me able to actually maybe someday absolutely actually do all this stuff right um but you had mentioned um an example there in finance um yeah as I as I kind of wrote in the research note that we put out not too long ago on the calls away Marketplace

um I think this Market the entire Market is really starting to build and I think it's going to accelerate from here a lot of the players to date though that deliver Callo AI capabilities are doing it in highly industry vertical ways like solar Farms or doing you know around

energy or or you know life sciences would be another example um you at caal Lens tend to have customers that are across all different kinds of use cases and in customer segments and industries right so more of a general purpose platform where regardless of what your

use case is um you can actually start to build you know these AI capabilities and the reasoning and everything you talked about is that something that you see as um a big differentiator for causal lens yes so so that's right yes uh if you have built an AI data scientist well

you expect them to actually be able to to be able to just be able to to work on on data in any industry right um it's kind of like if you were to hire a human well it doesn't you know there's nothing there to say that well this human can

only work in finance or can only work in energy right and so that's kind of the mindset uh with which we're building our product if we are serious about uh democratizing AI about giving everybody access to um uh if we're serious about everyone having access to

their own AI data science team um then this is the only way um now um the way we approach this is that uh so different data scientists have different skills right so you can find a data scientist that is specialized in causal AI right uh well great you know we have Kaya which

Darko Matovski, causaLens |AnalystANGLE

SiliconANGLE theCUBE

에이전트 AI 소개 및 비즈니스에 미치는 영향

2025년 에이전트 AI의 부상에 초점을 맞춰 AI와 비즈니스의 다음 영역에 대비하는 방법에 대해 논의합니다.

에이전트 AI 네트워크는 인간이 자율적으로 목표를 달성하고, 문제를 해결하고, 의사 결정을 내릴 수 있도록 협력합니다.

현재 비즈니스는 업무 중심 자동화를 위해 AI 어시스턴트에 의존하고 있습니다.

AI 에이전트는 작업 기반에서 목표 기반으로 진화하여 문제를 해결하고 단순한 작업 이상의 것을 달성합니다.

에이전트 AI는 여러 에이전트가 협업하여 복잡한 조직의 목표를 해결하는 것입니다.

데이터 분석의 과제와 해결책 AI 데이터 과학 에이전트

데이터의 기하급수적인 증가가 데이터 과학자의 선형적인 증가를 앞지르면서 그 격차가 점점 더 벌어지고 있습니다.

대기업, 중소기업, 정부에서 데이터 과학에 대한 수요 증가.

기업 데이터의 상당 부분은 인적 자원 부족으로 인해 분석되지 않은 채로 남아 있습니다.

AI 데이터 과학 에이전트는 인간 데이터 과학자와 함께 일하면서 그 격차를 해소하는 솔루션으로 제안됩니다.

신뢰할 수 있고 신뢰할 수 있는 AI 데이터 과학 에이전트를 만들어 비즈니스의 가치를 실현하는 것이 Causal Lens의 사명입니다.

데모: 고객 이탈 예측을 위한 AI 에이전트 활용 사례

AI 에이전트를 사용하여 고객 유지율을 이해하고 개선하는 방법을 보여주는 데모입니다.

상담원은 계획을 수립하고, 탐색적 데이터 분석을 수행하고, 결과를 요약합니다.

접지 에이전트를 사용하면 비즈니스 이해관계자가 데이터 과학 결과물을 활용하고 가치를 창출할 수 있습니다.

상담원은 이탈 가능성이 가장 높은 고객 세그먼트를 식별하여 타겟팅된 개입을 제공합니다.

이 데모는 정량적 데이터를 사용하여 비즈니스 문제를 해결하는 AI 에이전트의 힘을 강조합니다.

AI 에이전트의 신뢰, 투명성 및 설명 가능성

신뢰 없는 자동화는 무의미하며, AI 에이전트 도입에 있어 신뢰는 매우 중요합니다.

인과적 추론은 AI 에이전트에 대한 신뢰를 구축하는 데 매우 중요합니다.

기업 또는 프로세스의 디지털 트윈을 구축하면 도메인 전문가의 검증을 받을 수 있습니다.

플랫폼은 비즈니스 상황을 학습하고 규제 요건에 맞게 조정합니다.

신뢰는 인과적 추론, 인간으로부터의 학습, 조직적 맥락에 대한 적응을 통해 구축됩니다.

Decision OS: 에이전트 AI를 위한 플랫폼

Decision OS는 AI 에이전트와 사람이 협업하여 더 나은 결정을 내릴 수 있는 환경입니다.

기존 IDE에서 에이전트 우선 접근 방식으로 전환하여 AI 및 데이터 과학을 구축하는 방식을 변화시킵니다.

이 플랫폼은 전문 에이전트와 사람의 감독을 받는 멀티 에이전트 시스템을 지원합니다.

사용자는 특정 사용 사례에 맞게 자체 에이전트를 구축할 수 있습니다.

재무 최적화를 위해 맞춤형 에이전트를 사용한 고객 사례를 소개합니다.