The Data Foundation for AI: Daniel Cohen-Dumani of Experio AI on Balancing Centralization, Governance, and Agility

In this episode of Evolving the Enterprise, Dayle Hall talks with Daniel Cohen-Dumani, CEO and Founder of Experio AI — a leader in enterprise transformation and agentic AI innovation. Daniel brings over three decades of experience in technology consulting and data strategy to unpack how organizations can structure, govern, and future-proof their data for the age of AI.

From breaking down centralized versus decentralized models to understanding the rise of AI agents, Daniel shares clear, actionable insights for leaders seeking to balance agility with governance and empower teams for long-term success.

Transcript

Copy

Dayle Hall:

Hi out there. Welcome to the latest episode of Evolving the Enterprise. I'm your host, Dayle Hall, the CMO of SnapLogic. This is the show where we explore how organizations are transforming with AI and multiple aspects of innovation.

This season, we've heard repeatedly that data is the foundation of every AI strategy. You may have heard that yourself. There are a few brands out there claiming that. SnapLogic, too. But the big question remains, how should that data actually be structured and governed in a world where AI and agents are now heavily reliant on that data, particularly in environments where there might be regulatory compliance or constant changes trying to get access to this type of data? Should an enterprise centralize everything? Decentralize? Should they have data lakes, data mesh? How do they actually make all these data sources work together?

To help us navigate this today, I'm joined by Daniel Cohen-Dumani, who is the CEO and founder of Experio. They're basically a recognized leader in enterprise transformation. Daniel has guided global organizations through complex digital strategies. Today, he'll share how leaders can make the right choices to future-proof their data and AI ecosystems, a hot topic, Daniel. Welcome to the show.

Daniel Cohen-Dumani:

Indeed. Thanks for having me. It's a pleasure to be here today.

Dayle Hall:

Yeah, it's going to be a great one. Obviously, it's a big topic. Before we actually kick in on some of the questions, give us a little bit of background on yourself, a little bit about Experio, and how you came to be at this inflection point in our industry where AI has pretty much taken off?

Daniel Cohen-Dumani:

Absolutely. For my background, I've been in technology consulting for three decades, starting my career in consulting. I'm from originally Switzerland and moved to the US in ‘98. In 2002, I started a consulting business on my own, focusing on knowledge and data management solutions. And that's when I got engaged in strategy around how to manage data at scale and knowledge at scale.

In 2023, I got the bug of AI and decided consulting was in the past, and I wanted to focus on solving a problem that I've seen prevalent in the entire consulting industry, which is the inability to find content and information. So that gave birth to Experio. We're the first agentic AI solution for the consulting industry. I'm super excited to talk about data strategy, which is really at the core of what we do.

Dayle Hall:

Yeah. When you say things like you've been doing this for three decades, me, too, mate. So we're both old hands at this. But it's always good to see some of this new technology take off. And it's crazy because I talk to my kids, both in high school now, and I just think about what they're going to be experiencing- and what I'm experiencing now and how exciting it is, but what they're going to be experiencing when they come out of college and what the work environment's going to be. It's exciting.

Daniel Cohen-Dumani:

Exciting. It's scary, too, right?

Dayle Hall:

It definitely is.

All right, let's start with some of the fundamentals before we get really into the AI piece and leveraging it and building ownership and governance and so on. Let's start with the basics around data architecture. It's still very important, probably more important than ever. In your opinion, why is data structure so critical, just for the enterprise in general, but obviously as we start to look at leveraging AI? Give us the basics.

Daniel Cohen-Dumani:

It's a great question to start and set the stage about the importance of data. When you think about AI, all the outcomes are truly gated by the data structure, right? If you don't have a clean way to capture the data at its source, have rich metadata and relationship between those datasets, any AI model will start to hallucinate. The value that you're trying to derive from AI would evaporate literally. Architecture is the foundation of giving AI some context and value that can provide better outcomes.

Dayle Hall:

Yeah. Do you find when you talk to customers, clients, or the people in and around the industry, it feels like- because we are in it all, we're in it every day, we're thinking about these kinds of things, we talk to customers, clients. Is that generally understood? Do people understand that that is still critical? And do people have a sense that our structure's fine? Do they really know what's going on under the covers?

Daniel Cohen-Dumani:

No. 

Dayle Hall:

Okay. All right. Good. That's what I was hoping you'd say.

Daniel Cohen-Dumani:

Yeah. There's a common understanding that data is important. I think that has become prevalent over the last 5 to 10 years. But they don't understand why. And then I think it's bridging that gap that's been a challenge over the last 2 to 3 years.

Dayle Hall:

And so I think, again, as I talk to people on these podcasts, as I talk to customers and prospects, there's generally an acceptance that they know they need to improve their systems, their architecture for data. A lot of the challenge is just that they still have- most enterprises, maybe some of the newer start-ups don't have as much of a legacy setup, but a lot of these bigger enterprises have a lot of legacy systems, legacy data source. Sometimes they can't make that connection. Is that the biggest pain point in the enterprise right now, or is it something else?

Daniel Cohen-Dumani:

Yeah. One of the biggest pain points is truly what I call this fragmented organizational memory, which is, where's the data and why is it so siloed across different systems, right? I think we're still following that paradigm of, I buy a CRM system, the data will be in the CRM system, or I buy an ERP system. It's highly siloed and fragmented. And we've seen a trend over the last 10 years of, okay, we need to start integrating all of those systems, but it's still very difficult to find the true source of record, right? So if you're dealing with customers, I'm sure you have customer data everywhere. What is the true one source of record? How can you tell AI this is really that place where you should look for customers? Most organizations have no idea.

Dayle Hall:

Yeah. When you talk to these organizations, is that the baseline that you start with, like,we just need to know all the pieces of data, where you have it, that's the only way we can get going. I can imagine if you're a client or a customer and you think about that, holy crap, we've got to define all this before we can do anything. And then there's this discussion around if you want to use AI, you start with the quick wins. So you get a quick win and it grows from there. How do you balance the quick win with, holy crap, we have to figure out where all our data is before we know what we can actually use? How do you balance that?

Daniel Cohen-Dumani:

Yeah, a great question. I think when you talk to executives and say, to get better artificial intelligence in the organization, you need to clean up your data, they will look at you and say, it's going to take us 10 years. It's a massive task. And it is, totally. I think instead of that, the approach has to be, all right, let's focus on one specific use case. And I said this, identify what is the source of the data or the datasets involved in answering good questions for that specific use case and start small and use that as a baseline to expand further.

Dayle Hall:

Which side of the fence do most customers or most clients currently sit on? Do they sit on, we really need to get going, because they're just trying to get some kind of project up and running? And how hard is it to persuade someone that you really need to understand all the pieces of the data, if you were going in to advise someone today and they're conflicted?

Daniel Cohen-Dumani:

I would say do both, because you know that the long term, you need a unified data architecture, but knowing how hard it is, start small. So think about your long-term goal, unified data architecture, but maybe start with one small use case at the same time because you're going to learn a lot from there. And maybe you're going to realize that maybe we can decentralize those domains of data and knowledge in small pieces and mesh them together as opposed to try to solve it all at once, which is scary for a lot of organizations.

Dayle Hall:

Yeah, for sure. One of the things that we hear more about with AI is the business units, the functions, the business users, are driving some of that requirement. Because the data is so critically important, do you see the business is driving some of these initiatives, but obviously, IT and the technical side of the house have to be involved because of the data? Or are you seeing more of the IT know that they have to make some of those changes around the data, so they're still driving the bigger initiatives? Who's coming to you first? Who are you talking to more? Is it the business side, or the technology side?

Daniel Cohen-Dumani:

We're talking more to the business side, of course, because they have challenges and business problems to solve. And they could care less where IT is. They just want them solved ASAP, right? Then their conversations would lead ``to, okay, where's your dataset? How are we going to access them? What does that mean? What's the meaning of that data? And that's where I think there's constant conflict of back and forth. IT will tend to slow it down. They think they can do it themselves. They don't want outside vendors to come in. We see that conflict appears almost every time.

Dayle Hall:

Yeah. If you listen to this podcast, what I try and think about is, what could someone else take from this in doing their regular job? My question to you around that specifically, and if someone's listening to this, they may be faced with this challenge, how do you help the business users get closer to IT and have them move faster? And then how do you persuade or influence IT to say, look, it's not about whether you build this yourself, it's about moving fast, it's about better way to solve business challenges? How can you help or how do you think about helping those two pieces come together so you can actually get a better outcome?

Daniel Cohen-Dumani:

It's a great question, Dayle. I think that it really hasn't changed over the last few decades I've been in this business, but I think AI has brought it up to a certain degree. And I think that the critical aspect is quantifying the pain of those business users and what is the return on investment in solving that pain. We've always talked about ROI on technical projects, right? But when you think about AI, that’s much more critical because the return on investment can be dramatically higher than anything as you've ever implemented before, just of the nature of what AI can solve.

So I think the business folks have to focus on quantifying, how painful it is and how much better can our life be? How much money can we save? How much revenue can you deliver, or how much more revenue can you deliver by having this solution as a means to put some pressure on urgency to solve it now?

Dayle Hall:

Yeah, it's interesting because I'm sure if you can scope a project like this, focus on the business outcomes and how it will improve, I don't know, could be efficiency, could be top line, could be bottom line, doesn't really matter, but I feel like that's where the business is more unique to help the senior-level execs make decisions on funding, where I think sometimes IT is seen more of- they actually want to improve things, too, but it's a little bit harder for that organization typically to directly link it to business value. So actually, the two teams working together is probably a unique and mutually beneficial relationship wouldn’t you say?

Daniel Cohen-Dumani:

No doubt. And I've seen it. Clients come to me and say, we love your product. It sounds like it's going to solve value, but we can't do anything until we solve our data problem. We don't know what the data is. And I said, that may be true, but I feel like always start small with anything you do. Don't try to bowl the ocean because you are never going to get there. You’re in a world that's moving so fast that making a five-year plan to fix your data architecture is becoming meaningless. I think you have to move fast, find a way to be agile and move at a fast speed. Otherwise, your business won't be here five years from now.

Dayle Hall:

I like that because, again, you and I have been around a long time. We've been through tons of moving to the cloud, CRM, ERP, all these technologies. Typically, those projects have been massive multi-year projects. And now, like you said, with AI, if you don't start making changes now, don't make a five-year plan, everything's going to be different next year.

But how can a business really balance that? How can they literally sit down and say, okay, we're going to make this plan, it's going to be a six to nine months project, but they still have to have a road map? How do they prioritize and resource that internally, in your experience?

Daniel Cohen-Dumani:

I've seen that change truly over the last probably five years, but accurately over the last two years, is the willingness to try and experiment. Back in the days, you were doing proof of concept, and I think that those tend to take a long time. But with AI technology and agents and products, you can do a proof of concept or even a proof of value in two weeks, a week. So you can deploy something that will show value extremely quickly.

And I think the businesses that are being successful are the ones that are willing to take a risk of, let's experiment. I know we're going to continue with planning, but in the meantime, let's try to see if we can solve this problem and see if there's new way of doing this, specifically in the age of AI that we all live in. You have to be willing to take the risk and experiment.

Dayle Hall:

Yeah. I feel like across the enterprises that I've talked to, because AI has such a big impact and influence, there's still some concern, obviously, and again, this is why it's going to come back to that data side, but there's definitely a willingness to experiment, where I think moving, changing ERP applications or implementing a new marketing automation tool felt like there was a lot of assessments and thinking and planning and strategy. And now, because of AI, there's a little bit more if we can control some of the access to the data, we can do a proof of concept way quicker.

Daniel Cohen-Dumani:

Definitely. And I think the IT projects often, in the time we've been around, top bottom. Someone at the top says, we need to fix this, and then it flows down to the bottom of the organization that have to say, okay, now we're going to have a new ERP system. AI is bringing this flip upside-down approach where people at the bottom say, you know what, I'm going to experiment and make my life easier today, and I'm going to push it up because we can do that. And you've seen the adoption of AI.

There was a survey, I think, that came out about three weeks ago that was showing that 68% of employees are using AI without telling their boss that they are. It's staggering when you think about it. People have found ways to go and overcome all the blocks you can do to actually leverage AI because they realized it's going to make their life easier. We’ve never seen that before. I think, before, it was always- we talked about shadow IT for a long time, but it's taking abortion like we've never seen before.

Dayle Hall:

Yeah. It feels like- and I have heard the term shadow AI, too, built on the concept of the shadow IT, but it feels like there's definitely a bigger, more groundswell of support for people wanting to use this because I think across the business they can see value. So the 68%, that's a really big number. We can get onto governance and so on in a second, but how do you balance the need to move quickly and try things with putting controls in place? Because sometimes if you've got controls or like a council or a board that makes decisions, that can stifle innovation and which seems like the antithesis of what AI should be. How are enterprises managing that, and what have you seen work?

Daniel Cohen-Dumani:

The way I've seen it, and I think that lays out well to deploying Experio AI into organizations, is every organization think about, okay, if AI is going to access our data, how are we going to manage security and access to that content that's being exposed to AI? And that's true and that's complicated, right, because we used to come from a world where in a document management system, you will have access to the top folder and then so forth, like a hierarchical-type security. Those kinds of security models don't work well for AI. Maybe you can see the summary but not the detail.

So it's turning upside down how you think about accessing data. Our recommendation has always been, let's create an initial version of this solution, or let's deploy where everyone can see everything. Throw the value, and then we're going to come back and think about what's the best approach to secure access. Are you going to use the same antiquated method you used in the past where it's folder based, or are you going to think it through about, we have so much great content we can provide to our organization? Before, it was behind a security wall. Now we're going to be more thoughtful about what we can let people access to. So you have to rethink completely about this data security, what can see what, in the age of AI for sure. And that's a big job.

Dayle Hall:

That is, yeah. We're going to get into that a little bit more in a second, but just coming back to the data side and where to find the information, you've said that you can have initial projects where you know where the data is to get going and then really trying to figure out your overall architecture. For bigger initiatives and moving forward, is there a difference between organizations that maybe have a very centralized data structure or that may be using more decentralized, they have a data mesh set up where different departments can access different data at different times, less centralized? Does that impact how successful AI can be? Or does it just mean you just use it in a different way? How are enterprises managing the two?

Daniel Cohen-Dumani:

Yeah, an interesting question, Dayle. When you think about the different architecture, a perfect world is everything is centralized and ready to go and this clear definition and no data redundancy. Everything is great, right? But the reality is never that pretty.

I think it depends very much of the maturity of the organization and I think the size of the organization. The bigger you get, the harder it becomes to be fully centralized because there is so much data being consumed and created every day. And what I mean by data, I just don't mean database, or data like documents, image, videos, all of that is data.

So I think that we're definitely seeing both approaches. I think the data mesh is becoming more and more prevalent nowadays, just because of the sheer amount of complexity of centralizing everything and the time it takes to centralize everything. Smart organizations understand that we are going to create specific domains and create a mesh of domain specific. So we have the ordering department here. We have the marketing department here. And they all can be responsible to all their data. We're going to define a contract between those different data sources and some rules on how you can communicate. We're seeing more and more of that coming along. And I think AI is driving that move because you want data and you want data fast.

Dayle Hall:

Yeah. Again, I think while they may have been some reticence to allow access to other parts of the business, they didn't want anyone else messing with it or whatever if they've already set up a decentralized function, AI, the opportunity, the only way we'll really get benefit is to allow access to it. So I think that is helping.

Do you think that building agents or LLM-driven applications, is that changing? That's essentially what you're saying, is that's changing the debate around data architecture because people know that they need to have a different setup or they need more access to data to really benefit.

Daniel Cohen-Dumani:

Yeah. Agentic AI is definitely taking a step back and not in a bad way. In the beginning of gen AI, if you wanted to do some rag or some way of grounding a model on your data, you needed a centralized place and we needed to tap into a well-organized, hopefully semantically organized data source.At the heart of Experio, we use knowledge graphs. So there's been a lot of talk about, how do you ground a model with good data? And knowledge graph has emerged as this driven technology.

Now what's interesting is agents are allowed to go on their own autonomously, go in your ERP system and get the information it needs. And as long as the agent is capable of determine with high probability or the right approach where to go get that data, now we remove the need for this completely centralized data. What we need is a good semantic representation of where your data is, maybe some common datasets that have been brought together, but you don't need to bring everything. When you want to get the detail, the agent will know on how to connect to this ERP, CRM, whatever system, to be able to go grab the detail of the data that was not available in the big repositories.

So agent is taking us a step back for the good way or relieving the pain or the challenge of having this big data warehouse. So now we can have small datasets, well organized, and then agents going and taking what they need when they need it.

Dayle Hall:

Yeah. I felt like- and it wasn't that long ago, a year ago, when generative AI really started to take hold, there was obviously some concern about access to data, hallucinations, and so on. I feel like we've moved past that a little bit. There's a little bit more trust.

Do you think most enterprises will fully embrace this agent model, particularly a lot of these autonomous agents that are going to not only just go get the data and give you some insights, but actually make decisions? Is there an openness for enterprises to do that?

You just mentioned yourself, it's giving us a step back to make sure that we're setting this up in the right way, the security and governance. But are people embracing autonomous agents as much as they did with gen AI, or is it a little bit more skepticism?

Daniel Cohen-Dumani:

There's definitely skepticism, but there's also hype on agents that we've seen in 2025. I think there's still a misunderstanding of what an AI agent is. For some people, it's just workflow automation. An AI agent is not just a trigger that gets triggered when you receive an email and do something on your behalf. That is a workflow. That's been doing that for 15, 20 years.

I think an AI agent is more specific. It can define a plan. It can achieve the plan autonomously, and it can determine what tool to use to achieve the plan. And that's including calling other agents to do things on its behalf. So agents are really more complicated than most people understand it. And for that matter, the debate is still on. Most enterprises are now saying they're experimenting with agents. I would say most of them are still trying to automate more than agent.

Dayle Hall:

That's my point. Look, I'm in tech. I'm a marketer. So we love creating a story and we love saying all the things that's going right. But when I see a big CRM vendor say they've got 10,000 agents in production, I just don't buy it. That guy, Marc Benioff, is the best marketer on the planet. I give him that.

But it's not the reality of when we talk to customers and prospects, when we talk to our partners. Yes, people are experimenting, but it's like there is so much hype. I feel like those kinds of statements don't actually help people to actually trust what's going on. It's almost like they do it to just create FOMO. Everyone's got to jump in and create something. But I feel like that could have a negative effect. If they don't get these wins, if they buy into the hype, they don't see business success, it actually could be really detrimental.

Daniel Cohen-Dumani:

It is. And I think we're seeing it. It starts with this semantic misunderstanding of what an agent is in the hype driven by Salesforce, and everyone that's doing agent, that's claiming usage of agent. When you drill down and look at those agents, they are not agents. They are workflow with LLM enabled, which are just, okay, we use an LLM as part of a workflow process. Fine, that's not an agent. But they call it that way.

I think that we're still in that year of transition. I think we're going to see in 2026 and beyond real agentic use cases that solve real business problems that are truly implementing the concept of an agent. And yes, some organizations are doing it very successfully. But a lot are failing, too, failing to realize value.

Dayle Hall:

Again, I'm not mentioning the company names, but there's a CEO that came out recently and said that- first of all, he said was going to get rid of himself for a generative AI market, but said that he was getting rid of a lot of people because they were going to have agents to do it, and then six weeks later had to backtrack because you can't just throw this into the mix. And I think this comes back to then the kind of output and balancing risk and governance.

So as we've talked about who is interested in the AI models, it's both the business side and the IT side. And in general, IT usually is responsible for the data. But with AI being more prevalent in the business, in the functions, who's the best person to own data, particularly, say, in a decentralized model? And how does that really affect the accountability of the business unit or the IT team to make AI successful? Who is ultimately responsible to make this work? 

Daniel Cohen-Dumani:

It's definitely the business side that has to be responsible. In my point of view, it's rarely the case, because they're the closest one that's actually producing that data. They are entering data in a CRM system, or they're creating documents or videos or whatever, you name it, in terms of data. And they have to govern that data generation. And they have to own where and how it's being stored and leveraged, which means now you're seeing more organizations that have blended IT and business function where they have IT people embedded in the business, helping them achieve certain things like owning the business rule on decentralized data, and governance around that data for sure.

Dayle Hall:

Yeah. In your experience, as you've talked to people, what are you seeing as the biggest challenge, risk, issue with a distributed model when you're thinking about governing the data to make sure it's feeding the right models? You don't have to name any companies, but what are the biggest challenges or what fails are you seeing across some of these customers' mistakes they're potentially making?

Daniel Cohen-Dumani:

Yeah. It gets back to balancing the centralized versus decentralized. I think if you decentralize without a contract or underlying governance, then that's going to fail because every department is going to do whatever they want. Everyone's going to start holding their customer data and start adding fields and not talking to anybody else. That is sort of the risk of decentralization, is things get vogue, and then when you try to put it together, you're going to have duplicate customer data, two addresses for customer.

I've seen that with a very large organization I met with three months ago that says, we're struggling just because we have all those systems, we're not able to reconcile customers. We know we have customers in those three systems, but we don't have any unique identifiers on those three systems. That's an example of decentralization went wrong. The business owner implemented a system without talking with anyone, and now they're left with an algorithm bringing it back together.

So if you decentralize, you have to put a clear definition of roles and responsibility, rules of the game, because it all has to mesh back together at some point. So think about meshing in the world of WiFi or wireless. If people didn't have a standard, you couldn't mesh. You can have your own access point like serving 10 clients, but it couldn't mesh with the other one. And why they can’t mesh, because there is a standard that was defined on how to connect. So it's the same thing.

Dayle Hall:

Yeah, it's a good way of thinking of it. I came from wireless technology in my past, so I definitely understand that.

Moving away from the technology side and the data structure, how it's architected, let's talk a little bit about how organizations, are they culturally ready for some of these changes? Because I feel like it's a hot technology, we want to take advantage of it, but sometimes it doesn't necessarily come down to certain groups that are more risk averse. But it's definitely a cultural shift to allow access to certain pieces of data, to get IT and business functions more collaborative for the ultimate goal of the business to leverage this latest technology like AI or agents.

Do you feel that because of the potential, the organizations will just naturally allow their culture to allow this to happen? Or are there things that you see that an organization needs to change their mindset or change some of those cultural values to really take advantage? And can they actually do that? Would that hold them back from truly taking advantage?

Daniel Cohen-Dumani:

I think it's not as much as culture as it is the readiness of an organization to embark in more decentralization. I think there has to be a willingness. I think you find some organization where IT is large and very powerful and very gated in terms of keeping a hold of this and preventing other folks to even do anything, creating major roadblock.

I think the culture has to change with the IT at the top, saying, we know that we can handle everything. We know that data and software is deployable now faster than ever, including building your own. You don't need to buy. Sometimes you just want to build it your own. We have to enable this innovation coming from the business side as fast as we can, and IT becomes the guard there. We're going to set up the wall, we're going to set up everything. This is a major cultural shift, I would say, more on the IT side.

Now on the business side, of course, people have to be trained and receptive. So think about literacy on data and AI. I think you need to have some of that literacy and understanding within the business so that they can be involved in understanding how they're going to connect with the centralized function as well.

Dayle Hall:

Do you have examples? What does good look like? What does making sure that there's accountability across the business, that they have the right systems? If someone was to ask you, what are the three to five things that an organization should do, and again, it was great if it's a readiness and not really a cultural thing, but what are the three readiness things that the organization really needs to think about? Again, if you're listening to this, what are the three things that the person's going to go, yep, I'm going to go and ask my business now, these are the three things?

Daniel Cohen-Dumani:

I think culture is still important, Dayle. If you're a control and command organization, it's going to be really hard to shift to a decentralized model where we're going to let people experiment. If you’re a control and command organization, you're going to have to adjust to allow some level of autonomy across your business unit and departments and division.

The second thing is it has to start with empowering business folks to be more like IT people, to be trained. It used to be a clear separation, your IT or your business. I've seen smart organizations say, we're now putting IT people inside the business organization, reporting to the business owner. In this dual relationship, we’re IT, but we're in the business, we know these things and we know how to do it well.

Now this is where people are moving. They're a nimble team with some IT folks in IT so that they can do their own things and move fast. And then the third thing is overall governance and control and training along the organization to be able to handle this paradigm shift. But we're seeing it with AI. I've been in countless organizations where we've seen this team of three business users went on and created an app using this low-code, no-code tool. It's awesome and it's mission critical and it's working. They did it without IT, which was the problem because now there's no governance, there's no control, there's no liability. But you've seen other organizations allow that, and I think that's what we're seeing more and more.

Dayle Hall:

Yeah, that's good. Okay, we're coming towards the end of the podcast. What I want to finish with is something that I generally like to finish with most of these podcasts, which is really getting your insights on what you think looking ahead. What's coming down the pipe? What are the things we're going to be excited about?

But in general, what are enterprises going to be thinking about? If you're a leader of an enterprise, how would you advise them right now in terms of structuring, not just their data, but the governance, how you bring together these functional and IT expertise? What would you say to them around preparing for the future to really be able to take advantage of this AI wave?

Daniel Cohen-Dumani:

I think you have to think about it in several ways. We talk a lot about centralized versus decentralized model. I think enterprises have to look at this decentralization option, look at meshing data moving fast, and making that shift of embedding IT folks inside of their organization, inside the business organization. I think that's going to become critical as we start embracing agents and agentic AI to have a good balance on agility, speed, with governance, and I think it will become sort of a governing body of decentralized capability across the organization. ut of course, it's hard to say. I think it's going to matter to see enterprises adjusting to the demand of their business. 

Dayle Hall:

That’s interesting. And is there something, I don't want to say a myth, but is there something that you hear from talking to clients, customers in the industry, is there a myth out there that you want to debunk? We talked about agents are everywhere already, and I think the reality is there’s a lot of testing. There's some, but it's not as ubiquitous as we think. Is there anything else that puts someone's mind at ease, don't worry, you're not way behind because everyone's still testing? What is it that you would advise leaders as, don't worry, you're not behind the eight ball yet?

Daniel Cohen-Dumani:

Yeah, there's a few myths. I think the first one will be, it's okay, we can wait, we have time, it's all going to be okay. I think that is a myth that we see a lot of organizations thinking that they have time to embrace, and that is something that they need to debunk quickly. They don't have time. They need to get on the bandwagon and move along.

The second myth is for those people embracing AI, that the power of those large language models will keep exponentially evolving. I think this year was very interesting. We've seen new models, but I cannot tell myself as a user that it's so much better. I think agents have become better. I think agentic AI and deep research have gotten better. But the models themselves are just not exponentially better from GPT nothing to 3.5, 3.5 to 4.0, 4.0 to 4.10, whatever. I think we've seen a slowdown.

I think until there's a brand-new architecture that goes beyond the transformer, I think we're going to see just a slowdown in exponentially new breakthroughs in models, although, who knows, right? Nobody can predict that. But it’s sort of my point of view. That’s what I've been seeing over the last year, which I think gives a little bit of time. So don't think things will just get that much better faster. It will get easier, that's for sure. But there's a myth that thinks exponentially is going to get better like that. And I think that that's a mistake to look at it that way.

Dayle Hall:

Yeah. And is there something that you are really excited about, something that you're looking forward to seeing become more prevalent, more ubiquitous, something that within even your own experience, your own industry, that you're excited to see where AI agents could go in the future? 

Daniel Cohen-Dumani:

Yeah. When you think about any business, is this ability to become a true copilot or coworker of any human where you can delegate truly tasks that you don't want to deal with. I don't think we're quite there yet. I think it's still manual. But the fact that you could just have an agent that says, you know what, take care of this for me and give me the final results when you're done, and they can understand everything that you say in a very short period of time and with voice, hopefully, I think I'm excited about this. I think we're going to see more and more of very small autonomous agents that work really, really well emerge. And I think people are going to get excited and start building on this to get to a point where, hopefully, everyone has an army and agents at their disposal to really do the busy, boring work and focus on the creativity and interesting type of work.

Dayle Hall:

That's great. Look, I appreciate your time. It's been super insightful. A lot of the podcasts that I've done recently, very much AI and agent focused, but I think sometimes you underestimate the impact on the data side and how it's set up so it's absolutely a mission-critical part of making this success. So Daniel, thanks for joining us on the podcast today.

Daniel Cohen-Dumani:

Thanks for having me, Dayle. It was a pleasure and great conversation.

Dayle Hall:

Great. Let's reconnect in 6 months’ time or 12 months’ time, something like that, and let's see what's changed. I have a feeling this is moving so fast. There’s been a lot of change.

Daniel Cohen-Dumani:

I would love to.

Dayle Hall:

Yeah, great. Thanks for being part of the podcast. And for everyone out there, thanks for tuning into this episode. We'll see you on the next one.