How To Take Enterprise AI Pilots To Production — With Russell Jukes

0:00 Why are so few AI pilots moving to production? And what's the best way to get them there? Let's talk about it with DXC Technology Chief Digital Information Officer Russell Dukes in a conversation brought to you by DXC. And Russell, it's great to see you. Welcome to the show. >> Good morning. Good morning and thank you for having me. >> You bet. All right. so, you have told me in the past that what we're seeing in AI today is kind of like watching other people's Instagram pictures when they're on vacation. When people talk about what AI is doing for them, it seems great. It seems like they're living their best life, that nothing could be wrong. You read the press releases, you read the hype. AI is solving problems, it's curing cancer.

0:40 you hit a button and all of a sudden all your problems within the enterprise are solved. but then, when you actually speak to people about their real life, ain't as pretty as as those, you know, so-called Instagram pictures are portraying. So, just give us a look at the reality of what is going on with enterprise AI today and why you think there's this disconnect between the public messaging and the fact that so many companies seem to struggle to get their their pilots out to production.

1:10 >> you know, there's lots of answers to that. Like lots of answers to that. But we should start with it didn't get any easier. Like the the technology landscape, we didn't switch A for B. And suddenly A has gone and now we're just dealing with B. We added B to A. So, we added AI. We We layered Yeah, we layered AI into our applications. We layered AI into our processes. We layered AI into the SaaS applications that we buy from some of our great partners.

1:42 And so, we layered it in on top of what we already had. The headless thing is another a great example. Let's go headless. Okay, well, nothing went away. The UI went away, but nothing really went away from a technology standpoint. And so, we got into a real situation of well, we now just have more to manage. And with more to manage, drives more complexity. We'll talk later, I'm sure, about, you know, is there going to be a AI legacy debt? I'm sure there will be in the in the future.

2:12 But, it got more complicated. And so, if you can imagine, you you had, let's say you had 300 applications in your environment, and then you added AI. That got that got more that got harder to manage, right? You have control planes to manage. You have multiple constituents that that now believe it should just be easier. But, that changes us from at a traditional adoption kind of metric. You know, where we say, "Okay, I deployed the technology to you, and now Alex, it's your job to adopt it. And I'm going to measure you with a adoption scorecard. And if you don't get if you don't get green on it, then I'm going to beat you with that scorecard until you become green."

2:57 And so, that doesn't apply anymore, right? Because you just got given a capability. And that capability can do a lot of things. So, then what are you going to do with it? And so, yeah, there is a everything should be great, everything should be amazing. We should build or buy or you know, integrate AI into our capabilities. And some of that is around production. You know, like we we get in the technology industry, you know, we generally get paid to make things go live, right? Nothing really, you know, other than that it's a science project. Until it goes live, it's just a science project. And so, when we go live, that means that actually users can interact. And that going live means it should be production grade AI. And that looks a lot different than a chatbot or a prototype or an MVP, or all the other names that the IT industry has given something to mean that we're just trying it. Please don't hold us accountable, but it's not really production grade AI.

4:02 So, I think, you know, there's a lot of answers in that, but it just got it just got harder. It got more complicated. >> But okay, the story that's been told, and that's why I think there's this discrepancy between the vision and what's actually happening, the story that's been told is that the technology today is getting smart enough where I don't know if you would say it figures out some of these solutions on its own, but basically that you effectively just plop it on top of a piece of software, and everything is solved. Like that is the the that is the hype. what you're talking about is basically like you're still using all the different software tools, maybe you're still using them all internally, but you now have just have an AI layer on top.

4:42 So, you know, is some of and obviously enterprise is making progress here, but in many cases it's not as fast as a lot of people anticipated. And so, is is the case here that it's basically we still don't or the technology industry still doesn't have that unified sort of centralized AI that understands everything you're using, and therefore can make sense out of it, or is it something deeper? >> you know, I think it's actually something deeper, right? I think it's it's it's, you know, we love users in the IT community, right? Because they you know, without them, you know, we wouldn't have jobs, but with them, we create all the complexity. But I believe that actually that complexity, when you when you go to just build a capability and deploy it to people, they have to have a moment, and I think I mentioned this to someone else the other day, everybody has to have their like AI moment. You know, when they they they do something, my kids had it, my wife had it, you know, my friends and family have had it, when they go Wow!

5:49 This is amazing. And they have that moment. And when they've had that moment, now they're on the arc. They're on the curve of learning, and they're going to try things and do things they never did before. But until they have that moment, they kind of like, I don't really understand it. I don't think I should trust it. I don't know why should this work. I've got job to do. I've got deliverables to do. You know, so you've got it like you've got to generate. And so, I believe AI, you know, adoption in a company in the enterprise is a thing, and it has multi-facets. It's not just give you a bit of training, or give you a platform, or a tool, or give you some agents to answer questions, or give you a knowledge document. This it's it's a whole bunch of things that go into kind of like AI enablement. And I think that is where the enterprise is. It's really in making people have that moment.

6:43 And then now getting into how does it help finance? How does it help HR? How does it help legal? our legal suite here at Daisy, we we rebuilt an entire legal suite. Right? We We've used AI to rebuild it. And we've used AI to rebuild it with AI. So, now the the lawyers are are kind of going through that moment for themselves. Right? So, that I think that's where the complexity comes in. I think that you know, we've laid it on top.

7:10 we've done MCP servers and access data. And now everyone asks this basic question. Alex, when I when I type this question to give me the pipeline for the company, is it telling me the truth? How do I know it's telling me the truth? Who made sure it was telling me the truth? Do I need to check if it's the truth? You know, so it all these things happen that that never really happened before because we were in a deterministic software world where you would find the report, click the report, and it would be the same report that you would trust every single time. And that that isn't the case anymore.

7:44 >> Okay, let me see if I could set this up the right way then because you know, talking about that chat GPT moment, a lot of CEOs have had their chat GPT moment. You know, people at the top of companies, they tend to just try everything. And so they've been that you know, they might have their own personal instance of let's say, you know, dots or muse and it like speaking of agents right now and said, okay, I'm going to try it and then what they do is they then go down and say you have to figure it out.

8:13 Maybe a step removed from the complexity of of actually the integration side and and the the nitty-gritty of software. And so we're in this moment where we've seen those capabilities, but they're still, you know, plenty of studies about and new stories about enterprises who are struggling to get their pilot to production, right? Or or enterprises who are saying we're going to spend as many tokens as we can and then maybe that's not such a good idea. There's this like surge of enthusiasm and then there's the pullback.

8:45 So maybe you can help us just understand like is that is that like area where things are the most complex is taking that big vision that maybe a senior leader has has come to because they're they've had that moment and it turns out that the actual implementation of the AI for specific functions is harder than a lot of people anticipate. I'm just trying to put a point on it here. >> Yeah. I guess it's the path you choose for your company. You have to choose a path.

9:16 Do you let everybody in your company just go at it and you connect up all the data and you just say go at it. Or do you right? I mean like like that's a path. Or do you just say I'm going to only enable AI through my SAS and software providers. So it will all be embedded in the software that we do. So that's another path that you could do. Or you could say, you know, I'm going to let the users create their personal agents so they can create their own personal teams and become more personal productivity. But I as the CEO, I'm going to create the professional agents that have more guardrails and more security in them and have access maybe to more deeper level of data and we're going to have fewer of those. We took that path. Right, we took that path. So everybody in DXC can go into multiple platforms and create their own agents to drive their own productivity and that's the path that we chose. But there are only you know, less than a hundred professional agents that are running the company in terms of core data and core actions. And so that's the path we chose. So I think to put that final point on it, like you have to choose a path.

10:21 If you don't, I think that's where it gets that misalignment between we want AI, we want to be productive, we want better margin, we want lower risk. And okay, how do I make that happen? And so nothing's slowing down and so, you know, divorcing yourself from the the wheel that goes on every day, I think is the right thing to do. But choosing a path, you have to choose a path. >> Right. And so there's there's the trust that you give you know, people in the in the path that you give people. But then there's also the form of AI that you want to that that companies are using now to see value. This is a really interesting thing that's developed over the past well, I would say past year.

11:04 Because if you if you bring us back to let's say November 2025, AI coding was good but not still not able to run these long-range tasks that we've seen develop since December 2025. And so this concept of enterprises implementing AI and you can tell me if I'm wrong here, has gone from, you know, let's roll out like a you know, an internal information chatbot, right? Where you know, people within the company can access this chatbot and get and and get access to information that they would otherwise have to go like layers up to find. Or let's let's use an API of an AI company to you know, be able to synthesize this information and present it to clients. So, that was like where we were a year ago. And in that era, that was when like you saw these studies like 95% of AI pilots fail.

11:58 Since then we've seen like a couple interesting things happen. We've seen the vibe coding era take off because now you have those long-range coding tasks available and built on top of the technology, of course, has been agents. So, for enterprises who are working with AI today, is what is the standard form that they are using the technology for and the standard problems they're they're trying to tackle? >> Let's just stick in the coding. Just to start with, right? We'll take this in pieces, but the the the coding has got a lot easier, a lot faster, but actually a lot cheaper.

12:35 Right? But just like we did test-driven automation, we have to do outcome-based coding with AI. You know what I mean? Like yeah, you can feed it requirements, but you should be testing that as you go to make sure it's achieving the outcomes that you need. there's a lot of techniques in there. You know, there's a lot of techniques to manage caching and long-running prompt queries that you're doing in your coding and the loops that you can get into.

13:04 Right? And how many loops are you going to do and how many let's say fixes are you going to put in place. That coding piece is real. There's mass productivity in there. And you know, it's it's interesting when people say, "Okay, I'm going to I'm going to burn all my tokens." I'm like I think the industry is going to reward discipline, not consumption. Right? And so, but the the question is it's not zero. And if you were spending let's say I go to build an application for a school environment. and I'd say, you know, it's going to take me, let's say, 45 days to do it. And now it takes me 4 days. Now, the difference between should it cost me 45 cents >> >> or $4.50? Everyone's like, "Oh, it's $4.50. It's crazy." I'm like, "It would have taken me 45 days." Let's put that in perspective here. So, you know, the the tokenomics of it and the consumption and what was it before and what is it after? That all has to get modeled. So, if you're running an enterprise like I'm like you have to have the models in place to be able to say, "Okay, I'm going to go do this work. What do I think it's going to cost me in terms of human labor? What's it going to cost me in terms of non-human labor?" And I have those lines in my budget, you know, token lines and human labor lines and everything else to to get to the what's the business value of what I just did.

14:25 Now, I'm not a huge supporter of use cases. I don't really go after the use cases. I think they kind of narrow you down into something that you try to fix and and the technology that we're deploying now, you know, doesn't want to be narrowed down that way. but to go back to it, so I think coding has huge steps forward constantly and I think there's you you know, really good productivity. All my teams are using it.

14:49 Right? They they're using, you know, the different vendors like I said and we're getting great results from it. But there are some techniques, like I said, that you need to do. On a user point of view, I don't think we've solved the UI problem yet. I don't think we've kind of solved it and I've started to come up to the maybe the belief that traditional UIs will disappear. And I don't know this. This is me kind of speculating, but traditional UIs will disappear and the UI will be created on the fly depending on what you're trying to do, Alex. So, if you're in there like, "Oh, I want to analyze my financials for my company." And da da da da da. Okay, well, maybe that's a different version of a UI that gets created for you versus I'd like to do my mid-year performance reviews with my team and maybe that's a completely different UI that gets created. So, I think we haven't really solved the how we deliver this AI capability to users yet and I think that's causing some structure structural issues for people. Now, also, we spend decades building software and confining people to processes. So, we use the software to confine people to a process and then we helped people automate that process by providing drop-downs and other such things and we made them super productive.

16:16 But, what's also happened was when you press submit, you don't really know what happens afterwards. You kind of like, "Yeah, I pressed submit on the PO. Then what happens? I don't I I don't know. I did I did my piece." And so, with AI, there isn't the submit button. There isn't the drop-downs and the next and the save. There is the more open, what is it you're trying to achieve? Let me help you try and achieve that. And so, for us at DXC, we've been building very specific agents for people. I think other companies went through this journey as well where they built lots of agents and deployed many to the whole company and then began the kind of effort to consolidate them down into less and less and less and make it easier for the user. So, I think on the user side, people are going through that journey. but, I am a huge supporter of allowing individuals to create their own personal agents for their own team.

17:15 I think they need to get their hands on the technology and try it and learn it and understand it and I think that is a more accelerated curve for the learning. >> Yeah, I mean, from a cultural standpoint, this is a big thing that that we think about here is just that you know, you have you you they're going to be there's going to be a divergence of companies, you know, those that allow, you know, their employees some degree of autonomy to try this stuff and encourage it and don't don't get scared by the increased capabilities that, you know, some of their rank and file employees are are finding, you know, with the technology.

17:49 And then there are going to be those who are just who are going to see people with new ideas and new capabilities come through and and they'll be like, well, slow you slow down, we're not ready for that yet. I think those those second type of companies will really fall behind. >> What do you >> it will be more of a I think it will be more of a leapfrogging. You see what I mean? Like you fall behind a little bit and then you leapfrog because many of the lessons have been learned by others and you can kind of make a jump and then you fall behind. You know what I mean?

18:19 Because I think that's really how it's happening. You know, I I have calls all the day all day with different CIOs and different colleagues that are in different companies and there are ideas being thrown around and, you know, does this really work this way and then, you know, you we kind of drift off into the safety and security topics of it all and, you know, I think CISO's jobs and we have a great CISO here, Mike. we have a great relationship and you know, I think his job's got incredibly harder.

18:46 Right? He's trying to protect the enterprise. He's trying to protect it from the outside and from the inside. And I'm deploying new technologies constantly, right? So, his his job's got really hard. And so, you know, that relationship should be close in an enterprise between CISO and CIO and really kind of working together on on what are you giving people. Now, do I let people in my corporation deploy agents to everybody else in the company? No, I do not.

19:13 >> Mhm. >> No, I do not. You can build them for yourself, but I don't let you go deploy them across the whole enterprise because like I said, that takes a production grade AI mindset and a set of things that we have architecturally that you must do. Now, we haven't touched on the whole what happens if AI fails systemically. What happens if What What's the business continuity plan if you build AI into all of your background processes and it's doing the let's call them the must-do tasks or it's doing tasks, Alex, while you sleep at night and they're ready for you in the morning and it all fails.

19:50 We haven't really touched on the business continuity of that and I think that's, you know, as an industry that that will start to How does that work? And maybe that will take us to a more of a Well, hold on. Do we have sovereign, offline, private AI? You know, the all of those kind of other things to kind of protect yourself in a continuity space. >> Right. Okay, we'll definitely get to that. And folks watching, this is you know, this is part of our big technology guide to series that we're starting here on the channel.

20:16 And we're doing it with DXC. We're going to have a interview coming up about security in particular. So, definitely stay tuned and watch the channel for that. Looking forward to putting that and we'll also be We have accompanying pieces coming in the newsletter. So, just a a PSA on that front. All right. I I Let's Let's just go step by step here because I think this is important and and worth spending time on. when you talked, Russell, about the the movement from deterministic processes in SAS to an agent world which is more probabilistic, doesn't have those standard submit buttons.

20:57 It can sometimes be the interface can be, you know, a blank screen. You know, I wonder, you know, we've been talking today about the some of the difficulties that companies are seeing, you know, as they try to implement this new technology. I wonder if you've basically identified the core of the issue here, right? Which is that it is a re-envisioning of software and it's not as seamless as moving from let's say traditional SaaS to AI-enabled SaaS or or chatbot as as the hype might make it seem to be.

21:29 Is that what Is that basically where we're zeroing in here? >> I mean, that's kind of predicting the future. I don't know, you know, like it's hard to predict kind of next week, right? You know, the week after that. But I think that you know, when when you now go to a build versus buy, if you choose to build, what is it that you are building? Are you building a traditional application but using coding capabilities and AI coding tools to do it much quicker, more efficiently, more secure, with better quality? Is that what you're doing? Or are you building an agentic solution?

22:09 That isn't based on deterministic screens and fields and drop-downs and workflow. Right? Are you building something very different? And so, we've been experimenting with both. We've actually rebuilt things as a you know, traditional application but done it really quickly and much cleaner and better and then move that same application into more of an agentic solution. And then how does that work? And how does the business logic, because that's generally what our applications hold is the business logic, where does that business logic now sit in an agentic solution? And so, yeah, that that's some of the questions that we've been we've been working through. But yeah, I don't know I don't know what's going to happen in the software industry in like 6 months' time, but I do know that it won't back down on the whole AI integrated into their solutions.

23:01 >> For sure. Okay, so so then you tell me, what is what is the core problem holding back these companies from going from pilot to production? >> I used to measure adoption and I don't. I measure fluency. So, that's my measure. When I'm asked, "What was the business value?" I say, "That's a great question for the business, not for me."

23:35 Like I help capture the information and everything else, but the business value resides with the business, not with the IT technical team, because then it becomes a technical challenge and a technical deployment and a technical solution. And then everyone says, "Well, does it work?" So, you know, that's that's part of it. And I don't think that's any different than it has been for decades. You know what I mean? Like the the business set the requirements. Now, the requirements aren't a document, a traditional you know, that you I don't know Alex, have you ever sat in those sessions where the you sit down and you're like, "Right, let me just work out exactly what you want and we'll write it all down." And we write it all down in detail and then we say, "Is this what you want?" They say, "Yes." And then you build it and they go, "That wasn't really what I was thinking." And you're like, "Oh my god, that how did that happen?" You know, so like we've all done that, right? As IT professionals and we come back and think, "Okay, let's shorten these projects. Let's do Agile so we don't have that problem."

24:32 But that's not reimagining a software technology into an agentic solution. And so I think we've got the piece where we are deploying platforms, agentic platforms, Claude, ChatGPT, you know, whichever one you choose or multiple. I think we've done retrieval and summarization capability with those. So, we've deployed chatbots to retrieve and summarize data for you. People have started on the agentic I'm building agents that have context about the question that you're having.

25:06 And then people have gone, I've built thousands of agents, hundreds of agents, and now they've concentrated into No, I've only got three or four for my company, and everybody can talk to those three or four, and orchestration happens in the background. And so why I'm explaining this is I think the challenge is you have to go on that journey. If you try to like skip to I'm just going to build agentic solutions and everything will work, I think that's where it fails. I think you have to go on this journey, so you get the understanding, the base understanding, and you take everybody along with you.

25:39 So that's the the kind of learning I've had of like No, I don't think you can skip building lots of agents and just go to one. I think you still have to build lots of agents, and maybe then you hide them behind one. You know that So I think you have to go through the learning. So that's my really summary of like why is it all failing? Why are people not really getting the value? Now, how you measure that value? I'm seeing a lot of noise on like how do you actually measure the value of an AI deployment?

26:06 Well, that gets down to is it the AI that's working in the background during the night? Is it the chatbot that's helping you answer a benefits question? Or is it the workflow engine that's reviewing your invoices and telling you which are exceptions and asking you to approve and do the exception? That depends on the type of AI. So I think like that's where we've got a little bit confused. We've got these kind of business value realization templates, and everybody's got a template, and they're like, Okay, fill it in, and at the bottom it'll tell you if you had value or not. And I don't think one size fits all, just like all the software that we deployed out there doesn't. So I think that's really something you've got to go through the journey. I don't think there's any way to skip it. And if you And if you try to skip it, I think you'll miss things along the way and and and gaps in your learning.

26:56 We did a we did a large agreement with a learning management system provider to really get down to curated small snippet learning about for people. So, I'm in a certain role, other people in a certain role, can they they need to get education, it needs to be fast and short, but it needs to get to the point of what they're trying to do when they're trying to do it. So, there's Like I said, there's an AI enablement program that I think everybody should run, and it should have multi-facets to it.

27:25 >> Yeah. So, you know, from a technical standpoint then, what does production-grade AI look like? >> Hm. >> I think a lot of people are wondering this one. >> Yeah, it's I built some stuff and I deployed my first large-scale deployment of AI, I did it as if it was a SaaS app. And it was a complete disaster. Complete disaster. but I God, did I learn a lot from it. I mean, my unbelievable amounts of it. And I didn't make that mistake again.

27:53 so, production-grade AI that you you think about the control plane. So, the I have most of, you know, we have great partnerships and we have most of the major SaaS providers here at DXC. and we use them internally. Now, when they deploy AI inside of their application, it comes with a control plane. And that control plane allows you to manage that, right? And so, you can apply the governance, you can apply the controls, you can ensure the security, you can ensure the data sovereignty and security of it, you can get into DLP and all the other things that you want to do and you have these control planes.

28:30 If you build it yourself, you then have to work through, well, where is my control plane? How do I manage that? What am I going to do with it? Now, from a user point of view, they want to know certain things. Is it safe? Can I type this? I get this question all the time. Is it okay if I type this? Yes, it's okay if you type this. Are you sure no one's going to see if I type this? Yes, it's okay if you type this.

28:50 So, the input needs to be safe and secure and people need to feel safe and secure about it. But then you need to like manage that on the back end. Data retention. Where is the reasoning? Is it in the right country? Is it in the wrong country? Okay, hold on, you got to go work through that with a client. So, like I said, there's there's there's a lot to production grade AI. And if you think about you built an app on your own, Alex, and you just said, "Yeah, it's great." And then you went to a company that builds apps.

29:23 And you went, "Oh, yeah, I kind of missed all those. I missed all those kind of things. Integration to monitoring. Truth testing. we have we have 100 questions. I have 100 questions with 100 defined answers. And I run that against agents all over the place to see if they give me the exact answer I'm looking for. And if they don't, we take us Take a minute. We take a breath. Why did it not give the real answer? This is the real answer. We know this is the factual answer. We We're running these tests all the time.

29:53 That's part of running production grade AI. The harness is another point. We have an Oasis product here. Great harness. and they're building it to make sure that that harness really controls what the AI is able to do and not able to do. So, there's there's a there's a lot to it. I think the answer is there's not just one answer. There is a lot to it. If it was super easy, everybody would everybody would just do it. We would never buy anything again.

30:20 >> >> Imagine that. But here's the thing that I I'm really wondering, which is, you know, we've talked You've talked a lot about trust today. and and I just want to know, how do you get to trust with AI? You know, it's something that I'm personally struggling with. in my own teeny tiny use cases trying to deploy this stuff. Like when I set up an agentic process, right now I probably Look, I'm not a great example of this, but you know, I think a lot of people are feeling this. When I set up an agentic process, my pattern for trusting it is I test it out on like really low-stakes type of activities, and then I cross my fingers.

30:59 And if it passes the test, then I put it on higher a little bit higher stakes. And if it passes the test, I cross my fingers, and then I put it on on like, you know, high-stakes tasks. And I'm still not fully there yet. So just tell me a little bit about how you get to trust. >> my level of trust and a regular user level of trust are different. Right? So, like I said, we have a limited amount of professional agents that we deploy in the company that have access to lots of data, can do lots of things.

31:31 And the rigor and the testing that they go through, and it's not just a test it and deploy and we're done, it's a constantly testing things to make sure that it's still working. Is very different than people building, you know, a personal agent to summarize their email last night or review how many open requisitions came in or, summarize the POs that are waiting for them or, you know, whatever those tasks are. So, the rigor that goes in is a lot now.

31:59 Trust is a funny thing. If I ask you, Alex, like, when you ask a question to somebody in your company or your friends or whatever and they give you an answer, how do you do you trust them? And what made you trust them? Was it because they gave you good answers previously or because you think they know more than you? So you're like, well, actually you know more than me, so I'm just going to trust you. Or do you trust but verify every single time?

32:25 When your friend says, here's the answer and you're like, no, I need to No, stop. You need to tell me where did it come from, right? And so, you know, and I do that sometimes with my friends as well. And I'm just like, no, hold on a minute. Where did you get that from?" And everybody's like, "Yeah, this is where I got it from." And then you're like, "Well, I don't think I trust that source either." And so, from a user using the capability point of view, you build trust. That's the first thing.

32:50 You It doesn't just happen. You You build it over time. They also need to trust the people in the back room, right? That are configuring all of the guardrails and the control planes and everything. That's generally the IT organization. And when I say the back room, I don't mean the basement. I mean like behind the curtain. and you have to trust that organization. And to do that, there needs to be transparency. And so, the IT team needs to constantly have the transparency of this is how it works. This is how I've tested it. This is how I'm controlling it. This is how I'm protecting you. And that is just a That is a constant reiteration of trust.

33:32 But like I said, the production grade stuff that we deploy is is is a lot more rigorous. the production grade stuff that we have on the client side, because we run some of the world's critical infrastructure, is even more rigorous. And then the the user people creating agents and thing that is that is slightly less rigorous. >> just cuz you mentioned it and I can't get it out of my head now. can you talk about like the total failure of AI preparedness that you brought up before? How do you account for that?

34:02 >> The total failure? >> Well, you were talking about how like if AI completely shuts down while you're while you're sleeping. >> Yeah, so >> I I don't know if I if I got the phrase right, but it does interest me. >> Well, so we had a moment, I forget how many months ago it was, where you know, the US put out a government all order that said no one can use this model. >> Mhm. >> Remember?

34:23 >> Oh, yes. >> That And I was like, "Okay, well, okay, great. Everyone switched to another model." What happens if like you didn't have the capability to switch to another model or you didn't know how to technically or you'd built it into things you didn't know that you'd built it into. And so the model thing is fluid. Let's just be very clear. So it is is fluid. People have model gardens that, you know, context aware questions are routed to different models and there are AI gateways out there to help you do that kind of routing and technology.

34:57 But you have to have that continuity in place. So you have to just go through a very standard continuity process of, okay, if I'm if I'm using AI to execute a business process and it fails, >> >> how will I know? What will I do? What will my backup plan be? And so do I build a second AI system? Do I build a third AI system? Do I just have different offline models and online models? Do I need my own compute Do I need And you just have to go through it. It's a It's a very standard process because of the continuity process of it.

35:31 But we had that moment where that model was completely taken offline for everybody. And then it came back, right? It was like 45 days or whatever or 30 days. But you know, we we had that moment. That's got me into the okay, what's really running with AI and do I have the ability to switch between control planes, orchestration, models, cost points, and et cetera. And that technology is evolving. So there are companies out there doing, you know, gateway technologies and everything and I think that's and you kind of segment in the industry.

36:02 >> Enterprise AI stack. it's always evolving. But what does it look like to you? Like what are the core components if I'm an enterprise and I'm trying to build an AI stack, what do I need? >> if you're build if you're building building or you, you know, like you can buy >> Yes. >> You see what I mean? Like you can totally buy a stack. >> Yeah, let let's do build versus buy whether and then we can talk about what the stack looks like cuz I want to hear your thoughts on build versus buy. Does it make sense for companies to build this from the ground up, maybe with a partner, or should they just be buying solutions off the shelf?

36:34 >> I guess that comes down to the relationship you have with the partner. >> Mhm. >> Right. So I have a relationship with Amazon. and we're more in a co-innovation relationship. You know, we've gone from kind of like buy a supplier to provider to partner in terms of like what we're co-innovating on certain products. and because of that that makes that easier. You know what I mean? Like it's like, "Hey, this isn't working. It needs to work this way. Can you make these changes?" That's like a fluid relationship. there are other SaaS providers that are building AI, really good quality AI in. and so I guess that's part of the speed to market. If you have the technical capability then you should exercise some of that capability. If you don't, there are some great capabilities from the providers out there and your time to market and the and the results that you need will come quicker. It won't change the curve of everybody suddenly having to go on and learn, have their moment in AI, how does this work? It won't change the questions on trust. You know, all those things will still be the same.

37:42 But if you have the technical capability, you should be trying that capability, exercising it, exercising that muscle, and building something. And it can be very complimentary. Like I said, I've got great partnerships with great AI technologies and capabilities, but I also build some of our own. So I don't think there's a I don't think there's a one-size-now. The stack is super interesting. If you put the two traditional stacks next to each other, you know this is a database, this is our middleware, this is our application layer, this is a you know, like that. And then you're like, "Okay, now I want to build an agentic solution."

38:17 Well, what does that look like? Okay, well, I still need a data store. Does it need to be the same data store as I had before? Or do I store it somewhere else? Is it a different kind of database or is it the same kind of database? Maybe I use you the same. So, you everybody should go build their kind of this is my traditional stack and now this is my agentic solution stack and have them side by side. And I think actually when I've done that with our business partners, it's really helped everybody understand, oh, when we if we're going to build it, we're going to go from this one to this technology stack. and then you can put whatever you want in there.

38:53 Like I said, you might choose to buy an AI gateway. You might get one from an existing provider. You might do orchestration yourself or you might get the orchestration layer in the control planes. You might have multiple control planes. Right? So, like I said, I've got most of the largest top providers and they've all got a control plane. Now, it makes it a little bit complicated for the CIO to manage all of those, but then you get to pick them in your technology stack.

39:15 Okay, I'm going to I'm going to integrate this control plane for this particular area of the business. So, yeah, I think the the the more persona-based you can get. So, we have coders, we have finance professionals, we have HR professionals and then what is the technology stack that they're using is the is the way that we've kind of tackled it. >> is there like a starter pack, you know, for companies that are getting started? >> Absolutely. It's I I I you know, I I do a great rate on the starter pack, Alex, if you want to buy one. There's a there's a great rate on the starter pack, but like it's not you know, there's no secret sauce in there. I think the secret sauce is like we've just gone through a conversations of how do you go through trust? How do you build production grade AI? What's the continuity? How do you manage multiple control planes? How do you do manage multiple LLMs? How do you You see what I mean? Like and when you put all that on with your existing technology stack, that's where the complexity comes in.

40:15 Like, okay, how do I how do we get all this to work? So, like, no one thing is particularly complicated. When you layer them all in, and then, like I said, you're going to do another podcast on security, that'll be a fascinating podcast to to listen to for everybody. And so, because I think that's you know, that's blowing the doors off things, right? There's just like, you know, a world out there that wasn't out there before, and the CISO has got a, you know, really challenging topic. So, then you have to go take all the security elements into place. So, if you go build something, okay, how do I know it's secure? How do I know I'm protecting my data? How do I know I'm protecting my people? Have I stopped any injection technologies? Have I stopped any phishing technologies? I mean, we haven't seen a set of, you know, phishing agents show up yet.

40:59 >> Yeah, Russell, as you're talking, I'm just thinking like, man, this technology's moving at a million miles an hour. It requires, you know, a rethinking of the way you implement software, of the way your organization functions. There's all these security implications to it as well. And you have the So, in some companies will have the C-suite who are like, "Where Where are my results?" And you put it all together, and it's challenging. >> Yeah, it is. And but that's part that I think that makes it fun. So, for all the technologists that are listening, you know, yeah, it's kind of fun. You try new things. We're doing a lot with voice. Voice is super interesting. I I actually believe the new interface will be voice.

41:36 >> Mhm. >> And that's why I believe the visual interface will be on the fly. Because it will build based on what I'm talking about, with what I'm questioning, and how I'm doing it. but I, you know, we we aren't there yet. We're not like, you know, in a dynamic UI building. And I think if you showed that to a regular person that has to execute a process in an enterprise today, I think it would be just too much. I think they'd be like, "Whoa, whoa, whoa. What are you Like, like, it's changing on the screen as I'm talking?" Yeah, I'm like, yeah, that I I believe that's where we'll end up going, but we're not We're not there yet. But there is a lot to the stack.

42:10 There is a lot to do And like I said at the very beginning, I think we we we did this chatbot thing, we did this retrieval and summarization thing, and we just layered on more technology, and and none of it really just suddenly disappeared. >> Given the pace and all the solutions, and the eagerness of companies to implement them right now, is there going to be AI technical debt, a form of this in like the next year or two?

42:36 >> you know, people have talked about what's the agent life cycle. >> Cool. >> You know, like you know, and I and I was asked the other day, do you treat agents like team members, or are they like software? I'm like I immediately had an answer, and before I it before I finished the answer, I was like, hold on a minute, let me just think that through a little bit more. Cuz I had to like really and generally I'm like, okay, well, what do I treat them as a user, but what do I treat them like as a CIO?

43:06 And then the answer was well, the ones you build for your personal agents, for your team, you should treat them just like your team. The ones I build, they will have their own identity, because I need them to have their own identity, because they're accessing certain things that we don't want to take from other people's identity. Right? So, there was like, okay, let me think through the answer on that kind of stuff, but >> >> I think that you know, the enterprise will get into really quick shape. It's moving really quickly. Like, don't get me wrong. So, every morning I wake up, something crazy's happened. I think the security place, like you said, is just is a very is a very active place right now, and and people need to go manage that. We will we will we will do our best. The CIO job is a fun job. I actually enjoy it. I really enjoy it. I've got a you know, I've got a great leadership team around me.

44:00 I've got great peers in the business that understand this, great CEO that really gets into the technology, and I'll just you know, I'll say this about him, You know, he went to the Claude Anthropic boot camp. Right? So, that's when when you go there, you come out with a different level of understanding of what AI is and how it works and and how everything should happen. So, if you've if you don't really understand it, you should take the time to understand the technology. and then, like I said, you should you should ask what's the path.

44:29 In your company, what what what path are we going on to really solve these problems? There's lots of different paths you can choose. but you have to choose one. >> I just want to close with this. You know, the the idea, we've talked about this through the whole conversation, the the idea of AI, you know, of course, there there's much more powerful capabilities, but the way people talk about it is it's a simplifier, not something that makes more things more complex. I'm going to I'm just going to ask you to do another prediction and and do something forward-looking, whether you like it or not, I guess.

45:03 >> Okay. >> But, you know, is is this age of complexity or this moment of complexity going to be the rule for enterprise moving forward, or or is this is it going to be a passing moment? >> I believe it will be a passing moment. I believe it will be a passing moment. I believe that, you know, the introduction of new technologies won't ever slow, right? But our understanding of how to get the value and how to control it and how to make it safe and how to trust it will catch up.

45:42 Right? We're in this moment where we as people are kind of behind. You know what I mean? Like AI jumped out of the thing. Everyone's like, "Oh my god!" And then and we and we kind of like the technology moved faster than our ability to consume it. And now, I believe that, you know, we will catch up and then it will start to feel like AI again. We're like, "Oh, okay. Yeah, this is how you do it. This is how you do production grade AI. This is how the continuity works. This is why we measure fluency and not adoption.

46:07 This is how we deploy AI. This is how agents talk to agents and the protocols. Like we actually have two standards of business conduct in this company. I've got one for employees and one for how we want agents to work. Right? And both of them were created by HR, not me. Right? Because I said yeah you like that's about to happen. But how do we want agents to talk to people? Okay. All of that will become normal. And so I think we will catch up and then it will be Okay. All right. We kind of know what we're doing. And then you'll get into the Okay, how do we get real value from this? I I I don't think we've really squeezed or been able to measure the amount of juice for the squeeze at this point.

46:51 >> Russell, it's great speaking with you. Thank you for all the insight and wisdom and for the fun conversation. Really do appreciate that. >> >> So. >> Thank you very much. >> All right. Thank you again. Oh, I got to ask you. If people want to learn more about DXC, where do they go? >> you can go to the the website or you can find me or find anybody else in the leadership team and like I said, there's a lot of conversation. We have a lot of great companies out there. So we we got a lot of great services too.

47:17 And so, you know, these in this era a great technology provider is is really worth it. It's really worth the help. >> Love it. Well, thank you, Russell. And thank you, everybody, for watching. We'll see you next time here on Big Technology.