From Pilots to Impact: Scaling AI Across the Enterprise

[Music] Thanks everyone. Um, my name is Kumar Krishnamorti. Um, and I lead the tech strategy business for PWC. Uh, you know, it's, u, it's a great time to be here. I've been having a chance to uh listen to some of the discussions uh in the morning and a lot, lot of what we want to talk about today is the pivot from, we AI is certainly the topic to jour, but how do you scale enterprise with AI? And um, there's some interesting analysis. I'll set up the opening and I'll introduce uh the panelists here. We have three great panelists, but we've been doing some research and if you look at earnings reports, market analysis, CEO statements, the number of mentions of AI is going up year-over-year, more, more often, and quite a bit, right?

But then when you start looking at the spend patterns, now put aside high-tech companies, put aside a few um energy and utility providers, but when you look at the spend patterns, the capital expenditure, which I, I'm speaking to a group of very highly um smart people, but capital expenditure, which is a signal, it's one element of a signal that's largely remained flat across all industries right now. I, I think this gap tells a powerful story. The gap, the story is now the decision on what to do with AI, how to scale enterprise AI is less about the potential and more about the value creation. Right? So how do you create value with AI? We, to really create value, there's a lot of discussion around use case purgatory and it's about going more than use cases and the notion that you have to drive integrated portfolio choices. So you have to connect the dots, you have to orchestrate across the portfolio. Integrated portfolio choices is becoming real. What does this mean? How do you improve margin expansion? How do you drive customer retention? How do you think about revenue growth? Right? I mean, these are topics that enterprises and businesses deal with, and that's got to be the axis of emphasis on how you think about it.

The last point is, all of this is in a change that is happening in a very, very jagged frontier, extremely jagged frontier, right? We talk a lot about the speed at which model makers, the Open AIs, the Anthropic, are releasing products. They're releasing products like four, five times a year. We can't keep up with some of the uh releases. You got software makers taking some of these and embedding it into their product suites, SAPs, the Oracles, the Salesforces of the world, and they are releasing products about, you know, once a year. And then everybody in this room, the leaders in this room, you're dealing with a technology landscape that's built over years, uh, layered with functionality, layered with supporting the business. Now you're trying to adapt all of those changes into your environment. So you've got to think about how to drive that change as you scale and transform. And finally, the reality is people have to change. Each one of us, last two years, we've been using OpenAI. I had a very interesting conversation earlier today about how our evolution has happened. And I, I'm a much better user of AI today than I was two years ago. But I'm an AI adopter, right? How, how does that change percolate all the way through the organization? And finally, we have regulatory bodies. We heard about the health discussion. They probably evolve at 2x, right? It takes 20 years before regulation upheavals and in some cases even not after that. So in this notion of environment that we are dealing with, uh, h, how do you scale AI and how do you drive enterprise transmission? That's the topic we're about to talk about.

Let me quickly introduce the panelists. Uh, Roger Barger here is with us. Uh, he's from Oracle, um, SVP of OCI. He runs the AI/ML team at OCI, and as you know, OCI is creating quite a bit of splash in the market, so very, very interested in hearing from him. Um, Sanjay Sriata, he is the um, Chief Digital Strategy Officer at Genpact, but that's just his daytime job. Uh, he's also um, runs a venture firm, Massa Group. Um, he's a prolific speaker. He's, he's a guest lecturer at universities like Kellogg and Wharton. Um, he's, uh, he's part of several think tanks. He's a mentor to CEOs and startups. Super curious. And finally, Dimitri, congratulations. Um, he's a finalist here. Um, and um, Feroh, Ferrovial, which I'm sure you've been introduced a couple of times to what they do. Absolutely interesting. But what I thought was interesting about Dimitri's background is he, he is in an executive position, but he brings a range of experience. Right? So he has been uh a strategist. Thank you for that. Uh, he's been um an entrepreneur. He's had global roles, right? So he's, uh, comes from a very global background and um, he's also on the advisory board to several uh places like the CIO um council for the Wall Street Journal. So I'm very, very curious. So Dimitri, with that, perhaps I throw the first question to you. I started the introduction from that side and I'll start the questions from this side. Um, perhaps, you know, just a little bit of commentary on how do you scale AI from the way it is, the use case everybody talks about it, but, you know, building it big and scaling it big is, it's a little bit different challenge. And as a leader in an organization, how are you thinking about it?

Okay, to mention was, I was seeing the picture of Bill Pappas. I'm covering Bill. That was originally in, in, in this panel. I'm from Greek to Greek. I'm super happy to to to to give my opinion. Will be probably in tune with his vision also. Um, you're right. No, we have a very complicated role at this moment and, and I believe never till now, we have seen so many different layers at the level of the organization running such a different speed. And this is something that we need to first of all, dest. Do we want to maintain that taxonomy of activities running in an asynchronous way simultaneously? So those that are digital advocates are exploring, are not doing shadow IT. They are coming from an idea to prototype in a matter of minutes. Yeah. And we can't probably scale that to the entire organization. We have others that are understanding the potential impact and they are really early adopters, and others that are um scared and quite concerned about the role and they are progressively adopting with a different pace and speed. Um, that is forcing us to identify which is the method and the structure that we want to use to lead all the adoption. Absorption is at different rates. It's a different rate. No. And what we want to do is to not reduce the speed of the entire organization, try to create inertia from the bottom to the up, but at the same time, do not lose the control. We don't want to put guardrails in all the decision in order to restrict the possibility to discover which are the potential opportunities, but at the same time, we need to gain traction from the bottom, otherwise we're not going to be able to implement the transformational use cases. No, this is something happening right now at the scale, modifying the way we are framing the portfolio management in large organizations. We have the typical portfolio running the digital transformation, and we have allocate on top of that a new layer of initiative that are represented by a technology. And if you are not preparing the digital transformation scheme, you are not going to be able to absorb the potential of the augmented transformation. New strategic framework, no? And that's a, theonomy right now is costing a lot of opportunities because we need to synchronize that with our partners. I'm seeing a lot of colleagues, colleagues, even our, our own experience, we are in many cases much more prepared, construction company, than many of the partners we are working with because they have not taken the decision to manage their portfolio of services in those three different speeds. They are waiting for a consolidation of a new service model that will adopt IDs uh to develop the software while we are doing that at the scale and we took the decision and everybody is associated to to that um potential improvement. And so, it's a dilemma where we need to associate the opportunities with different layers and with different speeds and understanding that we are going to create friction in the way many of our employees and department are interacting with it, with other business because everybody is not synchronized yet. Yeah, that's wonderful.

Sanjay, uh, I'd like to come to you next. Uh, picking up on what Dimitri just shared, and you have a very unique perspective. You, you sit with um, Genpact, right? So you see the transformation and the layers of transformation that uh, Dimitri talked about, but you also bring a very unique perspective because you're also advising think tanks and startups and you have your own venture fund. So you're seeing the shades of transformation at different speeds evolve. And um, h, how do you help and why the delta? First, it's great to be here. Thank you. And um, thank you for the opportunity. Um, I'd kind of give you a balanced view. So I give you a very optimistic view from having um, from being from participating in eight startups now that are all in the AI, data, and agentic AI space. So really see a very future-looking view of the world, albeit small. Um, and the work at the think tank is with boards and CEOs and large company CIOs, COOs, and also strong believers, a lot of focus, a lot of investment. Maybe before we go to like, why the world is great, just give you the an, the perspective on the question you raised, the delta that you see as an operator at Genpact um, and by backdrop, we run business processes for corporations across verticals, financial services to consumer goods, life sciences to healthcare, manufacturing, etc. And in so doing, we have to go in and uh, essentially underwrite the work we do with data, tech, and AI. And so in many ways, we are kind of, I'm operating in a fractional CIO role, not for the whole company, but for the slice of the business that we run for them. Um, and so I'll give you a perspective from that that maybe answers the question on why, where the, where the bumps are. And I'd say there are three big things I've learned. Um, or we've learned. Um, I think the first one is, uh, AI, generative AI, now agentic AI, I think has fantastic potential. Number one stumbling block is data. I will tell you that we, most companies I work with do not have data at the level of foundational strength, a cultural asset, and cleanliness and hygiene that allows them to be able to leverage, uh, what will be a commodity eventually. So I'm not going to be any better user of a model than than Dimitri will be. But my differentiation will potentially come from the fact that I use the model, which he can get access to as well. And I'm just making a point here. But I can use it to orchestrate around my data, which is my book of business, and I have a lot of stuff. Then getting that ready is a big issue. And in fact, now I've almost gone to the point where if the data is not ready, I don't even start an AI or regenerative AI project. So that's number one learning. I tell you, the number two big gap is, uh, operating model. And what I mean by that is, and I've been in the industry for a while. We've, we've all done this for years now. You know, you think about the automation that we all did. We take an end-to-end process, you break it down into its parts. You take every single part, you automate it, digitize it, you slap them all together. You got something that's faster, it's better, it's potentially cheaper, it's definitely more scalable. It is ideally higher quality. And that works. And there's some work that's remaining to be done. It's similar to the work that was done before, and human colleagues can kind of attack it. I think what happens when we apply generative AI though is it actually fundamentally changes the process itself. The work that is done ends up being different from what used to happen before. The way it connects upstream and the way it connects downstream is now entirely new, and the work that is remaining for human colleagues to do is an entirely new workload. And so if we don't think about an operating model con, at the same time as we're deploying some of these technologies, we've got a great uh implementation done, but no economic returns likely because in the end, returns come from good utilization. And so maybe that's the second big, hit us in the face in a bad way, kind of lesson we've learned. And now, you know, I don't even uh kick off anything till we actually have a, you know, an a CHRO in the room and actually have a discussion how we think about operating model. Number two. And I think the third thing would be scaling. Um, I think scaling is an entire journey. Much has been spoken about it, so I don't want to go there too much. But the reality is that there's a big disconnect for most Fortune 100, 500 companies from the board level and the C-suite, all get it. And actually, the early career professionals that came out of colleges in the last couple of years, actually beyond getting it, they actually live it. And so against those two thin ends, the big middle is actually quite uh lost. And I don't think you can just walk away from that and say, well, people pick it up over time. I think going through a very thoughtful mechanism of skilling and reskilling to be able to use this technology is part of the journey. And we've had relative success and obviously a lot of kind of bumps in the road. But I'd kind of just uh, Kumar, just elevate those three things. I think getting data ready before you start AI, number one. Really thinking through operating model at the same time you're actually deploying AI solutions, number two. And then actually thinking about skilling and reskilling as a big and an involved part of deploying AI. And I think if you fix those three things, and actually look, tech is no longer the long pole in the tent. I, I love uh the components as you broke down and I want to come back at some point and uh explore a little bit more on the operating model and the orchestration point you talked about. But Roger, if I could, you, you're now picking this up from a very, very different vantage point. You, you're the model makers. You, you're the ones that are putting out the different releases, the the components, but then this all has to kind of flow into the enterprises for the value creation to happen. Uh, would be great to get your thoughts and how you think about this and, you know, where some of the technologies headed to help with some of this as well.

Sure. I think, I think one of the things I'd like to hit on first, I mean, the elephant in the room, some 78% of AI projects fail or stall at some point in the enterprise. And if you start to ask yourself the question of what happened and why are the, why are the ones that were successful actually successful? And it's actually not the tech itself, it's about the people and how the tech was interacting with the people. And I'll just share a couple of examples within my company for both what we're offering to our customers, what we're using inside the company for ourselves, and try to highlight a common thread. We're a data company. One of our best assets are our databases. There's obviously a set of highly skilled customers who are able to use SQL or to do use our analytics projects. Um, these individuals are highly sought after, highly in demand. But as we look at AI, we can actually use natural language to SQL, natural language to analytics that allow people to actually interact with their data through natural language. It incredibly expands the user base and it fits naturally into the workflows that they were already doing. An internal project, we're also a software development company. We have a lot of software engineers, many at various stages of their um, career and skill set. Obviously, the most junior engineer would like to be on par or to hold their own with their much more senior engineers. So we have a project called Code Assist. And it doesn't just do coding. It does debugging. It actually generates documentation because frankly, coding is only about 20% of what engineers do. A young engineer can be looking at a project spanning hundreds, if not thousands of source code files and ask natural language questions like, what is this package do? How many times is this module called? It gives them these superpowers, but it's actually fitting naturally into the workflows that they're already doing. And so what I have noticed, what really makes these projects successful, either for your customers or inside of your company, is does it fit naturally within the workflows they're already doing? Is it actually augmenting the work that they're doing? Just giving them superpowers without completely trying to disrupt it or transform what they're offering. To me, these are the elements of what makes a a key successful um AI project. Um, user experience, I would want to get into that. But also, if you look at some of the projects that are that are value in the billions, is it any, are they doing any different AI? No, it's the user experience. It's actually fitting and working well with the individual. I think the greatest bar that we have to, and I've got, I've heard many good points here from my panelists, but if you want your projects to be adopted, look across your enterprise, see where the labor-intensive workflows are happening today that's actually bogging down your people. Make sure you can understand what their workflows are and how to naturally complement them with the AI we'll be talking about here today.

You and uh, Sanjay brought up an interesting thing um, and we should, u, I just wanted to probe into that a little. That you talked about the notion of reimagining the workflows, right? Sanjay, you, you articulated in a way that saying, hey, what you do today, it's not the same way we reimagine the workflows. And Roger, you, you kind of highlighted a set of reasons why IT projects won't fail because, you know, there's a set of tools. Perhaps I open the question first to you, Sanjay. Is when you think about reimagining the workflows, there's an aspect of innovation that kind of underpins that. You have to be able to innovate the workflow. What, how, how do you think that change has to happen? I mean, uh, what, what needs to change for that innovation spike to happen? And then Roger, I'd like you to respond to that if you don't mind.

I, I just want to start by saying that this room is an excellent example. I, I would suspect and I'm assuming you are in kind of the role of the chief, in some ways, the transformation officer in which your titles are. And part of that role is to, is to really help your peers and your management teams and your employees understand how AI fundamentally changes the course trajectory you're on. And I think this is such an important part of what, what uh, what you do, uh, and we have to do on a day-to-day basis. You know, I'll give you a couple of quick examples that come to mind. Um, I was in Singapore meeting with one of the largest banks in Asia, and they, uh, the chief operating officer of the bank told me this. He said, "Say, we've had a sales department for 42 years and we have a services department for 42 years. Very different workload kinds of things they do, specialized, really top-notch, and they're one of the largest banks there." And he said, as it turns out, you know, we deployed an agentic AI PC, and when we ran the agentic AI solution, what we realized is one agentic AI that covers end-to-end, so start of sales all the way to actually completing the servicing, actually not only can do it, number one, uh, but actually if I have one agent, actually does a better job than if we break it into two parts and have one in the sales department, one in the services department because you can imagine continuity of data points, relationship, all the bits and bytes you need to remember from the first conversation, right, all the way through till the transaction gets. And so they're having a conversation of saying, listen, uh, number one, um, 42 years notwithstanding, should we be thinking about collapsing our sales department and services department at the bank into one? Because in the world, uh, in the future of the world, we will be moving to agentic. That's it. It's very clear. The world is moving towards that, and we'll be part of that journey. And when that happens, is our organization structure set up today? And I think just more broadly from there, you know, companies are trying to think about, should we be organized by products or by workflows? And so what is the future of the organization structure actually look like when you actually have agents, uh, AI agents running around in the workforce doing a variety of different things? Who's, what is the role of the CHRO who's actually looking after human colleagues like me versus the CIO that's administrating these agents, managing life cycle for the agents, when do you hire, when do you terminate, when do you fire, what access controls you give it, which version of Salesforce or Oracle, you know, admin level versus user level they get in. All of that life cycle management today is the CIO's role. And you know, today we've got 1000:1% agent to one human employee, that's kind of the ratio where it's at, maybe. I mean, this is going to change and it's going to change very fast. And when you get to a scenario where you're 1:1 and then growing to 1 to 10, that inter-agentic economy, that agent-human workforce, all of that has massive implications on, I'm just talking with a C-suite structure. And in fact, what does a CHRO do versus what does a CIO do? So I think these are matters that are different for different companies. There is no one-size-fits-all answer for all. But I think the, the path of action, the direction of travel here is to sit down and really bring out what agent AI can do or generative AI can do or what AI can do for our corporation and really help our peers understand that. But then the job doesn't stop there, or to just deploy the technology. The job continues to actually drive the transformation and operating models and, and actually the scaling and in structuring the workforce.

Well, that's great because you're now starting to say, if I, it's like unspooling a ball of yarn, right? And all of a sudden it goes in different directions. Dimitri, I'm going to let Roger respond and come to you to say, as enterprises, when you're thinking about a scaled enterprise transformation and AI, how do we pick this? But Roger, first, I'd love for your thoughts on this.

Yeah, I'll just build on the answer that we just heard. Um, first and foremost, a lot of our organizational um, structure is a function of this, the tasks that individuals had to do and, and the skill sets that they had. As we introduce these new tools, these tools can be used by a broader class of individuals who may very well merge groups together because they, they, they themselves can actually use these agents to carry out these tasks. So I do think we will not only see individual business units merging into a single augmented with AI, again, hopefully in a way that's actually natural for them. We'll also see new job families arise. This is what's going to be interesting longer term. Everyone, we've seen data scientists and machine learning engineers in our group. Soon we'll see prompt engineers. We'll have security experts who are very keen on how to actually keep these agents secure and other aspects as well, which we're still just tracking ourselves. But these will become the new business units that actually manage this technology and actually with completely different workflows.

Wonderful. So Dimitri, just, u, to bring this together, to you, as you think about, like, you know, this gets more and more complex as we kind of like try to unwind the spool of yarn, right? How do you, as, as an executive in an organization, you're now trying to orchestrate this across as a transformation, you're capital planning, getting the, the people ready, getting the change ready, um, and by the time you get the planning done and get it through the process, technology has changed underneath you. So you're dealing with that as well. So how can you share some thoughts on how you think?

It's interesting because we, we have deified in the last two years many of those frameworks that we were adopting in order to scale properly um, and bring the technology to any place in the organization. No solid foundations by the level of the platform or at the level of the data because many of the foundations from the perspective of the stack that we were using has been evolving along the last two years. Um, we need to be very, very conscious about which are decisions that we are taking from the perspective of that layer. So what we are trying to do is define a common understanding of which are the principles that we are going to use for the design. First of all, are we going to be agnostic or not to the LLM? Are we going to have proprietary data um, in a progressive way? We are going to try to build a competitive advantage in that particular domain. That is something that we are going to try to scale across all the regions, across all the business units. Are we going to try to progressively identify if we are going to pilot from a make or buy? Because as soon as we are progressively increasing our productivity, probably we should not rely so much in third parties. Are we gaining speed or are we winning efficiency? Are we followed by the right partners or we have to completely reshuffle the way we are selecting partners at the scale? That's something that is an independent program that we are running right now. Maybe we need to go and interact with partners that are small enough in order to be able to adopt as fast as we are trying to do all those new technologies. Let me give you a very concrete example. I was giving guidance to my entire organization six months ago in order to adopt at the scale Cursor. Um, we celebrated that several months later, everybody was actively working with that platform plus GitHub Copilot. But a couple of members of the team started to work with Windsurf. We started to see all the uh results and all the analysis, and it was objective, quite tempting to jump from Cursor to Windsurf. It was a critical decision, not so much compared with the first one. In a couple of weeks, we took the decision to move everything. Could be a new product too. That's the point. So, we need to be frugal. And that component is not anymore piece of our foundational red line statement that we are going to spread to all of the world. I'm going to assume that in any side of the company right now, I have members of my team testing in a controlled environment other platforms that could scale rapidly and represent the standard. And that was not reality one year ago or six months ago. It's gonna be a moment where those elements are going to be stable enough, like for example, the MCPs or the new protocol to interact between the the agents that will not be the problem. Probably yes, but we are in such a volatile environment that many of the assumptions that we have used in order to scale the tech stack and to bring directional instruction are not solid enough. No. So we are trying to maintain very broad assumptions that are guiding us, assuming that we are going to have to rearrange that along the way and try to maintain as much as possible the proprietary elements in the value chain to to to not lose the control and to be able to shuffle from one reality to another faster than the market. But we assume that others that probably has been testing other solutions or taking the the most probably will have much more resiliency or capacity to react, no, to react fast and to not destroy what you have built. It's important in an environment where I assume that we are going to have most probably in less than two, three years, thousands of agents. The problem of operating at the scale, those agents is going to be much more complicated than operating 6,000 of RPA. Yeah, because what else is much more simple to be tracked and to be observed than to put thousands of cognitive elements in the middle of processes where probably you are going to change a policy. You are going to be affected by a new legislation. No, are you prepared enough in order to introduce a chain at the scale in your agentic platform?

You, you're raising an interesting uh point, not only about the agents, but it's the continuous management of those agents. How do you do that? I, I'll, I'll come to the audience in a minute, so think about any questions you may have had to the panelists. Uh, but I want to explore one more topic. Uh, that's a pivot to scaling AI in large enterprises, and it's the notion of technical debt that you're carrying. And it is interesting that the three of you come from interesting backgrounds to have a slightly different perspective. So Roger, I'm going to come to you first, right? Uh, when I look at it from Oracle, and you can talk about it both in Oracle and OCI, there is an inherent level of large technical debt in the clients you serve, in the companies you serve. You created some of it as well over the years, but, but that exists. And now the sea of change with the AI and ML that's coming, that is getting layered on top of it. Um, how are you thinking about that pivot of the change? Because that's another component that has to get broken through.

Yeah, sure. Um, first off, we've actually been putting AI in all of our projects and all of our products. I mean, Oracle Database was, is the first and only autonomous database. Um, thinking about how these customers are going to interact with it. Fusion has over 50 agents. So this integration of AI into legacy software has been happening already, not just because it was a nice to have. I mean, we truly view this is what our customers want and will demand. And if we don't do it, our competitors are going to do it. NetSuite has agents built into it. Um, we're offering agent software over OCI for our customers to build this. Um, so I think we've been able to actually move quickly on those fronts. Um, but what we are noticing, and I think the biggest legacy challenge, and maybe for all of our groups here, I don't know how you all are facing it, but enterprise risk, enterprise risk is probably one of the greatest challenges, the greatest areas to defend. As a data guy, I used to say that data has gravity and all of our tools and analytics will move towards it. Enterprise risk is the gravity which is going to draw all AI projects into its circle. And enterprise risk management, from security, compliance, uh, you name it, uh, monitor, monitoring, logging, that works in a much slower pace. Years, six months, we would be lucky, 12 months. And but yet we have to move fast as we've just heard. This field is moving so fast. If you want to stay competitive, I think one of the biggest areas of legacy risk is not technology, but it's the processes we've been using, using to keep our businesses safe. And you can't throw that away, that's not going to change. It's instead, how fast can you speed up the process by which these new technologies can not only get vetted, secured, and released to customers in a very safe manner? That to me has been the hardest thing. In fact, one of my mentors, Jim Gray, used to say, "May all of your problems be technical." This is one of those areas where hopefully all of your problems will be technical. Odds are they're not.

Wonderful. Uh, Sanjay, you pick up this problem from two domains, as a service provider, Genpact, but also in your think tank role advising CEOs and also um, in some of your new ventures. Um, so you can advise the whole spectrum of how to think about the tech debt um, and how to navigate through the change that has to kind of get embedded. Um, would love your thoughts on that.

Well, look, I think on on technical architecture or enterprise architecture. First of all, I love what Dimitri said, which is innovate hard and fast and actually even decentralize some of this. So you got a lot of experimentation, innovation teams coming up with whether it's Windsurf or whether it's Cursor, and you know, let the best kind of settle in. I love that idea. And also love the idea of make good, uh, have a, have a mechanism to make quick decisions on this versus that. So when you get to a fork, you don't let it fester on. You don't have 17 versions of the same thing. So I love that. And then, uh, Roger, I love the point about how we're incorporating AI into pretty much every product. Uh, just to like put another perspective on the table, maybe to get some tension and kind of different, different ways to think about it. Look, one of the biggest issues and enterprise archite, enterprise architecture, old phrase, completely new meaning now in the context of AI. And the reality is many of my um, our think tank members um, and I are in a boat where we have an AI that is kind of built to the level of, kind of we take a large language model, we put it behind a firewall, we put some metadata layer around it, put some retrieval augmentation, and that's like version one. We use it for like pay, uh, HR stuff in our own case. So you'll have other examples. Then we have, you know, application providers that are embedding AI into their systems. So Service Now and others, for instance, that are doing that. Then you've got like core system providers, Oracle, SAP, many of the larger companies, everyone's embedding AI in there. We'll have agentic systems which are going to come from either we're trying to bootstrap them ourselves or we're bringing in specialized startups. Right? The point I'm trying to make is, you, you talk about legacy debt from the past. Let's take a moment and realize what we're doing right now, if we don't do this thoughtfully, is going to create the largest debt we've ever seen because you cannot have 16 different versions, you know, at the edge, at the core, in the middle, on the side, and there's an agent for everything. And yet we don't understand how does the agent in in Workday on the HR side interact with the agent in Salesforce on the sales side, with the agent in the legal software on the contracting side, and then eventually with the procurement system and the sourcing agent, right? And having this thing work through is going to be become really issue, uh, become a big issue. And so the main thing I would tell you, and this is what we worry about at the think tank, is how do we think about enterprise architecture in the age of AI? And actually, this legacy debt is actually only going to increase because we'll end up, if we don't plan for this correctly. And by the way, the answers are going to be different for different companies. For some, it might well be, you know, using it in the core ERP. For others, it'll be, you know, let's keep the core thin and let's put the intelligence on the edge. And so there's different models out there. But whatever decision gets made, I think it's important that you make it consciously because if you don't, you're just going to have a AI sprawl. And between guardrails and guidance and regulation and compliance and performance and security and updates and versions, it's going to be a complete nightmare.

Yeah. Between, I, I think the point you earlier mentioned on when you think about reinvention, you want to think it in the context of what your enterprise architecture is and how you're going to navigate through that. And that, uh, integrated view has to come together. Dimitri, you're in the enviable position of dealing with the tech debt, right, while driving the change. Um, h, how do you navigate these conversations and the prioritization of this as you think about scaling AI at the enterprise level?

So, I think technical debt traditionally has been a burden that has limited possibility to adopt the latest technology. And it's interesting because with generative AI, we are taking advantage from both sides of the coin. If you have technical debt in areas where you should deploy AI, probably you are going to be limited. But at the same time, you can use AI at the scale in order to self-discover and self-heal many of the areas where you were approaching the problem with a traditional unconventional way. We are going to refactor the code. We are going to try to apply patches. So we can build augmented processes in order to improve dramatically the way we are facing our technical debt. You know, it's interesting at the same time that we are relying more and more. This is my opening Pandora's box in order to create a bit of section, provocative um, on average, 60, 70% of the startups in the tech domain after the first couple of years do not succeed for many reasons. This is statistically something that is a fact, sadly for for many of them. We are relying more and more in that tradeoff and in that race. We are immersed. Either we are using massive platforms, we develop our own solution, we rely on frameworks, or we deploy our ideas through a set of startups that that at the end could be a wrapper. It could be a solution that is using an open source, but many of them are not going to exist in two or three years. Are we taking into consideration the technical debt that represents relying your future augmented processes in input that not, not going to exist? Bad definition. Are we scanning that structurally? Are we making the proper assessment of whether it's cheaper right now, could be super expensive in three years from now? I think this is going to be a new wave of opportunity to reassign how we introduce in the portfolio analysis the risk of relying on third parties that right now have an amazing idea, the capability to prototype using this tool and bring to the market something that apparently is going to solve my problem. But in two years from now, nobody believes that safety, 70%, 60% are going to be there in order to be SOC compliant, to fulfill all the regulation, to be sure that they are fully aligned with the GDPR and to give us the global coverage that the environment forces us to fulfill in a corporate environment.

What you're saying is interesting because as an executive, the complexity of decisions and executing a strategy is kind of like exponentially increasing. Absolutely. Let me, uh, put this out. I see a hand there. Oh, we, we've got a few minutes for questions, but let me start right there. Can we get another microphone on this side? I saw a hand being raised. One. Not. Okay. So, first of all, Kumar, awesome panel. You saved the best for last. So I want to pick up on first what Roger said about 80% of projects fail. It's the dirty secret and it's all about people. And then go back to what Sanjay said about how, you know, we're living in a world where the agent to human ratio was low, but it's going to scale very rapidly. Okay, it's already scaled very rapidly in the startup world. We're seeing companies get to 100 million ARR with like five people. But the question I have for you is, we saw Merna, a local company here in about a week ago, two weeks ago now, announcing they're merging their HR department and their tech department. Now, are they crazy? Is this where we're going? Is this, are they just early? Have they been very early adopter? But is this the future of what we're going to see in the enterprise? Who wants to take this?

I'll start first, but with with Merna. Um, so I've attended some investment forums, some groups talking about the merging of HR um with the tech department. A lot of what HR does, HR, first off, is one of the most overloaded groups often in a company. Everyone's trying to hire the access to information. They have resumes. Resumes will be gone within the next few years. We will not see them. Much of what we need to know about individuals is online. It's on their LinkedIn profile. We actually get feedback on what are good LinkedIn profiles. The information is largely digital, and you, with technology, you can actually source good candidates based on the information that's out there. Yes, you might need a high judgment HR individual for executive hires and other areas where there's considerations which require a soft touch. But again, I think we mentioned it earlier. There's going to be consolidation where we look at the potential of what the technology can do, the inefficiencies within a particular organization, or how to enhance them. Bring those two closer together so they're serving each other's needs and more alignment just because of efficiencies that it unlocks. Probably looking at the workflow and looking at what the people do, like I said before, finding out what's inefficient and say, let's augment that with AI. So I think that's the first offer I'd like to share with you.

Maybe I'll just add to that quickly. I, I think it's true that some elements of the job will get combined. So life cycle management of agents and recruiting and managing employees, like that whole cycle is so similar. And a multi-agentic agent-to-agent and agent-to-human economy that will clearly be the case. But I think as AI does with pretty much any profession, it allows you an opportunity to step one level higher up. And so whereas a CHO or my CHO is a dear friend of mine, uh, as an example, spends all this time really worrying about 110,000 strong employee workforce, for once, I think he'll have the ability, and I'm just projecting out a little bit, as some of this automation comes into play, to really sort of uplevel, kind of the day-to-day compensation management, recruiting, the LinkedIn pursuit, all that sort of stuff. Go one level up. What is the culture of the company? How do we get an ethos in place? How do we actually drive the transition of the workforce? How do we get adoption of AI in the context of text-to-human workflows? You know, I was down in one of our um, board members, and we were in one of their stores looking at a humanoid, uh, deployment that they've got going on. It's amazing if you think about these like big warehouse stores, the ability to move things from up to down with a humanoid with no safety, just all kinds of considerations. And they were saying that actually what ends up happening is where the cameras are not there, employees actually go and just kick these humanoids over and they fall on the floor and break. So, there are real issues that I think have to be solved. And I just think that like with everyone else, the CHRO will have an ability to rise up to the next layer of value framing. And that's true for anything with AI. Does that mean they need new skills to be a CE? Oh, they will most definitely need new skills. There's no question about it.

I think there was a question here. Uh, yeah. Did he get a mic? Just trying to see if there's any other show fans I can come next. We can get a mic there. Okay, there you go. Can we get a mic to the lady? I think they activate from um.

Hello. Yeah, go ahead. Sorry. Um, one of the things I'm trying to understand in all these major changes is how the leadership role, how the CIO function specifically is changing. One of the things that helps me understand is are there specific use cases of AI that you and your executive function um, you know, have spotted as something that's selling? Like, what do you, what do you use AI for? And and what does that mean about how the, your roles are changing?

So I think we need to, in a role like the one I represent and many of you, we need to maintain a very active AI literacy. I need to practice, I need to test, I need to be able, uh, to to answer at a user level in order to demystify what I'm going to have to not impose, transfer in a gradual way to all my peers at the executive committee. Um, I need to sponsor at the same time, probably the most difficult and the most delicate use cases that will create friction at the level of granulation because those will require much more nesting stage debates, analysis where we are not talking about technical issues. So we are talking about which are.

The repercussion at the organization level of implemented that change and what I need to do at the same time is to bring to the entire organization with my team that mentality of AI first in terms of how do we manage resources? How do we put in motion opportunities in order to reinvent the way we are framing new businesses the way we are seeing growth to synchronize that is really important.

For example, if we are facing a potential M&A opportunity and as you can imagine an M&A opportunity brings a lot of departments under the pressure of the time in order to close a potential deal. If we have an augmented department trying to take the maximum of one particular use case that for them is going to give the answer on time and we have another department that are not using that we are going to have bottlenecks and maybe the deal is not close maybe we are creating those kind of asymmetries to equalize that AI literacy across the entire company requires a lot of visibility a lot of use cases to be showcased at the level of the company and to identify what is really um the opportunity for each department to represent like the pivotal point just to make them understand that they have to be self-sufficient trying to introduce that in the in the Berto's wheel of their own improvement it's important because we were talking before HR and the case of of modern know combining we have been along the years seeing how HR has been dealing with different type of persona with partners with employees working at the office, working remote, but right now human is not just a unique component. The cognitive component and how that cognitive component is helping in the department has much more implication in the productivity of the entire staff that many other partners in the equation. I believe sooner or later the owners of a specific type of agents that are providing transversal capabilities to the company are going to be more in the hands of HR or augmented resources department that in in my side because to define the policies to define how to equalize that capability across the company it's something that require to understand the impact of that capability in the rest of the human employees no and they have more tools to do that we'll we'll take one last question can I can build on that just just a little bit because that was a great answer and I I just want to just emphasize I have seen this happen so many times over a decade ago I was at Microsoft my team launched Azure ML Satcha sent a message down everybody is going to embrace AI and propose AI projects across the or it takes that kind of top down experimentation will people know right away absolutely not we actually set up a regular conference where ideas were shared every semester of successful projects so people could see the patterns a machine learning university was created by my team where every semester we would bring a few cohorts in and actually train them on the best practices. Most recently while I was at Microsoft recently Satcha did the same thing about Genaii. We see the same thing happening at Oracle. You can't culture change is hard and it takes multiple aspects to effectively carry it out. But if your companies if you want to get your companies there it does take a very concerted and focused and top down effort to try to move the organization forward. Yeah. Great. Great. Because Shopify did that about a month ago. Great.

So, one question and then I'll come to closing comments. Um, good afternoon. Thank you for being here. Um, SEPU at PwC alum and now swimming in the startup ocean. My, uh, and this is why I'm asking the question I want to ask now. Uh, Roger, I know that you've seen this firsthand from an industry perspective. Kamar, I know that you deal with clientele like this every day. My question is around executives that deal with rapid turnover specifically in the chief digital, chief data and CIO uh realm. Given the rapid turnover rate across industry, my question is uh how are you each of you tackling succession planning in your organizations and what is the legacy that you want to leave uh when you move on to your next chapter? That's a loaded question, but I I'll take maybe one of us. 30 seconds. My 30 secondond answer would be that my legacy or my goal if I'm successful should be that this role goes away that digital is integrated into the rest of the company that the IT is our senior executives think about it just as much as I do and a specialized function that's off in the corner doing something uh is not needed anymore. We're not there yet but that's the aspiration and goal. Yeah, in in the spirit of time, you can find us u sure afterwards, but um we can certainly have a follow-up discussion on this. But maybe to close this off, thank you all of you. Maybe a rapid fire question for each one of you. If you were sitting here in two years, um can you think of one thing that AI is going to change in your organization? And then the second part of it is what are you worried about how AI is going to impact your organization? Maybe I'll start with Sanjay, you first, Roger and Dmitri.

I think for I think for me I just leave it this. I'd say two years from now I want to be one to five. So I want to have five agents for every employee in the company. Okay. That's a very aggressive goal. Um, and it's going to strain us at the edges. Um, and then what was the second question? What are you worried about? Oh I'm worried about everything. Uh, the inter agentic economy technology is not there. Life cycle management is not there. The handoffs between agents. There's no mulesoft for agent the way it is there for SAS. What is the future of SAS by the way? We have all this legacy debt in SAS. Is SAS really going to exist? So there's a number of questions to be asked. Wonderful.

Roger. Wow. Um, sorry I'm stalling. Sanjay stole all the good answers. No, that last question. I So what what I expect to see I expect to see software development is going to largely change. People who are very well-versed liberal arts can actually speak very eloquently about the product that they want to build will now have the ability to be builders. And I think that's actually exciting because now you have anybody with an idea or creativity can actually make this happen. I think what concerns me the most is major companies are not going to move fast enough. There's incredible investments in SAF SAS software which is about to be radically displaced by agents. And so great companies, great people, great technology if they don't move fast enough and embracing this and thinking about how the inter this software will be interfaced in the future could could be displaced and that represents a great loss of value. That concerns me.

An easy one. So um my guess 5 years um the overall cycle of development of our activity will be run autonomously and we will be able to simulate any piece of infrastructure before putting any brick on the ground. That's going to happen for sure. We will have a digital proxy of our organization fed with the intelligent of the past and maintaining that intelligent for the future. Um most probably we will have all of our supply chain. That's a hope. That's not a belief. It's a hope. All of our supply chain will be connected because we have been able to democratize this technology to the SM and means of the world. Wonderful. I know we are a little over time and unlike the Oscars, they couldn't sound the music on us to shut us down. But thank you everyone and thank you for your attention. [Applause]