AI Adoption Was the Easy Part. Now Comes the Hard Decision.

0:00In May, the Pentagon announced agreements with eight technology companies to deploy advanced AI on classified networks. It was a major shift in access, but access is not readiness. A model can be available and still fail at the moment it matters. If the data is trapped, permissions are unclear, infrastructure cannot reach the edge, or no one has defined the decision it should improve. So on today's show, the missing step between acquiring AI and producing a mission outcome, and why federal leaders must treat the model, data, security, and infrastructure as one system. Joining me are three leaders who bring different but complementary perspectives on federal AI.

0:37Ahmed Ali, Google's Chief Technology Officer for the intelligence community, who understands how advanced models can operate across sensitive and classified environments. Suri Dorvasula, Senior Vice President and General Manager of Dell Federal, who brings the infrastructure and mission delivery perspective. And Walton Smith, WWT's Managing Director of Public Sector Cloud, who helps agencies connect those technologies to practical use cases, secure architectures, and measurable outcomes. This is the AI Proving Ground podcast from Worldwide Technology. Let's jump in.

1:18Well, gentlemen, welcome to the AI Proving Ground podcast. Walt and Suri Ahmed, how are all of you today? Wonderful. Quite good. Thank you for having us. Doing great. Looking forward to the conversation. Absolutely. Well, like everyone else, federal leaders are being pushed to adopt and move fast on AI. That's no secret for anybody here, but AI is a very broad topic. There's a lot to take in there. Suri, from a federal perspective, just to start, You know, is the biggest challenge right now from your perspective access to AI or is it more about deciding what problem AI should be solving for all of us?

1:52It could be a little bit of both, right? Government has plenty of access and ability to get to AI. But effectively, you know, what are they going to do with it, right? You've got to look at adoption. You've got to look at, you know, the outcome and the need. And, you know, why are we actually using AI? So, you know, I think the question becomes, you know, what are we solving with AI versus just the access problem? And so, you know, it's out there and it's prevalent. But, you know, what the ultimate use case is and what government's doing with AI is a big question. When you talk about the user community, ultimately, AI has a big opportunity to help with mission outcomes, government needs, and ultimately citizen services and citizen enablement.

2:41So there's a lot that can happen, but right now, the access is there, and it's a matter of what it's being used for at the moment. Yeah. Ahmed, I've heard you describe AI as the everything machine, which I thought was a unique way to describe it. expand on that a little bit. And why is that actually adding complexity? You know, that sounds like an ideal thing, but it's actually making it harder to adopt this stuff. Yeah, it's, it's kind of interesting because when, when something purports to be good at everything and it's, you know, a, you know, a general super intelligence, you, you kind of get paralysis with where do I start?

3:19Like, what's my, where do I apply it at first and to kind of like build on something that Siri said, which is, you know, there's, there's almost like this belief in, you know, step one, adopt AI. Step two is this big giant question mark. And then step three is mission awesome. And, and people usually have a hard time defining what step two is. And so it's kind of like incumbent upon all of us that, you know, are involved in the AI space to help government folks figure out like, what is that step two after adopting AI that leads them to the mission awesome. And so like, you know, when I think of like the point around the everything machine is that like, it creates potential paralysis.

4:12But the other problem is there's this natural skepticism that anyone would come into because Because back in the days when they used to sell snake oil, the guy would stand on top of the soapbox and it's going to cure your lumbago and it's going to cure this and it's going to cure that and it's going to cure that. And sometimes AI can kind of sound like that because it's an everything machine. And so it's really, really incumbent upon us to keep that promise as the transformational nature of a technology while putting some grounding around here are very specific things and specific workflows in your organization that you should really, really focus on applying AI to first.

5:01Yeah, well, I'll bring you in here. Maybe fill in that question mark box for us, the step two. What do federal agencies or state and local agencies, what do they need to do after the adoption phase to get towards that third step of Mission Awesome? Sure. It starts before the first stage. Our public sector customers, the data that they're dealing with is, in most cases, very sensitive. It can't be put, in many cases, on the public web. So we need to take a step back and say, what are those outcomes that Ahmed and Suri are talking about? What are the best way to bring the models to the data to make sure that we can get the results we're looking for?

5:41So what we've been really focusing on to make sure that the total cost of ownership gets in the ROI they're looking for is to break down what is the problem we're trying to address? How does AI handle that problem more efficiently, effectively, and quicker to be able to solve the problems that we need to keep our country safe or to provide services for our citizens? So it's all about that pre-planning to understand what's the best model, what's the infrastructure needed? Can we go to the cloud? Does it need to be a hybrid model? Or for our more sensitive workloads, does it need to be in a disconnected state?

6:17And that's why we're really excited about the partnership with Dell and Google because Dell is building us the gear we need to be able to solve those multiple use cases. And then Google bringing Gemini 2.5, Cineb 2.x to bear and all those different use cases gives cutting edge solutions to our public sector customers where they need it, when they need it, in a cost-effective manner. And Suri, when we were having a couple of prep calls, you mentioned agencies need to clean house. Is that clean house as it relates to data, governance, policies, it can span a wide gamut here. Is that what you think Walton's talking about in terms of that pre-phase one, they need to clean house and get everything in order?

7:01Yeah. I mean, it's not really clean house as much. And of course, it doesn't have to be a perfect data estate per se before government starts, but they need enough visibility across the enterprise to then say, okay, what am I going to do with AI? What is the problem I'm trying to solve? What is the outcome that I'm running after? And do I have all the data assets in a governed, secure, controlled environment such that I can give AI the visibility to all of my data assets so that ultimately I'm driving for the appropriate outcome, right? When, you know, it's like anything, if your house isn't in order, then you don't know what to organize first.

7:46You don't know what you're trying to clean up. Ultimately, making sure that we have, you know, a clean view of the data assets, that we've removed all the junk, if you will, in such a way that you now have an environment that is going to, you know, ultimately let AI ingest the assets that are meaningful to driving towards the outcome. You know, it goes back to the very beginning of this conversation, which is, you know, we need to know what the outcome we're solving for and then making sure that we then bring the data assets that are going to help produce that outcome in a meaningful way. Today, government is siloed across a variety of, you know, data silos, enterprise systems, whether they live on premise or in the cloud, they're just sporadically put everywhere.

8:32across multiple cloud providers at that. And so when we look at it, we think, okay, well, how do I gain access to all these data assets that are living everywhere and then put the right governance around it, make sure that I've done the clearinghouse and making sure that I've got the right assets enabled such that AI can then produce the most accurate, reliable result that will produce the outcome. Yeah, Ahmed, build on that a little bit. Commercial teams can often just spin up cloud environments, test hypotheses, et cetera. But federal agencies are dealing with some of the most sensitive data on earth.

9:06So it's not always as easy as that. What's your perspective on what Surrey was just talking about in terms of their sensitive data, but it needs to traverse kind of the broad range of where it's headed. Yeah. I mean, so the way I think about it is our traditional path to handle processing for sensitive data has been to bring capability per data silo. And the thing that we have an opportunity to solve for, because that strategy in and of itself is going to be very hard to replicate with AI because of the, you know, exquisite requirements all the way on the infrastructure stack from the electron to the user interface.

10:04Like they all have to be engineered really, really well together in order for AI to work best. And so trying to replicate that within every single data silo is going to be a really, really hard, not to mention an extremely expensive thing to do. And so the way to really think about it is to think of an AI almost like a person that you can authorize to participate in multiple data silos and be, you know, almost like read in to all the different data silos. And you have the trust and the controls built into it. that it is smart enough to not reveal something to a user who asks a question from a data silo that it has access to, but that user doesn't.

11:03And so to me, it's actually almost like the ultimate solution to the traditional problem of to leverage multiple pieces of data. your options in the past were I either had to shove it all into one data repository, and then I could query against all of it. But then you've got like sharing issues and whatnot. Or my other option was like, we just didn't share the data. And so there's an opportunity with AI to have essentially an intelligent broker in the middle that you can trust to have access to all the different silos, while at the same time, giving services to end users that are appropriate for what they are authorized to see.

11:55Yeah, Walton, where does that intelligent broker sit? And are federal agencies generally thinking that way? Or is that a mindset shift for them? It's a mind shift change for these agencies. They wanted to protect their data. Their data is their crown jewels. But looking at Gemini Enterprise, the success it's had with the Department of War, the adoption rates that they have to be able to build those agents that Ahmed's talking about to be able to help tackle some of those repetitive jobs that are having to be done to free up people to be able to do stuff that's actually important as opposed to some of the busy work is crucial.

12:36So we've seen the ability to take Mandiant, put that on top of Elastic and Splunk feeds and have Gemini investigate any attacks on a system. Kind of like a sock in a box to be able to allow Gemini to tackle some of the mundane things that are coming in to free up the soldiers to be able to go after the real attacks that are needed to protect. AI is great. AI is going to make our adversaries that much more effective. So we need to make sure that the folks that are protecting us have access to the latest and greatest tools and have those agents to be able to do the work to get us the scale we need to continue to move forward.

13:16One of the bigger issues that we're going to see is that the power needed for this, we need to continue to build the data centers that can support the 705 standards and some of the other clearance needed to be able to put these model locations. So that's going to be a problem that we need to address moving forward as well. You know, there's something interesting that you just said, Walton, because it kind of really spoke to me because while I will not divulge my age on this podcast, I have been in the information security industry for a long time. It was a critical part of the early part of my career.

13:58So long ago that when I was doing it, I was referred to as an information assurance officer. I'm sure some people listening to this recording will know what that is. But I have seen like multiple waves of technology adoption that are all meant to fundamentally secure computer infrastructure. You know, I was, you know, way back in the days before, you know, firewalls were a thing. And I was involved in the first, you know, deployments of those in the government. And then, you know, okay, firewall is not enough. We need an intrusion detection system. And then we need logs and then we need this and we need that.

14:39And then over time you go and you ask someone who is responsible at the end of the day for the risk profile of the cyber assets of a government agency. And you say like, do you feel safer? And I've never had someone tell me yes. And usually the reason is because regardless of the advancement of the awesome technology that has been built by multiple really, really innovative companies in the cyberspace, there's still this fundamental concern, which is like, I can't get a quick answer to a simple question. like, am I being attacked right now? Or am I like, what is my vulnerability profile to this horrible thing that was just announced by the cyber infrastructure and security agency, or whatever the case may be.

15:44And AI, from my perspective, is the first time I've seen in my entire career epoch where we have the opportunity for a senior person who is responsible for the cybersecurity risk for a government organization to be able to ask a question that is very, very simple, like, what is my cyber posture right now? And to actually get a very quick answer to that question. And that to me is one of the most game-changing things about the cybersecurity aspect of it is we finally potentially have something that can keep up with the volume of telemetry that's coming in and make sense of it in a way that a senior leader would need.

16:36Ahmed, the only thing I would change is they're being attacked all the time. The question is, are my defenses working and where should I put more troops to protect? Because they're always being attacked. It's are my defenses working? Suri, over to you. And no, but that's the most exciting part, right? Especially with the agent capabilities with AI, you know, to your point, Ahmed, about, you know, being the person asking the question, you can also get the answer along with remediation immediately, you know, kind of presented back. And then the question presented to the person asking is, do you want me to go make the changes and fix the problem such that it's not occurring anymore?

17:17Or would you like me to evaluate other scenarios? Like the next question is going to be the most important, right? And continuing to dig and iterate on that is the most exciting opportunity for all of this advancement in technology. We now have more virtual hands-on keyboards. We have a 24 by 7 operations. We don't have the same fatigue that we've always had. As an assurance officer, you're looking at data and there's data dilute, there's security, overwhelmed aspects of security. Now you've got this machine sitting in the background that's presenting you response, action, taking the action. And it's just an exciting opportunity to say, I've got more people here, more automation to help defend against all the threat actors that are out there.

18:07Well, sorry, you know, speaking of security, what other security controls do SOC teams, do security leaders need to be in the best position to your point, answer that next most important question? You know, the one that, you know, stands out, especially as we've been talking is around the, you know, identity and access management when it comes to agents and the agentified environment that sits behind all of this, right? Ahmed brought up a great comment about putting AI across the different data silos and then presenting data that is only applicable to the user asking it and not more than what is being asked.

18:48Well, I mean, that's governance, right? That is the security parameter. You're going to give unfettered access to AI, but ultimately, who's making the decision? Are they the right person to make that decision? What is the human in the loop that needs to authorize the decision. All those things need to be rethought of, especially as there's access that's going to be granted to machines and agents and specific taskers versus Ahmed, Suri, Walton. It's going to be a different role that's sitting out there that is a machine, is a bot that's doing work on your behalf. And you want to make sure that you've given them the right security profile, the right understanding of who's asking the question, such that you're not exposing gaps in your infrastructure and presenting data that is not supposed to be presented in that manner.

19:45Walton mentioned the intelligence side of it and the security there with 705 and all that, right? We can have a great data center that is secure, but then the tasker and the questions that are coming out, are we making sure that we're creating the governance inside of that to make sure that the people asking the questions should receive the answer that they should receive based on their role? Or are we giving up all of our information, right? Like thinking through all of that is is creates trust that creates governance over, you know, how the information is disseminated and then ultimately the action that's taken based on the the question, the answer and then the, you know, go do, if you will.

20:27So, you know, it's just a difficult thing that's going to take its own body of work because that's where we need to be in control of what the AI is able to do with the information that we're giving it access to. You know, Siri, that's so interesting you say that because it turns out that it's actually possible to leverage AI agents in an adversarial nature, similar to the way we do it in the human space. Right. So like, you know, how we some of one of the ways we handle governance in the human space is like, you know, we have people that are tasked with achieving the mission outcome and that requires sharing data.

21:06And then we have other folks that are called security officers and their job is to like, you know, really understand, you know, classification guides and all this other kind of stuff. And to basically say that it's supposed to be the expert that you can go to and say, like, is this data transaction authorized? Or can I expose this based on, you know, how this data is secured and categorized? is actually really possible to build very specific expertise in agents in nothing but honoring and enforcing a classification guide, as an example. 100%. So you have like one agent that wants to do something.

21:45And before they expose data to a user, they ask another agent like, hey, just as they would ask their security officer, like a human would ask a security officer. It's like an AI agent asking an AI agent security officer, is this okay? So you can actually build multiple layers of governance exactly like the defense in depth strategy we do in the non-AI kind of space. And that structure helps giving your agents a job description, right, to then say, here's your role. Now follow the guidelines of the role. And here's all the books that you need to read in order to be effective at your role. And now do the role as a side by side to every other agent.

22:22I mean, that thought process can be really heady, but ultimately it's so possible to go do. It's just thinking through it all that is really where that governance model comes into play. But, Suri, I think it's so critical to do that. If you go look at just about every data spill, every phishing attack, anything where adversaries were being able to get in, the systems were working. some human made a mistake because whether they were tired or cut corners or whatnot, the human is the person at fault in most cases. So leveraging the agents to protect our information and to protect our employees, I think will only improve the overall security stance that our federal agencies and state and local agencies can enjoy.

23:09Yeah, better than any human could have, right? At some point, you keep training it and educating it, And it's going to follow a very specific guideline with the limited gray area as you keep tightening and applying all the rules further and further, for sure. They don't need coffee breaks. They don't take PTO. They're working 24-7. Yeah. This episode is supported by Nutanix. Nutanix provides organizations a secure, unified platform for running applications and AI and managing data anywhere. The Nutanix software platform helps simplify operations for traditional and modern applications, freeing you to focus on organizational goals.

23:49Trusted by over 29 ,000 customers worldwide, Nutanix empowers you to transform digitally and run hybrid cloud environments consistently, easily, and cost-effectively. Walton Suri says this is heady stuff, but it's also possible. What's the game plan for actually implementing those policies, those guardrails? Is there kind of a common one, two, three type steps or is this a snowflake in each case? Well, every case is a little bit different, but working with Google, working with Dell, we come in, as I mentioned earlier, the planning up front to understand what's trying to be done, the structure that's put in place.

24:26As Ahmed mentioned, Gemini is a massively useful model, but making sure we're configuring and setting up to meet the needs of the customers is critical for the initial success. success and then continuing to train and educate the users to be able to understand, okay, we've tackled mission problem number one. Now let's use the depth and breadth of the model to continue to look at other use cases that can be improved upon leveraging the Gemini model. Yeah, I'm going to switch pace here a little bit. One of the interesting things that you said in one of the prep calls with us was that you compared AI infrastructure to a human body as opposed to just a collection of parts.

25:10In biology, performance does not come from one strong organ. It comes from the coordination of everything within your body. Is that the right way to be thinking about AI infrastructure, not just the model only, but something that serves the broader body and its function? Yeah. I mean, one of the things we've noticed is if I think of all of the different components necessary to deliver an AI outcome, power, data center, world-class infrastructure, the right accelerator chips, a good operating system, a proper frontier model, an interface to serve it, an agentic workflow, a way to do all of those things.

25:56It turns out that if you take the world's best engineers in each of those different components and they never talk to each other you will get a crappy ai outcome and so like that's the big thing that we've learned is like we can't treat every single part of an end-to-end vertically vertical ai stack as like you know either an upstream or a downstream supplier it doesn't work like that like they don't have to come from the same company i'm not saying i'm not suggesting that by any stretch of the nation i'm just saying that like there has to be tight coordination between like you know efficient use of power because if you don't use it efficiently zero chance you're going to be able to handle the workload required because there's just there's just not enough power to go around you know data center efficiency around cooling and liquid cooling and all these other kinds of things things like you know core chassis of infrastructure and it really really really matters the memory interfaces and the disk interfaces and all of these things like like all the way down to the like the real nitty-gritty details like milliseconds and nanoseconds matter of latency.

27:20It all matters. And then I can have this conversation with customers so many different times. They're like, hey, I want some Gemini. And I'm like, okay, great. Let's say I got you some infrastructure. I got you Gemini and exposed my Gemini API. And it's compatible with the open API that you expect. What are you going to do with it? And so if I haven't thought about how I'm going to integrate that upstream, I'm going to create agentic workflows. I'm going to do all those things, it's almost like you have like a, you know, a strong heart and, you know, you have, you know, a good, you know, a musculoskeletal system and you're eating right and all this other kind of stuff.

28:00But like, you know, your, your brain isn't working. Like, I mean, like you think about it in the human senses, the reason I compared it against that, it's very obvious that if one part of one system in the human body is not functioning, there's going to be a serious problem with the person's health. Like that's so obvious in the human aspect, but it seems like it was less obvious in the computing world until we started to look at AI that it's now really, really obvious that if you have power problems, if you have data center problems, if you have infrastructure problems, if you have model problems, if you have any problems in any part of the stack, like the system itself is going to be unhealthy.

28:47And you have to think about it from that entire systemic point of view. And to double tap on the meds, continue to talk about that. It's doubly important to look at who are the users of the information. If I'm sitting in a nice office with a fat fiber pipe to my desk, I can probably get full access to all the models I need and all the compute real-time access. But if I'm a warfighter out on the edge, I need that information from that UAV. I need better understanding of situation of where and what's going on. So as we look at tackling problems, it can't just be if you have a perfect environment.

29:26You have to look at the totality using a meds analogy, the totality of the human body to understand that at the extremities, they have to be able to get access to the answers as quickly as possible to be successful. Well, Suri, we're not, you know, we're not talking, you don't need a perfect environment, right? But you do need, you do need good enough. How should leaders be thinking about what that good enough state looks like so that they can start to, you know, put AI into the hands of, you know, to Walton's point, the users that are going to make the most impact? And I'm thinking about what Walton just said, or, you know, the warfighter at the edge, the extremity.

30:03And think of the warfighter needing certain amount of data to be surfaced at any given point from a UAV or sensor or what have you, right, in theater. And they're at the far edge, right? At some point, we have to be able to think about, you know, do the agents, does the AI really understand what the warfighter is going to need in that time, right? And it comes down to really understanding what is the mission, what is needed to support the mission. It comes down to understanding, you know, what is the data that needs to be presented and making sure that we've grabbed that data, we've put it in the right spot and we've allowed AI to understand it.

30:40And then, you know, and we're surfacing the right information that's going to be most impactful for the mission at hand. Right. And I, I love the movie. So I think of every, every sort of, you know, warfighter movie or, you know, sci-fi movie. And I think, you know, I've got, I've got, you know, I've got goggles on, I've got information being presented to me at real time, But what is that information being presented? It's going to be an agent that's going to know what the warfighter is looking for before the warfighter even knows to ask the question. It's going to be the healthcare professional that's going to get the information surface to them because the AI is sitting in the background looking at what the healthcare professional is doing to then say, hey, be careful or look at this or analyze this.

31:22And ensuring that we've curated the right types of data and train the models, train the agents to understand what they're supposed to do. That's where we've got to be really good at tightening down what is the use case, what is the outcome, to then making sure that we're presenting accurate information because it could be so critical for the mission outcome and the thing that we're trying to deliver on. And so the data doesn't have to be perfect, but we've got to train the AI to understand what it needs to look for in order to weed out that imperfect data and ensure that we're grabbing the right information, the trusted information to then take action or have it take action on what's being presented.

32:07you know i like to think of it a lot as um you know you think about like what are some of the best performing high-level folks in the world they have access to things like they have a chief of staff they have a board of directors of you know trusted you know folks that they can ask questions of and get advice. They have coaches. So imagine a world where literally everybody on the planet can have almost the same access to those high level capabilities that makes a senior leader an amazing decision maker. Everyone can have access to that and you don't have to be the CEO of Google or Dell or WWT in order to have access to all of that capability.

33:03like the soldier in the field can have that exact same capability. And then they're getting all the sensor information and just something as simple as a trusted entity to ask a question of and say like, I'm seeing, you know, these three things on the horizon. I'm thinking of shooting them. What do you think? Like, you know, it's kind of like that thing. Like I'm seeing these, you know, UAVs coming in that are threats or something like that. And so it's really nice to be able to think of having kind of like a trusted advisor at your fingertips and also a trusted executor of some sort that can take actions on your behalf.

33:49That's why I'm so excited about the partnership with Dell and Google. Dell is providing the hardware to meet the use cases, whether it's small, light form factors that can be on the edge to the latest and greatest from NVIDIA. And then Google has the family of models. So we can pick the right hardware, the right models, and put it with the data, near the data, to be able to serve those mission needs. So while the problems continue to grow, our ability to solve them continue to grow almost as fast. And it's a really exciting time to be working together. Ahmed, I want to go back to what you were talking about in terms of the access and trust.

34:27I mean, how can how can federal leaders actually provide broad enough access to these systems, tools, models, et cetera? But knowing that the federal government or any agency is dealing with various levels of sensitivity, how can they balance that with tight controls that can contain or ensure that trust? Yeah, I mean, so it's really interesting because we have trusted governance and control frameworks. The question is always, what is the implementation method and what is the assurance that those controls are being followed? And so when I get asked this question, I always point back to the fact that like AI is one of its core features is that you can give it guidelines that it can adapt to just like a person can.

35:30I mean, not the same level as a person, but like ultimately like, you know, an AI model and an agent will do different things depending on how you guide it and how you prompt it. And so from my perspective, it should be relatively straightforward to build a plan to say, I am going to inject my existing trusted information governance framework into the AI models that I do that. And that's going to allow me to have confidence that when I provide access to the relevant data, that the AI is going to process it in such a way that it honors my existing governance frameworks. And the cool thing about it is, you know, when you have a, you know, 100 page, you know, information governance framework, and you give it to an AI and say, read it and follow it, it's probably going to complain a lot less than a human employee would that same document.

36:40Just think about like 100 pages being generous, by the way, right? I mean, it's way bigger, more complex, right? And you look at it and you go, you know, again, I like to think of it in terms of sci-fi and what the future would look like, right? And just think of the captain that's telling the ship to avoid any fire and the ship is thinking to make that move, right? Why couldn't that be the future state that we're kind of driving towards where you have this trusted advisor, actioner, executor that's working alongside of you that has the governance framework that knows what right and wrong is based on the material that we presented to then say, hey, these are the things that we follow.

37:24So therefore, this is what we need to go do in the time of whatever, right? Whether it's medical discovery or warfighter activity, what are the guardrails that we need to have in place? And then act on my behalf and help me as we're trying to chart a course forward. That's a super exciting thing to think of. And we're not there today, but there's no reason we shouldn't be aiming for a much broader future where we do have that side-by-side buddy that's going to act on your behalf and make sure that we're executing for the broader mission. Yeah. I mean, so we're talking about trust on the front lines, the end user of AI here, but Ahmed, there's also another component of trust here and that's on the investment side.

38:08How should AI leaders be thinking about investing in some of these tools and resources, knowing that there's not always an immediate clear path to ROI? How can organizations or agencies, how can they develop a trust in those tools and resources that they're going to deliver value down the line? Yeah, that's probably if there was one thread of concern I have over broad adoption and value generation out of AI technology. It's this specific thing, which is there is a natural trust gap that basically says, like, it can't possibly be this good. Right. You know, like it can't possibly achieve all of these world changing results that like one thing can't be everything to me.

39:12And so what we have found is that in order to kind of like get past that is we have to get to a point where the organization is applying, choosing particular use cases to go after first, where you can actually measure the value that has been created. So an example of this would be like, let's say, you know, I am any large government organization and every large government organization, actually every large organization on earth, you know, has a, is a bureaucracy operating system. Like it's the core of how the organization functions. And, you know, these people have to send reports to these people and these people have to create these documents for these other people to review and, you know, all that kind of stuff.

40:08And just starting with automating and improving that entire internal machinery of an organization and looking at the results that that delivers, both in quantitative results of things, decisions happening faster, that kind of thing, as well as qualitative results, which is like people who have to create the content being happier because they get help creating the content. content and people receiving the content being way happier because the content is now better, uniform, follows the rules and is universally better. And so that kind of thing where you're leveraging it for the internal workings of an organization, that's an example of something that is right in front of you, closest to the doc, can apply AI to it and can easily measure it.

41:03So it's almost like a way to get traction and then to move on to the next thing. But build upon that and say, okay, now you've got this internal use case that you've put the right guardrails around. You can almost use that as a readiness assessment to say, okay, here's how we can scale it into more mission outcomes, more larger use cases. But it can be your proving ground, right? We totally believe what you're saying. We at Dell are using it that way today just to make sure that our user base is comfortable with the environment. But as they start thinking about it and using it in a protected environment, you know, the adoption is growing, you know, leaps and bounds and new ideas are coming up.

41:45Well, can I use this? Can I do this? Well, you know, maybe not yet, but let's go figure out how we can go implement that capability and grow, you know, the adoption across the enterprise. That's every organization, whether you're government, enterprise, et cetera, should be thinking about how to start adoption, use it as an assessment and kind of understand your path to driving larger scale opportunities. And when we look at successful organizations, it's kind of three things. You need leadership buy-in. The most successful companies that are looking at AI, their CEOs, their leadership are using it and talking about how they use it on a day-to-day basis.

42:21They're providing those playgrounds or training areas for folks to be able to go and test and try with limited fear that they make mistakes. And then training, because you've got, especially across the public sector, you've got a wide variety of age groups. You've got folks who just got out of school and chat GPT and Jim and I help them pass all those exams to people who don't have any understanding of what AI is going to do and how they use it. So we need to make sure we're providing that training from the novices to the expert users to make sure that everyone within our public sector use force has access to the tools and information to help them be successful.

43:02Yeah, strong practical advice there, Walton. Thanks for that. Well, Walton, Suri, Ahmed, thank you so much for joining the podcast today. This was an insightful conversation and hopefully we'll have you all back here soon to talk more because as we know, you know, nothing ever sleeps in AI land. Okay, thanks to Walton, Ahmed, and Suri for joining us today. The headline was that advanced AI is moving into classified networks, but this conversation makes clear that deployment is only the beginning. The real test is whether a soldier, analyst, or public servant can get the right answer from the right data under the right authority in the few seconds when the decision actually matters.

43:39This episode of the AI Proving Ground podcast was co-produced by Naz Baker, Cara Kuhn, and Kelly Chu. Our audio video engineer is John Knobloch. My name is Brian Felt. Thanks for listening. We'll see you next time.