Security Confidential S19 E6 David Linthicum

Manoj Tandon (00:01.47)
Hello everyone, this is your host, Manoj Tandon. Welcome to another episode of Dark Rhino Securities, Security Confidential. We have another fantastic guest today. Before I announce him, though, I gotta remind you to please hit the like and subscribe button. Show us a little bit of love so we can keep bringing you this great content. But without further delay, we’re honored to have David Lithicum join us today. He is a globally recognized AI, cloud, and cybersecurity thought leader. He has several of his own podcasts. And he has more than 30 years of experience in this field. He has advised global companies, startups, and government agencies on AI strategy, cloud architecture, and digital innovation. He’s a five-time best-selling author. I’m going to talk to him a little bit about that. Former CEO and CTO and one of the world’s most influential voices in cloud computing and artificial intelligence. We’re honored to have you, David. Thank you for joining the show.

David Linthicum (01:04.524)
Well, thank you for having me. It’s great to be here.

Manoj Tandon (01:07.025)
Yeah. You know, give us a little bit about your background. We were talking about it off camera a little bit, but you know, how did you get started in this business, if you will?

David Linthicum (01:10.242)
Ha.

David Linthicum (01:18.25)
Manoj, I’m very old. Very, very old. It started in nineteen eighty-five. So I think I’ve actually got forty years now. And actually was hired initially out of college as an AI analyst, an AI programmer with Boeing Computer Services. I worked on Lisp and Prolog and one, and you know some of the early AI stuff. And yeah, it’s still out there, believe it or not. And so, you know, went through that whole thing; it was decision support stuff and then went through a number of different jobs, different technologies. I kind of

Manoj Tandon (01:35.514)
I remember Lisp.

David Linthicum (01:48.021)
I get bored fairly quickly, and I start switching into different areas. I switched out of AI into land-based development, you know, because the PCs were coming up at the time, and I thought that was emerging technology; switched out of that to service-oriented architecture, switched out of that to web development because the web was suddenly a big thing at the time. This is in the nineties, and then switched out of that into integration focus. And so I wrote the book called Enterprise Application Integration, which took off like a rocket. That was three or four CTO gigs that I got from that.

You know, those things lasted three to four years before we sold the company. Took my shot at being a CEO a couple of times, probably got a B plus doing that. And decided I didn’t want to do that anymore. Found my own company, sold it. Went to Cloud Technology Partners as one of the founders. We sold that to HPE. Boink, I’m this is an amazingly boring story. and then joined Deloitte, which was supposed to be my last job as a chief cloud strategy officer. Stayed there about eight years; I retired two years ago. Aged out, actually, and I’ve been doing the influencer stuff since then and just having a great time. I’ve read a couple of written a couple of books since then: Cloud Computing Insider and Unlocking the Power of Cloud. That’s on top of the seventeen I’ve done over the last thirty years. Still do the cloud blog on InfoWorld, been doing that for fifteen years. And do a lot of writing and speaking and, you know, corporate events and go to conferences when I’m invited, things like that. And then I got my two YouTube channels that I update three or four times a week.

Manoj Tandon (03:11.674)
It’s fantastic. That’s a heck of a background there, David. And how did you manage eighty years at Deloitte? Given that background, I’m surprised that I’m actually surprised that eight years there seems like thirty years somewhere else.

David Linthicum (03:19.682)
Ha.

Okay.

David Linthicum (03:27.106)
Yeah, eight years. It was Deloitte’s an easy place to work. and I and I I talked about two of my people who I brought in there. And another one I went to Deloitte with a bunch of other people who came from cloud technology partners. And so we kind of teamed up and worked together and, you know, got into different areas of the firm. But as people will tell you about Deloitte, you know, kind of once you get in there, it’s easy to stay because it’s, you know, just kind of first-class everything. They have a Deloitte University; they have, you know,

Huge amenities, and you know, everything is, you know, very much in terms of accommodating their employees from the consultants all the way up to the partners. And so I enjoyed it. It was a lot of learning experience. I got to work with a lot of smart people. But the big thing I really loved was getting introduced to clients. So Deloitte has Global 2000 clients, and I was working with dozens of them. You know, had a lot of lessons learned that I couldn’t learn earlier because we were focused on small, medium-sized businesses at cloud technology partners and tech companies before that.

Manoj Tandon (04:14.904)
Sure.

David Linthicum (04:26.274)
So I got into working with some big, y huge accounts in terms of building their cloud and AI strategies and realized they’re kind of a different beast and how to do it correctly and how to deal with the politics within those organizations, how to deal with be becoming a leader and leading organizations, you know, matr you know, matrix management, all that kind of stuff. And it was a great learning experience for me; I came out of there much smarter than when I came in.

Manoj Tandon (04:51.362)
So, you know, looking at AI now, I’ve heard from people who are at Accenture, and they say there are eight hundred thousand employees and the markets want them to downsize because of AI. They’re saying, you know, a lot of the roles there can now be reduced. and has apparently been punishing their stock as such, or I don’t follow them personally, but this is just what insiders have been saying.

Do you see any r truth to that that firms like Deloitte and Accenture are gonna ha see a big downsizing coming up here?

David Linthicum (05:27.478)
I think so. You know, I even look at the analyst space. in other words, what and that’s someone what consultants can do. I’m just amazed at the tools that I have today where I can use them as a force multiplier to get into some larger analysis and get to larger answers in a very short period of time as a single person, you know, as a per one person versus a team of individuals, management consultants that are sitting with a board of directors. But I can go obtain the data.

I can analyze the data. I can figure out the trends within the data, the patterns within the data, and make pretty good predictions that would cost, you know, probably half a million dollars from Mackenzie or, you know, BCG or even Deloitte, with the tools that, you know, sit on our desktop. And that’s, you know, generative AI, LLMs, you know, based systems, as long as you know how to use it. You have to, you know, spend some time training yourself on how to make it happen. So I can see how it’s going to impact them.

Manoj Tandon (06:15.263)
Sure.

David Linthicum (06:19.328)
And I can see how it’s going to impact also the job market because the entry-level jobs that you’re going to do are typically going to be busy work, you know, formatting spreadsheets and things like that and creating, you know, creating small tactical models. Well, AI can do that as well, if not better, and typically better, certainly if someone’s trying to learn. And that’s going to have to be something we’re going to have to adjust to. It was funny, I put a post out today on LinkedIn, you know, about the larger tech analyst firms. And I guess I’m in the tech analyst space, but

Manoj Tandon (06:45.78)
Okay.

David Linthicum (06:46.936)
You know, I’m able to use SurveyMonkey and AI to basically do reports that, you know, provide, I think, 20 times better information and they’re not biased. In other words, I wasn’t paid to, you know, maybe come up with some sort of a narrative. And I think that’s going to be a force multiplier for lots of folks out there. So I can do things better than the analyst firms out there. I can do a lot of things better than consulting firms out there, certainly the architecture stuff. And I can do it as a single human being. And so if I can do it, you know, I’m not the smartest person in the world.

There should be other folks within these businesses that realize that we’re not gonna need these very expensive consultants that are charging, like, a thousand, fifteen hundred bucks an hour to show up and provide, you know, these types of services where they were needed four years ago.

Manoj Tandon (07:32.34)
And contrast this with where we came from in 1985. Was this even imaginable? The world of Lisp and Prolog. And, you know, AI originated even before that, in the 1950s. A lot of the mathematics was done on it. From where that was to where we are today, could you contrast that a little bit? How far we have come?

Could this was was this even imaginable?

David Linthicum (08:04.492)
You know, it’s funny, it was imaginable because I was in a lot of AI SIGs, you know, special interest groups, you know, when we had the we didn’t have the internet, we had to actually get together in someone’s office and meet once a month. And they were talking about AI really being two thousand one of space odyssey, the ability to have conversations and have intelligence that really kind of rivals that of a human or exceeds that of a human. And so that was always the objective.

It just we went back and looked at our list models and went back and did, you know, some of the neural networking stuff, and it wasn’t anywhere close to doing that. And by the way, it cost a million dollars just to do anything of worthiness in the space in nineteen eighty-five. But what we did know, we knew how to do it, but we didn’t have the power and the ability to spend the money to make it happen. So something like generative AI, we couldn’t contemplate it in nineteen eighty-five, but it would have cost trillions of dollars.

You know, to get to a computing platform that just basically acts as, you know, GPT three point five. And the reality is we knew about how AI could excel. We knew about some generative AI capabilities, certainly generative. I was building AI agents in nineteen eighty-five, by the way. And ultimately we

Manoj Tandon (08:55.951)
Sure.

Manoj Tandon (09:14.662)
Well, give us an example ofwhat’ss surprising to hear. What would an agent back then have done?

David Linthicum (09:22.358)
There are autonomous agents that had small that had knowledge models associated with them. And I used to put them in devices that were monitoring things at NASA when I was working on the sh on the shuttle payload project. And so if they had a device that would operate in the clean room, which I was working on, it would be basically a PC that was configured as a device, so I’d have the agents running on the device. And so they would be autonomous in themselves, and the reason for that is the connectivity was so bad.

So we didn’t get about; we couldn’t have networks we could depend on. And so the thing could operate any kind of a situation and make core decisions and learn, you know, learn as it go learn as it it went. That’s what agents do at the end of the day. Well, now we have agents that are bound to an LLM, but they were, in essence, you know, small models that were, you know, very tactically focused. I think I was using M1 at the time. And it was just out of necessity in terms of the architecture; I had to build these things- hundreds of these field interface devices. We call them FIDs.

Which had their own AI agents running on them because the networking was so unreliable and they could carry on operations even if connectivity was down. And so that was an architectural need. But I did understand at the time that it was more expensive and also more complex. But you know, we had the architecture, you know, agentic architecture is nothing new. We’ve known about that for a long period of time. And I built a few of those things. So you use it, but you use it for special situations.

Manoj Tandon (10:28.975)
Hmm.

David Linthicum (10:46.562)
Where it’s going to add value. That’s what, you know, how AI should be configured. And so, you know, when I look at an AI problem domain back then, and even now, I’m trying to get by with a minimum viable technology. I’m trying to spend the least amount of money, use the simplest solution that’s going to provide the most value to the business. And you know, that’s not always what the narrative talks about, but I think that’s the best way to use it. And back then we did the same thing. In other words, I wouldn’t not

recommend agentic AI to my clients, in this case NASA, unless there was some value in doing it. And it was based on the capability of it; he was able to turn value back to the mission. And that’s, you know, still around today. The architecture tricks I learned, you know, in the 80s and 90s, you know, were foundational to my books I wrote and how I teach architecture and, you know, the ability to kind of look at this stuff. At the end of the day, we’re trying to select technology and configure technology.

That’s gonna return the most value to the business. The technology has changed. It’s gotten tremendously better. And I think that’s great. We have the tools in front of us that allow this stuff to happen. But the same processes and thinking still needs to occur behind the scenes.

Manoj Tandon (11:54.613)
What was the transformation moment in modern times that’s given us the rise to LLMs?

David Linthicum (12:01.908)
I think it was the ability to spend the money, and it was access to the processing that we didn’t have. So, you know, I saw the academic papers, and I wasn’t really much into that world, but I would see them come across, you know, in terms of talking about they may not have mentioned generative AI, but it was basically talking about generative AI. And so they knew how to do it. And I would read those papers like, we can do this, but we’re gonna have to spend half a billion dollars to build the first model that’s willing to do that. And

The reality is, well, LLMs are so powerful because they’re built with the whole of the information that’s on the internet. Number one, we didn’t have the internet, you know, 20 years ago. We couldn’t digitally access all these various pieces of information. Now we do, and now we have the ability to spend the money to build these very powerful models, not because they’re powerful, but because the technology is unique, but they’re powerful because of the training data that’s available to them. And so it was kind of a perfect storm. Someone had the money.

To spend, in this case, OpenAI, but there are a few other companies that were doing as well. We had access to the information for the open internet. We’re dumb enough to put everything online that, you know, is out there that we can train from, able to crawl those spaces. And two, you know, they came together, and it came up with the first LLM. And that was kind of the Eureka moment for them because people who used it for the first time, they go, this is this is something different. I can have it write a letter basically using my voice. I can generate images, I can generate music, I can generate videos.

You know, all these capabilities that we knew were gonna be there, we just did not have the money, the processing power, and the access to the information that it needed to make that happen.

Manoj Tandon (13:39.349)
So GPU computing and, I guess, the Transformer paper at Google was, you know, those were, from what I have surmised, two critical pieces that came into play that allowed us to scale these models at industrial capacity without breaking the bank. Well, you still break the bank. If you want to build a foundational model today, that’s not for the faint of heart.

David Linthicum (14:01.4)
You still break the bank.

David Linthicum (14:06.432)
No, yeah, at least most of them are at least $100 million to train. And some of them can take a couple of months. I don’t know what they’re doing now; the benchmarks are changing a great deal, but only the big companies have foundational models because they’re so expensive to create. I don’t know why why why we need 200 of them now, which we is what we have, and we get a new one every couple of a couple of months, every couple of weeks. But that was it. So in other words, we realized that that was the power. We always knew.

There was something there; I don’t think people quite imagine, you know, how much that was going to have a far-reaching impact on us as people, the economy, you know, kind of changing the game and reinventing the way in which we look at technology. So it’s technology that, from a foundational thinking point of view, was always available to us. We knew it was possible to make it happen. We didn’t have the money and didn’t have the wherewithal to go ahead and build those models. Someone was brave enough to do it. And suddenly we have LLMs. And once someone did it,

You know, you had 20 other people do it in the next six months, and you had a hundred people do it in the next, you know, next couple of years. And now it’s at full-blown blast. Anybody can knows how to build an LLM. There are open models out there. You can use Llama. You know, you can build a reasonable LLM, you know, even on your local hardware, believe it or not. It’s gonna have some limitations, but we know how to do it. The cat’s out of the bag, and it’s just like you remember the four-minute mile?

No one would break the four-minute mile until somebody broke the four-minute mile. And then suddenly over the next six months, there were hundreds of people who broke the four-minute mile. So I think the person who is the pioneer is the person who realized I’m going to spend a lot of money, put a lot of risk in involved in it, and see if we can make it happen. And guess what? It worked out. But they were built on the thoughts and the thinking and the work of thousands of scientists, thousands of college professors, thousands of people who did the research.

that allowed them to have the jumping-off point in order to create generative AI and basically instances of generative AI that were going to be useful, namely an LLN. So here we are.

Manoj Tandon (16:08.411)
Well speak and speaking of jumping-off points, we come to today, and the AI world changes pretty much daily, almost. As you rightfully said, we’ve got several models, and depending on the day of the week, one is going to surpass the other in certain capabilities, and it’s a continuing race to what finish line? We don’t know. But, as a business leader, do you

Take your organization, because back when there was digital transformation, that was still kind of slow. And there’s still, believe it or not, people out there, David, that are still working on digital transformation. And now we’re talking about AI transformation, adoption of AI in the organization. In a world that’s changing faster than you can think, where do you advise a company? Where do they start? What do they do?

David Linthicum (17:05.176)
Number one, my f big biggest piece of advice is to calm down. I think everybody is as worried about their fear of missing out, that their competitors are gonna get some kind of a system running before they do and they’re gonna get wiped out of the market. Number one, understand your own requirements and also understand which one of the requirements in business problems that you have that are solvable using AI and which ones are not. And guarantee it’s not as many as you think. So

When you’re looking at an enterprise problem domain, you’re looking at all the systems that the enterprise is running. If you’re looking at AI enhancement opportunities for those systems, that’s typically going to be maybe 40% to 45%, sometimes 50% of the systems. It’s not going to be everything in the in the enterprise. Doesn’t need to be updated, doesn’t need to be AI enabled. And normally it’s going to be your edge applications, you know, productivity, your ability to do scheduling, logistics, management, things like that, not things that are core to your business. So figure out where you are.

And what you need to do. And I’m so surprised when I talk to people, they come up to and they said, you know, How do we do AI? And I said, Well, what problems are you solving? What business are you in? What requirements? What, you know, compliance issues are you dealing with? And where is your point of latency? And also, where do you think additional innovation or use of AI will add a force multiplier for the business? There’s always something to make it happen. And then figure out where it’s

Manoj Tandon (18:21.349)
How many people even know the answer to that question that you just asked?

David Linthicum (18:25.086)
I think it’s just look at it with the brain of someone who understands the technology and figure out, you know, where you know, one of the exercises we do in my architecture class, we read these case studies, and we go AI, not AI. And we talk about the differences in using AI versus not using AI versus force fitting AI. So it’s getting some smart people in your organization that either buy or rent that help you look at all your core systems and figure out what the opportunities are.

Where AI can be applied, and also probably more importantly, where it shouldn’t be applied. Almost every client that I have that’s looking at the AI transformation, which is part of the digital transformation universe, that people are that’s an ill-defined concept that we’re still playing with. But we’re going to make things better and more automated, in essence. Call it digital transformation, whatever. They fail to look at them; they show up with pet systems that they want me toAI-enablee. In other words, we have logistics management. Inventory control, and we think AI is going to have a big impact, and AI should be nowhere near those systems. They’re transaction oriented. They don’t need to make decisions. You know, having, you know, deterministic capabilities on these things is not necessarily going to have a core value. And it becomes contraindicated where AI is not going to be in that part of the system. They just don’t understand it yet. They just think it’s going to happen. So having someone look at what your transformation is, look at the larger architecture.

Look at where this technology can be applied and where it can’t be applied. We did the same thing with cloud computing 15 years ago. Now everything’s AI first. Well, guess what? 10, 15 years ago was cloud first. And where not every workload, not every data set needed to be in the cloud; maybe 30 to 40 percent would be the maximum that the enterprises needed to transform in the cloud. Yet everybody tried to make this mad dash into the cloud, move things into the cloud that shouldn’t have been moved into the cloud, and now end up having huge amounts of technical debt that they have to deal with. They’re spending

Manoj Tandon (19:58.283)
Yeah.

David Linthicum (20:17.966)
You know, as much many as much as 10 times operating those systems than they should be spending. Same thing with AI, but it’s even more dangerous. That’s 20 times, 30 times the cost of traditional environments. So in other words, if you’re AI enabling something today or using AI or force fitting AI, that’s going to be a key technology within your applications, within your business domain, you’re going to end up misapplying it, and you’re ultimately going to build inefficiency in your in your systems, and you’re just going to end up hemorrhaging cash.

Because guess what? I can make all those things work. I always tell them I can take your inventory system, I can enable it with AI, and I’m gonna be able to make it work, but it’s gonna cost 10 times, 20 times more than your system is running now, and it’s not gonna return that much value because AI is not gonna have a strategic impact for those particular for that particular system. So it’s the ability to look at all that stuff. My biggest fear is people misapplying the technology.

not getting the return on their investment, very much like did in the cloud computing stuff, and they’re gonna end up taking some of these businesses out because they’re gonna run up so much technical debt, misunderstanding, misapplying the technology that they’re gonna end up you know, in in ways where they don’t have the resources to do the things that the business needs to do. So if you’re spending money on one system trying to AI-enable it, that means other systems aren’t going to be AI-enabled. That really should be. And I think the misallocation of the money and the resources.

Manoj Tandon (21:16.832)
Yeah.

David Linthicum (21:42.499)
To throw AI at everything in there. It’s just a misapplication of the technology. You know, like I said in a video I did a couple of weeks ago, you know, all the CEOs are running around AI first, AI first, enterprise gotta do AI first. It’s a bad strategy. It shouldn’t be AI; it should be AI sometimes when you can justify it. And AI sometimes when you can pay for it. And I know I’m a big bummer when I say that and kind of a buzzkill based on all the, you know, hype that’s going out in the market.

Manoj Tandon (22:06.634)
Think so, but go keep going.

David Linthicum (22:08.182)
Yeah, in the market right now, but I think that’s the reality of it. I know people don’t want to hear it, but I think that message is starting to resonate with people.

Manoj Tandon (22:14.079)
Can you give us a verbal matrix? You mentioned in your architecture classes you guys go through use cases and you’re like, yes or no, AI. And you mentioned like logistics. Bad idea. Let’s keep humans with that. You already have a deterministic system. Are there other ones that our audience you can enlighten them on that, or how do you come up with that or areas that definitely don’t bother with, right?

David Linthicum (22:43.448)
Typically, things that are transactional, you gotta look at each bus, each application and solution on its own merit. So you can’t really, you know, talk just profile it based on what it is. But normally systems that are very transaction oriented that are making they’re making structured calls, that they don’t they’re not gonna benefit from dynamic changing of decision making. In other words, the knowing and the understanding and building on the model, probabilistic stuff. and

That would be an indication to me that it probably shouldn’t be AI. I’d have to look at it in detail to see what they’re wanting to do and some of the opportunities in the system. But if it’s just a sales order entry system, I can certainly use AI to, you know, track clients and, you know, do the CRM stuff that’s on the edges of that system. But as far as core to the system, probably doesn’t need to be AI enabled because what it is isn’t that complicated. And it’s going to be good enough for what it’s for. I’d rather take that money and spend it in other areas.

areas where, you know, a good example would be, you know, supply chain management. Typically that may be a good use of AI because ultimately you’re having a dynamic understanding of the different supply chain characters- things that are out of your control: the suppliers, raw material producers, shipping, you know, all that kind of stuff, which is gonna be changing a lot over time. And you’re gonna have to make decisions and

And understand the patterns in terms of what is going to be the most efficient way to build something, order from somebody, ship something, things like that. Normally, something like that, AI is going to have a benefit because, number one, I’m putting in a decision-making entity, in this case, kind of a fake brain, that’s able to make some tactical calls that normally humans would make about using different entities, using different suppliers. And so the complexity of that system would be mediated by an AI system.

And it wouldn’t be that I could do that with machine learning. I wouldn’t have to use a gen a generative AI to make that happen. So those are things you know to kind of keep in mind. But you have to look at the core mission of the system, what it’s capable of doing. And so we’re you know, agentic versus non-agentic, that’s even more fun to play that game. But ultimately AI is very good at working with systems that need dynamic changes, need to change their behavior over time. In other words, adjust to

David Linthicum (24:59.96)
Changes in the environment, state changes, changes in the data. Because if we have to do that as humans, and basically through the system and through some sort of automated system, we’re not gonna we’re not gonna be able to do it as fast and efficiently as AI. But that is probably maybe 30, 40% of the systems that you have within your enterprise. And also you have to look at the core and non-core stuff. In other words, is this application core to what we do, or is it some sort of an edge-based application? So if it’s sales order entry, inventory control, shipping and logistics, that’s core to the business. Okay, what is it doing? And typically we’re not building AI in those things right now. So it’s gonna be more money and time to build that. However, if it has a justifiable use case where AI is gonna be, you know, of value, then we need to study it, you know, figure out the architecture, create the logical architecture, and then map it into a physical architecture and see if it’s something that we’re gonna be able to do and something we’re able to build.

And then guess the amount of value that it’s going to be able to bring back to the business. And by the way, if it’s going to be less than I think, then that may be a reason not to build it. And so I would move on and put my money in something else. So it has to be a game where we don’t think we have a massive pile of money that’s going to be unlimited. No one has a money tree in the back of the business. We have a certain pile of cash. We have to spend the cash in reasonable ways that are going to have the most. Value strategic and tactical values to the business. And that means a lot of thinking in terms of how you’re going to use it. You know, one of the things that the AI vendors will come to me and they go, we’re really frustrated with the adoption of AI is not occurring at the pace that we thought it was going to occur. You know, what they heard a couple of years ago. Was that what you didn’t hear from me. And number one, they don’t have a pile of money, guys. They they they’re they they’re only able to spend 15% of their their budget.

On IT; that’s typically gonna be best practice. That’s not unlimited funds, and they’re not able to buy these systems that are ten, twenty times the cost of traditional environments. And they talk about, well, it’s strategic; they have to do it, you know, get in and become competitive in the space. I get it. But what you’re telling them to do is to spend massive amounts of cash on things that are very arbitrary, things that are very random. And so if they do that, they’re gonna make the wrong guesses. And they say, Well, yeah, we don’t see that happening. Look at cloud. And then they

Manoj Tandon (26:54.199)
Right.

David Linthicum (27:18.882)
They shut up. And I say, and we have, you know, we’re back there trying to fix these things now. If we do this with AI, you’re gonna kill those businesses. They’re gonna start hemorrhaging cash because the technical debt is not only just gonna be three or four times, you know, that of a traditional environment, it’s gonna be twenty to thirty times. And they’re gonna have to operate those systems, do so at scale, and they’re not gonna have the resources to put in other places where they should be spending the money. So think about that when you sell this stuff. And I don’t think people are doing that now. It right now it’s a big it’s a big party.

Manoj Tandon (27:46.231)
No.

David Linthicum (27:48.623)
You know, I go to all the conferences; everything is an agentic AI, a generative AI conference. A lot of cheerleading, gotta do this, make it first, AI first, enterprise and whatever. But it needs to be dumbed down a bit. Like I mentioned at the beginning of our conversation. Everybody needs to calm down and take a look at what the real value of this technology is for their particular use case. And many people think they already understand that, and they don’t.

Manoj Tandon (28:13.718)
Amen to that. I would agree wholeheartedly. Most people don’t understand it. I guess, you know, one area that I’ll come back to is like you mentioned, there are things core to your business that you really have to look at on a case-by-case basis before you let the AI beast loose on it. But we see often customer service getting turned into an agentic AI or an AI chatbot of some kind.

What are your thoughts on it? And before you respond, I’ll say this: to me, depending on the nature of your business, if it’s human touch oriented and you lose that relationship with the end client to an automated robot, then I would argue you’ve actually lost something of significance to your organization. But it seems to be

David Linthicum (29:08.248)
Yeah.

Manoj Tandon (29:11.7)
You were just saying it’s becoming ubiquitous where a lot of companies are just killing off their phone people and saying, Yeah, deal with the chatbot, man.

David Linthicum (29:24.312)
Yeah, well, in my architecture class, we’re actually designing an employee-free company that’s gonna have no employees. And actually the hardest part of that, and just just just to say we did. the hardest part of that is the customer service element of it. Everything else is automatable. In other words, I can sell stuff, I can ship stuff, I can, you know, even track the shipment, things like that. I can use dropship, I can figure out how to deal with the manufacturer. This is a case of a retail organization just selling somebody else’s stuff.

Manoj Tandon (29:30.613)
Ha ha

David Linthicum (29:53.231)
But I go, customer service. How are we going to do that? And I put that to the class. And that’s actually the toughest thing to do. I told them, I said, if we’re able to figure that out, that’s core to this system. So I can certainly try to automate a lot of the customer service system using chatbots and IBR systems and things like that. And by the way, there is the ability for AI to add value in those situations because it’s able to communicate with people with different dialects, different languages. It’s able to send, you know.

customized videos to people that it’s generated dynamically based on how it’s going to solve their particular problem in their language, in their dialect, you know, in ways in which they’re going to understan,d in in even considering, you know, who they are in their educational background. And all that stuff is available to you today, but you can’t remove the human out of the loop. There has to be the ability to talk to a human to solve the issue. And you don’t need as many humans if you do that, but that’s always going to be the case. And people will argue with me.

They say, well, AI can do this and this and this. I get it. I understand it has magical capabilities and we’re able to do lots of stuff we couldn’t do, you know, five years ago with the same sort of a customer service system. But at the end of the day, there still has to be some human interaction, or people are not gonna like it. People hate AI. I always tell people that, you know, who are in and out of the industry. They hate it. Okay. It’s building data centers. They don’t want the data centers there taking jobs; they don’t want their jobs taken. And now you’re having them talk to a chatbot when they’re trying to get their television fixed or their cable back on.

Things like that. And I do it all the time. I’m sure you do it all the time. You have to put some humanity in those things where you’re able to get to a human being to solve many of those issues. And we’re not going to be able to take the humans out of the loop unless we start dealing with non-humans, which we’re never going to deal with. If we’re dealing with robots, easy. But if we’re dealing with humans, we have to have the capability to have someone with empathy solve their issues and basically find the solution to the problem. Now we can take them down a hierarchy, in other words,

You know, try to send the documents, try to help them digitally. And sometimes it’s even better than waiting, you know, waiting in a queue for someone to answer the phone. But eventually you gotta have that human element in these systems that’s going to keep the humans interested in that company. Anybody who adopts, I always tell people that. I understand you can you can there’s a there’s package systems you can buy that are complete AI-driven customer service systems, and they’re amazing. But when you l launch one of those systems, people are gonna hate you. And

David Linthicum (32:15.192)
They’re gonna figure out pretty quickly that it’s AI, and they’re gonna figure out pretty quickly that you don’t care about them, you disrespect them, you’re not gonna you’re not gonna be respected. But if they’re able to get their answers better and quicker, it doesn’t matter; it’s the perception of the individuals. And I understand that’s a big bummer as well, but the reality is you’re selling to humans, and humans like dealing with other humans in some instances, and if they don’t get that, if they don’t get that interaction, they’re gonna consider it a failed transaction. And that in these days when you have social media.

You know, Google ranking, all that kind of stuff, people are going to figure it out pretty quickly. If someone has bad customer service, you’re going to know that pretty quickly. You know, cable TV, by the way, legendary bad customer service. They people know that when they call in there. So they’re already pre-disappointed when they start talking to those people. It’s like dread calling into my- you when my internet goes down, trying to get my internet back up again. So, excuse me, we have to take into consideration those things.

Manoj Tandon (32:55.086)
Yeah. Absolutely. And I’m glad you said that because there we see a lot of people who are throwing the first thing they’re throwing at it is customer service. You know, we can take all our help desk support and just totally, totally outsource it.

And what we actually find is that there’s a lot of, you know, things that could be addressed by the FAQs. Yeah, throw an agent at it. But once you get beyond that, and there’s a lot of it, and there’s a lot of nuances, and people at that point want to talk to someone. They want to talk to a human being and get real intelligence, not artificial intelligence, to help them solve the problem.

David Linthicum (34:01.464)
Yeah, they want to choose their own journey at least. And you know, I and by the way, I I hate dealing with humans when I’m dealing with airline tickets and things like that. I like automated, you know, use my app, use my you’re the web page, things like that. But I’m choosing that journey. But if I chose to talk to a customer service agent, I should be able to do that. You gotta remember people are gonna have different needs. Some people are visually focused, some people are auditory focused, some people wanna deal with the written word.

And deal with chatbots, things like that. Sometimes I’m able to use the tools that they’re able to provide me to find the answer quicker and easier and probably in a cheaper way than talking to a human. But you’re always gonna have to have the human in the loop anytime you’re dealing with any kind of a customer service or sales scenario. And people tend to kind of brush that by. And I think people are over-automating right now, and they’re finding out quick that, number one, people hate AI; they’re gonna hate you for using it, and you’re gonna lose a customer.

I’m not picking on companies here, but there’s case study after case study where they replaced their customer service with many of these chatbot-based, you know, customer service agents. And it blew up in their face. They had to hire the customer service agents back. And over and over again, I’m seeing that. And that’s because they didn’t anticipate the reaction in terms of what the humans were going to think, their customers were going to think about them using that automation. It’s disrespectful at the very least. And their employees hate it; their customers will hate it.

And you’re gonna find your business has gone away. And I think somebody has a good case study out there. It’s public now. I’m not gonna pick on them, but you know, they lost 20% of their customer base in a couple of months because they, you know, put in this automation thing. And by the way, they announced the fact that they were gonna do this automation, their courses were gonna be automated, and the customer service was gonna be automated completely by AI. And, you know, hey, look at us, how automated we are. And everybody completely across the board panned it.

And unsubscribed from their service until they fixed it, and they backed up on it. We’re seeing a lot of those things today. I just saw one study that talks about, you know, many of these organizations are stepping back from many of the original predictions that they have in terms of how AI is going to impact their business because they realize they can’t do it in a time frame that they thought they could do it, and there’s it’s not gonna have the impact they thought it was going to have. So they’re dialing back the amount of value.

David Linthicum (36:20.066)
That the AI stuff is actually going to be able to create. Now the tech guys are going crazy about that because they’re worried about them not buying their AI after they spent trillions of dollars in building the technology infrastructure. That’s probably going to happen as well. But the reality is, who couldn’t see that coming? You know, we adopt technology very much in slower ways than we think we’re going to adopt it. You know, whereas cloud computing, the web, you know, rise of service-oriented architecture, rise of the PC, you know, all these little technological, you know, innovations that we’ve been through. Is adopted at a much slower pace than they think is going to happen. Cloud computing, for example, when it first came out, Gardner said 80% of the workloads are going to be in the cloud in one year. I remember that report. And I brought it up and showed it on my video. And we’re nowhere near that. Maybe at 40% now, you know, maybe 50% when you consider SaaS. And the reality is they can’t move that fast. They can’t spend that much money. And they didn’t need to have all those workloads and all those data sets reside in the cloud. And people just got better

Manoj Tandon (37:00.596)
Yeah.

David Linthicum (37:18.496)
understanding and better ideas. AI’s gonna be the same way.

Manoj Tandon (37:22.132)
So now, switching a little bit to agentic AI, how do you think that’s going to change the way the product is on the web? And let me give you an example so I may not have asked that question properly. Today when I go to Expedia as an example, and I want to go book an airline ticket, I get shown 10,000 ads of various different kinds. And that’s part of the way they make money. Now, if I program an agent to go and do research on the flights, go ahead and book my tickets, those ads suddenly have no value. In the old model, I am the product.

Now you have an agent that might be doing some of this work, and it could care less what you’re showing it or not showing it. Do you have any thoughts on how that might change the model on the web or in certain areas?

David Linthicum (38:24.76)
Yeah, you’re gonna have to change your model. This is an example where a genetic AI should be used. It’s good at edge cases where it’s looking for cheap flights, you know, it’s basically looking at a website or websites to figure out how the flights are trending up or down, cheaper hotel rooms, you know, looking for real estate for you, trying to buy furniture, you know, whatever. And we’re able to do these pre-programmed things that are able to operate continuously and act proactively. So we’re no longer dealing with the web as the interface, and that’s the problem.

If we’re no longer dealing with the web or dealing with the applications, the agents are doing it on our behalf, then the agents are the ones who are gonna see the ads, and they’re gonna skip right by it. So guess what? You guys have to, you know, figure something else out. So we figured the fact that we could put ads up on every web page out there to, you know, make some additional income, and the social media guys have to live off that. And that’s something that’s not gonna go on in perpetuity. So eventually we’re gonna figure out how to abstract ourselves away from those interfaces. We’re not gonna see the ads, and they’re not gonna make money.

And I’m not crying for them. They have to shift around and shift their models. But, you know, lots of businesses are going to change because of the addition of this technology. Not that agentic is this, you know, you know, game-changing stuff where everybody’s gonna, you know, basically have agents run, you know, run everything out there. McKenzie is saying that twenty percent of his employees are agents. Come on, give me a break. But the ability to provide some more productivity opportunities for us by not having to go through the monotony of seeing these ads and be able to skip right by it. And so that we use automation to make our lives better. And we’ve been doing this for years. We’re getting better at automation. The advertisement stuff is not going to be an issue. I mean, just like, you know, a podcast. You know, if you don’t want to give your ads on a podcast, there’s an automated system to remove the ads and just play the text. And some people want to consume information that way. And I think that’s going to change some of the industries in the fact that the web stuff.

Even the software as a service systems are finding an impact as well because the agents are, in essence, abstracting the human beings away. And guess what? They’re charging per seat. So instead of a thousand people that are paying ninety dollars per seat, they have a hundred agents that are paying a hundred dollars per seat for each one of the agents, but they’re doing the same amount of work. They’re still dealing with the humans on the back end, but the agents are mediating between them and the SaaS system. So if that abstraction is gonna happen, we we have to just adopt you know, adapt for it.

David Linthicum (40:49.486)
We’re not gonna create a law that’s gonna, you know, outlaw agents, so people have to look at the web pages, see the odds.

Manoj Tandon (40:57.711)
Let’s chat briefly before we run out of time about AI governance and security. I mean, those are two we got AI, which is the wild, wild west right now, and this whole concept of governance, which can’t possibly keep up with the rapid pace of changes that’s taking place. But if a company is considering adopting AI in certain areas that could have PII or can end up disclosing data on the web

I mean, to me this is really dangerous. You you if you created a data lake, remember those from way back when that had all this stuff in there, and you let AI loose on it? Well, there’s no delete function there, man. I mean, once that data is in there, it’s gonna answer stuff based on that information. How what’s a company to do? Most companies don’t even know where their data is or what all that data is. So how do you handle that, David? Like I

David Linthicum (41:55.065)
You have to be very aware of what data is being used to train your models and understand that your models basically are functional are functionally equivalent to a database. So, and by the way, you’re absolutely right. So once it’s trained on some of this data, it’s going to be very difficult to pull the data out because that pattern’s already been consumed and it’s already been turned into a thousand million different parameters. So just use your head in this. If it’s training on data.

That is some sort of data that has compliance, for example, healthcare data, PII being the big one, then you have to treat the model as also a recipient of PII information, including being aware of the law and what it says. And sometimes I think that’s illegal for you to do that, unless you have the unless that is certified as a piece of technology. You have- I can’t remember the details- but you had to have a manager that’s associated with it and all that kind of stuff. So you

We know what the rules are. We just have to understand how it applies to a new set of technology. And when you train a knowledge model, it’s nothing more than reading the data and turning it into parameters. And so we’re going to assume that the knowledge model is going to be the recipient and container of the data. And if that data is under some compliance, you have to deal with it. We have to deal with governance and the data side. In other words, dealing with compliance and how we’re dealing with databases, that same stuff has to exist on the knowledge side.

And it’s going to be a replication or a copying of the data, and you’re going to have to basically deal with it. You know, deal with it as such. And I was- people always say we’re gonna use fewer governance people and, you know, security people now that we’re going into AI. No, you’re gonna need twice as many. I can’t every time I map out a project and look at the operational impact, in other words, operationalize the systems we’re building, there’s always gonna be twice as many security people and twice as many governance people as they currently have, just because of the problems you just mentioned.

So those things have to be governed, those have to be planned, those things have to be managed. And the way in which you govern a model is gonna be very different than the way in which you govern a data governing a database.

Manoj Tandon (43:56.724)
Absolutely. And I think that point is lost on a lot of people. The governance of the model versus governance of people, I think it’s much easier to govern people. And especially when you look at some of the models as they are evolving and they’re able to, I guess what the word I’m looking for is? They’re able to break the boundaries that you might have laid out for them. So you’ve told it the prime directive of solving this problem, but to solve that problem, it has to go past the boundary that you may have set for it. And it’s not intelligent, but it has enough pattern recognition to know, I’m gonna go do that anyway. Now you got a problem.

David Linthicum (44:37.226)
It does. I mean, I have a I had a case years ago when I was doing a system for a company that had an NIH grant, and we had anonymized data, in other words, data that came from patient outcomes that did not have PII information associated with it. We uploaded it into a model, and guess what? The model figured it out.

How it could backfill the PII information and had a 95% success rate in going out on the open internet and figuring out who those people were, brought it into the system. So when I looked at it, suddenly this thing’s out of compliance, guys. You can’t have PII information here. No one entered it in. They just had the command: can you match up the data, you know, with existing known populations out there? And they didn’t even know it was going to do that and went ahead and carried it out and was able to figure out, like, it had access to the web and access to patient records and patient databases, able to figure out.

Manoj Tandon (45:19.231)
Exactly.

David Linthicum (45:30.05)
You know, that this information was there and suddenly they violated the law. Someone got arrested for that. Those sorts of things are going to happen all over the place because people just do the same thing with agents deleting the data, and they’re doing, you know, dumb things because they’re not putting governance and thinking around, you know, setting these things loose. You gotta remember AI has dynamic behavior; it changes around stimulus; it’s able to learn, it’s able to adjust, it’s able to, you know, figure things out and get better and better and better at what it does. That’s why we’re using it. However, that same getting better can be weaponized against you.

If you don’t figure out how to put governance around it, then I don’t think people have a clue in how that’s going to be done. In other words, they’re looking at We know how to do it within the data. You know, keeping PI information at legal data compliance governance, things like that within the knowledge model. I don’t think you have a clue how to govern those things. And there’s I just wrote a book on governance, unlocking the power of cloud. And you know, myself and my co-author, Meredith Sign, you know, we talked about that, the ability to kind of put these processes in place.

You know, create a culture and look at the tools and technology that are able to do it. And it is not an easy thing to do. It’s a complex array of things. You have to deal with something at the knowledge level. You have to figure out the future dynamic nature of the model, how it’s going to change itself over time, and the ability to put limitations on that growth. And that’s something people have trouble thinking about. We’re used to dealing with static things that unless we change it, it’s not going to change. AI is completely different.

Manoj Tandon (46:56.061)
I’ll have to pick up a copy of your book there. That sounds like an interesting read because that problem is a very real one. And you know, in a lot of small businesses where we’re just seeing the- I kid you not- Microsoft’s putting up the pop-up. Are you okay with letting Copilot be active? And then you say yes, and it says, Are you authorized by your company to allow this action?

Yes, and the person clicking that has no idea what that thing’s about to go do. And before you know it, now you let Copilot loose like that. Well, you can actually search up what somebody else’s income is in the company. I can’t make that up. You know that’s rare, all story. It happened.

David Linthicum (47:42.85)
No, it’s true. They charge you for the privilege. Yeah. Yeah. Copilot became something that Microsoft has to undo. One of many things.

Manoj Tandon (47:55.983)
But it’s very dangerous to turn those on just like that. And especially if your data is not segmented, your network’s not segmented, you don’t know exactly what you gotta be really focused with these things, I guess. It’s like anything else. You you it’s it’s aneed-to-knoww basis, maybe the best way to deal with AI. Does it need to know? And if it doesn’t

David Linthicum (48:16.718)
That’s right. And I think even dumbing down some of this AI, you know, enthusiasm instead of using large language models all the time, which we’re doing now, even to run agents, but the ability to use very narrowly focused small language models because they’re not going to be as dangerous and they’re much easier to govern. And by the way, I can build those in an hour, you know, for, you know, 11 cents of cloud computing time, where an LLM takes a month and takes a hundred thousand, a hundred million dollars of cloud computing time. So

It’s overly focused, overly using, overly trusting that’s causing the issues. We’re throwing everything in an LLM right now, and it’s really kind of never meant to do that. The people who built the LLMs, they didn’t really design them to do that. They designed them to be prompt and API access. I’m sorry, API access to do very tactical things. And what companies are doing is they’re pushing everything into a front and into a frontier model. And that’s not what they’re designed to do. And that’s where the danger occurs. In other words, you’re putting it up in a shared space.

It’s own; it’s gonna be owned infrastructure, owned by somebody else, maintained by somebody else, and somebody else may have the capability of seeing it, but you haven’t really done the auditing to figure out how that’s going to work. You just realize it’s very easy for you to drag and drop your database, you know, into ChatGPT and then ask questions of ChatGPT about the data. And I know it works very well, but you know, how many things did you violate? How much security, you know, parameters have you have you have you violated? And by the way, we catch people doing that all the time. In other words, it

Manoj Tandon (49:42.875)
yeah.

David Linthicum (49:43.727)
Yeah, I mean, and because they go, well, we even were told to use ChatGPT, and you know, use it efficiently. And so we’re not putting, you know, client data in there and things like that. And there was no thinking and no checking with the security folks. The C in the CISO didn’t know anything about it. That’s because everybody has access to these tools that are working in productivity-type tasks, that are running on their desktops. Everybody uses it every day. Everybody’s automating their stuff because it’s an easy way to, you know, work an eight-hour day in two hours. and then

They’re violating many of the security rules because there’s a lack of training, a lack of culture to understand what they’re doing. And that has to be first and foremost. So if you’re going to use these tools, which I encourage people to use, I want you to be productive. There has to be some understanding in terms of the security issues that you’re going to be dealing with. And it’s it’s it’s out of epidemic proportion now. All this shadow AI stuff is gonna kill us, very much like the, you know, Shadow Cloud was 15 years ago. People are, you know, uploading.

you know, a lot of the corporate data into the cloud providers. It’s worse in the AI system because you’re dealing with a co-located knowledge model. In other words, that n it’s a single knowledge model that everybody’s using. You’re in essence having a certain namespace in there that you’re going to be operating within, but you don’t know how the data’s being handled. You don’t know how it’s being backed up. You don’t know how it’s being dealt with. And you don’t know, by the way, and I think this is actually the case, if the model’s going to train itself on the data you’re uploading. I would- you- I would

You know, I train myself on any kind of data I’m exposed to if I’m building a model. So suddenly they’re gonna be trained on the corporate data, including the salaries of your CEOs. And so someone types, you know, in ChatGPT in a couple of months, you know, who, you know, what does you know what does Don CEO make? And hey, he makes a million bucks a year. How did he know that? Well, that didn’t come out of the 10-Q; it came out of the data someone was stupid enough to upload.

Manoj Tandon (51:33.108)
And on that note, David, we’re actually at the hour. We’d love to give you a couple of minutes to plug whatever you’d like our audience to know about.

David Linthicum (51:40.332)
Yeah, easy to find. Check me out on LinkedIn, out on X. Also two YouTube channels, Cloud Computing Insider and Dave Linthicum is not AI. You can find those on YouTube. Check me out.

Manoj Tandon (51:51.498)
Fantastic. David, it’s been we didn’t get through all the questions, but I knew that would that was gonna happen. But it was fantastic having you as a guest. Please stay in touch, and we’ll look forward to seeing you sometime in the future as the AI world evolves to keep us enlightened on all the core topics that are going on here. Thank you so much. Take care.

David Linthicum (52:13.742)
You got it. Thanks again.