09×08: A Realistic Approach to Agentic AI with Nick Patience, The Futurum Group

Although companies are just starting to deploy generative AI industry attention is already turning to AI agents. This episode of Utilizing Tech brings a realistic perspective on the agentic AI timeline with Nick Patients VP and Practice lead for AI at the Futurum Group. Welcome to Utilizing Tech, the podcast about emerging technology from Tech Field Day.

Part of the Futurum Group. This season focuses on practical applications of ag, agentic, ai, and other related innovations in artificial intelligence. I'm your host, Stephen Foskett, organizer of the Tech Field Day event series, including AI Field Day.

And joining me this week for the co-hosting seat is Mr. Frederick Van Hern. Welcome to the show, Frederick.

Thank you. Glad to be here. So my name is Frederick Van Hern.

I'm the founder and CTO of Hyphens, and we provide HPC and AI consulting services. You know, Frederick, uh, we've been talking quite a lot about a agentic AI this season. I guess that's the topic, so it's no surprise.

Um, but I guess, uh, you know, we're still in, uh, early phases of rolling this stuff out. I think that people forget how quickly this field has moved. Yeah, I totally agree.

I think people are still digesting what generative AI is, and guess what? Now we're talking about agen AI and, and agents. Um, I think it just proves how fast all of this is going in the AI world.

The question really is, is how can consumers kind of follow and learn about all these new technologies as they come out? I agree, and I think that's especially difficult for enterprise buyers who fear that, um, you know, there, there's so much news about this. They fear that they're being left behind maybe, but they're really not.

Uh, this is really early stages of the development of this technology, and that's why, uh, this week we've decided to bring in, uh, one of the folks here from the Futurum Group who really focuses on this, uh, Nick Patience, who is able to maybe provide a little perspective and a little realism. Where are we really when it comes to Agent ai? So, Nick, welcome to the show.

Thanks, Steven. Thanks for having me. So, yeah, as Steven said, my name's Nick Patience.

I am the, um, vice president and AI platform's practice lead at Futurum Research, another part of the, of the Futurum Group. Um, so I'm the kind of principal AI analyst here. Everybody's an AI analyst to a certain extent, but I, I really focus on, um, the things that are, that are fundamental, um, to ai.

And obviously GEN is part of that. My history, my background, I've been looking at AI for over 25 years. I started at another analyst company called 4 5 1 Research back in 2000.

Um, and I was early on focused on, uh, machine learning and text analytics and all those things, and I've really just stayed focused on that. Um, and obviously the, you know, the, the whole, the whole kind of interest level in the, in this space has exploded since, since late 2022 when, uh, when Chat GBT was launched. So let's start off there.

Um, you know, you've been watching this, uh, for a long time, as have we, and I think that sometimes we as well, uh, get sort of pulled into all the news and the announcements and the hype and forget that, uh, a lot of this is still off in the future. What is your perspective on the timeline, especially around agentic ai? Yeah, we're, you're right.

We're, we're incredibly early, um, with AgTech when you kind of think we alluded to just at the top of the podcast, the, the, the kind of compression of time, um, that's gone from, you know, machine learning to yeah. Other kinds of predictive ai. Um, and then, you know, all we're coming up right up against the third anniversary of the launch of chat GBT in November of 2022.

Um, and then just, you know, so just less than three years later, we're also now trying to, um, ask enterprises to embrace Egen when they've only really, um, beginning to understand the, you know, how to operationalize, uh, generative AI in the form of, you know, chat bots and, and people writing prompts into them. And then all the kind of the interesting stuff and the scary stuff that, that, that ensued from that. So I think if you, you know, if you're kind of thinking about an S-curve, you know, we're on very much the very flat, flat bit of the bottom.

Um, and, but there's a kind of, there's always a pressure on an enterprises obviously, um, you know, used to be really exclusively the domain of, of the tech industry itself and financial services companies, um, that had, you know, larger software development teams. But really every company is, is embracing technology these days. So they're all under this kind of pressure.

There's, there's fomo, there's a fear, the fear of, of missing out, um, meanwhile on every day they've gotta run a business. And so this is, this is, this is what they're, they're up against. And there's obviously, um, you know, I think also the kind of the timeline between what seems, um, you know, magical to them becoming normal, to them becoming boring, um, you know, used to take decades, and now it sometimes feels that it takes dates.

Um, so, you know, new models coming out almost daily. Um, and then, then tools on top of those models. Uh, and so it's, it's incredibly difficult one to, to, to keep up.

That's why they engage, um, analysts like us, a, a, a rum, um, to he to help them, uh, you know, get, get a kind of a, the big picture, but also some, you know, some specific guidance on it. Yeah, no doubt that it's early stage. What do you see as a agentic AI applications that are being delivered today?

And again, we all understand it's early, but do you see kind of a trend or kind of applications that are making a breakthrough? I think it's, it's similar to, um, every kind of AI trend. We've, we've seen from back in the predictive days, you start with the, the things that are horizontal, so they're not vertical specific.

Um, and every company has some sort of customer service, um, challenge ahead of them. It doesn't matter whether they're, you know, B2C, B2B or, or any combination, um, thereof. So usually that is the, um, the first kind of opportunity.

So when it was, you know, when we're talking about predictive models, you know, talking about classification of tickets and things like that, now we've moved way beyond that. Um, and now the ability to quote understand natural language, um, is, is, you know, with, with generative AI has opened up a whole slew of opportunities, uh, for people to kind of lease semi-automate, um, customer service at scale and, um, at a scale that they never, they never could. So if they, if they have only a handful of customer service people, um, but they have, um, thousands of software agents, you know, there's, there's, there's, there's a clear opportunity there to be able to deal with people, enable them to interact with na in natural language, um, and then, you know, hopefully, you know, resolve their issues or if not, then escalate 'em to humans.

Um, and so we see, you know, we see, um, see a lot of, a lot of that. I guess the other things, um, we're working towards is, you know, we're working towards some sort of workflow automation, but that's, that is, that gets very, um, very specific to each company. And so that's, that's a, uh, yeah, that's, that's more, that's more challenging, I guess, some of the more novel use cases we've had since generative ai, you know, came along.

And let's be clear, obviously Agen doesn't work without genive AI is the ability to, um, analyze, uh, unstructured data at scale and then search for hidden patterns, turn those patterns into some sort of actionable insight. Um, and, and, and, you know, and do that, uh, over and over again, um, like having, you know, thousands of interns. And so I think, you know, then, you know, and then we've seen the kind of rise of, um, co-pilot, like things, whether it's the original ones like from Microsoft or, or other similar, um, tools that just sit alongside us at work and do, you know, very simple tasks such as, you know, suggesting times for meetings and, and things like that, and who might want to be in it.

And summarizing meetings, which has now almost become standard again, think about how quickly that's gone from, you know, that can't be done to more or less, every meeting is recorded and summarized. That's, you know, matter of, you know, a couple of years. And so, you know, all those, those kind of use cases there where we've got the, everybody, every company of any size has got that problem.

And what I think will happen is similar to what happened with predictive AI, is eventually it will get verticalized. So, you know, if you are, if you are a car, if you're a car maker or you're a bank, you have quite different problems down once you get down beyond those initial, um, horizontal use cases that everybody has. And that's because, uh, AI is dependent on data, and if you are the car company, your data set is completely different than if you are the, the financial services company, more or less.

Obviously there's, there's finance involved in cars and things like that, but you, you get what I mean. And so then the application, um, becomes, you know, vertical specific and then almost, you know, company specific. Um, but I think we're, we're, so we're very, very long way from that situation, um, with AgTech and yeah, we're, we're, we're, you know, really, really early and we've looked into, um, some really narrow domains of, uh, I say customer support, customer help.

com/help. And we've, we did actually a project where we looked at how many of those help services are agen, um, or how many of them are, you know, not agentic in the sense that they rely on humans prompting them to do things, um, all the way through. And it's amazing how, um, you know, the lack of, of, of true agents, autonomous agents, um, exist in those kind of environments.

And bear in mind that is, you know, the tech industry itself, and that is, um, the help within their own domain. So this is not trying to solve, um, a massive problem. This is trying to understand what's going on with their own applications and things like that.

So it's, it's not a way of not denigrating anybody for that situation, it's just that the, we are, you know, the, the hyper waste is obviously gonna be well ahead of the, uh, of, of the reality. And, um, you know, I say part of my job is to try and keep our, our feet on the ground while also seeing where we're gonna go in the next, um, you know, three to five years. I think it's really interesting.

Um, I love that phrase, predictive AI as a better way to phrase, um, what we've currently, what we currently have, uh, because in many ways that's really what, um, most of our LMS are used for, or at least what they're doing. They're predicting what, uh, you know, what the interaction should result in, whether that's generation of code or generation of, um, support answers, uh, as opposed to ag agentic, which theoretically would, um, include some sort of tool, use some sort of external references and calculation. Um, I know that you, uh, one of the things I'd love to hear from you about is non generative steps in AI tool chains.

com/help, uh, URL, uh, those are called agents. In fact, they're usually called agents. I interacted unsuccessfully with the United Airlines agent yesterday, um, using their terrible, uh, predictive ai.

And, um, you know, I wonder if we have sort of a semantic challenge here. Is that part of your role as well, trying to help, uh, clarify what we mean by all these terms? Yeah, it's definitely in part, I mean, I think the, what we will get is a, um, there will become, there will be more clarification with the reason, I guess, obviously these things were called agents is, is, goes back to that, um, that customer service focus.

And as you say, you, you know, your kind of, your kind of experience is, is not, um, atypical. And so I think the gradually over time when we get to the point where, um, you know, the software is working behind the scenes, so the agent, the Genix software is, is, is, you know, creating, taking action, um, without a human, um, necessarily knowing or, or a human having to any have, have an interaction with it, that's when, you know, the, maybe the agent word might be more appropriate. I think it's, you know, at the moment, an an agent obviously comes from a customer service, a person, um, you know, that, that, that nomenclature where that originates.

But what if it's, um, executing, you know, 1500 workflow steps, um, without you knowing and, and something gets done, uh, in the background, then I think it's, uh, yeah, that's the kind of goal we're trying to get to. Um, and also when you mentioned on some of the, um, the kind of, uh, the generative AI and the, you know, the, maybe the, yeah, the probabilistic and the deterministic aspect, you didn't use those words, but the, but that, that kind of difference. I think what one thing I'd just like to point out that we we're looking at at the moment is, um, because of the early generative ai, um, use cases were humans typing things in, um, and getting results back, um, you know, those, those kind of, um, and that's probabilistic, that's losing a large language model, which has a model, you know, of, of, you know, scraped from, from the web.

Um, and that's really useful for doing creative tasks and creativity doesn't have to be, you know, literally artistic, but you know, that obviously it's, it's great in those situations, but creative obviously suggesting, you know, ideation, you know, you know, give me some ideas of what I should be talking writing about here, or, you know, we've got a meeting about this, you know, can we, can you produce an agenda on all that's all very creative stuff. Um, and, you know, they're pretty good at that. Um, and obviously there's a load of, you know, there's hallucinations.

Um, we know about quite a lot of them, and this is some things that are just a complaint incorrect and you have to work with them, but that's, that's great. Um, but there's also a need for, um, agen automation of deterministic processes. So the classic example is, you know, payroll runs on the 15th of the month.

I don't want a creative suggestion that says, why don't you delay that till 19th? Um, and then, and then that to cause an action that delays everybody getting paid. That's not, that's not to useful at all.

Um, and so those kind of things, there's an, there's an element there of, um, um, agent automation. I think, you know, this is, this is what you're trying to get to, this is what the agents are gonna be doing. They're gonna be automating these processes.

Um, and on the side there's this, there's this fantastic generative, um, aspect of it where humans can, can interact with natural language, but we kind of need to, organizations need to understand that there's a place for, for, for some of that. And there's a place for the straight up determin deterministic automation. Because if you think back, um, you know, I always like to think the history of the software industry is a history of automating repetitive human processes, starting back in the, you know, the fifties with mainframes in accounting, in finance, um, yeah.

And they were literally number crunching. Um, and then we've moved along all the way through, and we've always been focused on, um, usually been focused on structured data in relational databases and then data warehouses and then data lakes and, and things like that. Um, and we built up all these kind of software tools on top, um, analytics tools, business intelligence, things like all those kind of things.

And then huge application suites. And there were basically following rules and executing, um, you know, processes, but those rules had to be written by humans. Uh, they had to be overseen by humans and so on and so forth.

What we're gonna move towards is, is, um, you know, the more agentic, uh, future where, you know, there is, there are a set some, some rules and then some, you know, completely probabilistic situations. But the software is, um, in some cases managing it itself and executing on, on our behalf. And I think one thing, one kind of rule of thought heuristic, I guess for organizations to think about is, is if your problem involves a load of structured data, um, such as your customer records, your employee records and things like that, and that's where the automation is coming from, then that's probably gonna end up in a fair amount of deterministic processes.

If you are, um, if you are, you, the problem you're trying to solve involves a load of unstructured data. So we've got thousands of PDFs and we're trying to extract tables from them, and then turn that table into something useful, um, that we can then, then use, then you're gonna end up with more probabilistic kind of, um, challenges. Um, so it's, it's, it's just a way of, um, of framing things.

But I think we're only in the ag agent space, um, in terms of, you know, the software that's out there. Um, yeah, I think we're in only really just starting to think about that. I know this sounds silly, but you know, here we are in, in November, uh, yeah, we're only starting to think about that in the last few weeks.

So this stuff is, um, yeah, moving so incredibly quickly. I go to a lot of, um, technology vendor conferences, um, you know, and this time, you know, this week is another one that's Microsoft, um, ignite. Um, but I've been to, been to many others, um, this year, and we'll do again, you know, next year.

And I've been doing for obviously for a long time, the, so you see kind of, you see these kind of trends, but you know, these days it's stuff is moving almost daily. And that's, uh, that's very hard for, um, for, for organizations to keep up with. Um, it's also hard for analysts to do, but it is our, yeah, it is our sole focus, so at least we don't have an excuse of having to do, uh, a whole bunch of other things.

Yeah. It's, it's difficult enough to deal with and learn the new techno, new terminology, let alone learning the new technology. Uh, I mean, there's no doubt that agent AI can help with automation.

Um, the, the problem I think is, is that the technology behind AI is becoming so complex as time goes by, that more and more people use ai, but less and less people understand the technology behind ai. And in some cases, I, I, I believe people use AI and believe that it's more trustful than a human. Do we have a trust problem With agen ai?

Oh, yeah. I mean, I, I think there will be, yeah. I mean, because obviously it's ca it's so much more powerful, um, because if you had to rely on a human writing prompts in to get things done all the time, that's, you know, useful.

And it's, you know, up to a point. But if, if you, you know, if you get to the point where software is executing, you know, you know, um, software, then, you know, you can see how that could scale very quickly and become incredibly powerful and potentially, you know, dangerous. Um, so yes, there's definitely a, a, a trust, um, problem to be solved.

Uh, I think we are still not really, um, thinking about that at a, at a deep level because we, a lot of, a lot of, um, pilots that are in enterprises now are just, are just very much that they're pilots within sandboxes. They're not really dealing, um, with anything of, of, of enormous scale, um, that, where it could cause problems. But, um, I think, yeah, I think there's definitely, there's definitely a, a trust issue.

There's an, there's an old joke, um, such as there are jokes in, in AI that, um, you know, AI is anything that doesn't work yet. In other words, um, this is impossible. Um, I'll use some ai, and then once it works with ai, people go, that's not ai, that's just, that's just the way these things work.

I say, well, it is still ai, it's just is solved the problem and it's moved onto to another one. Um, so I think we're gonna have, um, we're gonna have a similar, um, set of issues with, with, with, uh, with AgTech, um, that the, the, that, that we had, uh, with, with kind of traditional machine learning. Well, you know, your, your point there about software, executing software I think is an interesting one because you're right that that's where the, the trust factor is needed most.

But it's not just executing software, it's software, writing software, and then executing that software and acting autonomously. And I think that that's really where, um, not just the, the, the risk and the, the threat, the trust comes in, but also where the promise comes in. I mean, if you, if you look at what these companies that are developing this technology are saying, they're saying that that's basically the promised land.

So, you know, AI will become truly ai. In fact, I've hear, I've heard a big backlash against that whole phrase, AI people don't want to use artificial intelligence to describe anything that's not, uh, verifiably and independently intelligent. Um, and they're saying basically that, that we will get there and we'll know we've got there when it is truly autonomous, when it is, um, taking action on its own when it is, um, creating its own motivations when it's writing its own software, when it's executing things completely on its own, um, that's pretty, uh, pretty concerning when it is such a black box, as you also pointed out.

I mean, you know, people don't understand how it works. Even people very close to it don't understand how it works and people seem to be enamored of it and already taking it as intelligent when it's, we really haven't gotten to that point. Now.

Um, what's the prognosis here, uh, for when this will be truly intelligent? If you mean artificial general intelligence, um, you know, the ability to do everything a human can do. I Don't wanna necessarily push you into that corner, but I, I, you know, when it's truly able to, uh, to act on its own.

Well, It, it depends what it is, doesn't it? It depends on the domain and depends on the problem you're trying to solve. Um, you know, you're talking about code generation there.

I mean, that's obviously been, you know, probably the biggest, um, apart from the kind of custom service stuff, it's had the biggest effect, um, on, um, organization's ability to, you know, to to, to automate something, um, in the last couple of years. And that's, that's, that's taken off hugely. Um, and that's, that's, I think, you know, it's, it might be challenging if you are an entry level, um, um, you, you know, a graduate that's just graduated, um, with a computer science degree looking to, for a coding job, I'm sure, sure, that is definitely an issue.

Um, but when you think of all the legacy code for the people who no longer with us who wrote all the COBOL and the, and the lisp and, and all these other kind of languages that, uh, um, are relatively important, but, but, uh, aging, you know, there's enormous opportunity there to, um, automate the, you know, the, the maintenance and, um, regeneration of that code. But I think the, I, I don't, I don't really, um, I must admit, I, on the kind of AI safety spectrum, I'm, I'm not particularly that concerned, um, that we are gonna head to some sort of, um, a GI oblivion, um, anytime, anytime soon. I kind of think the, some of the people that, um, pushed that, um, you know, are doing that for a reason, and that could be a kind of, uh, you know, um, can't think of a way to put it politely.

Um, but there's, there's, there's, you know, there's a reason why people might wanna say, you know, I told you so, uh, if something bad happens, but there's so many things that have gotta happen. Um, you know, and, you know, in order for, for, for software to have, you know, major real world effects, obviously, um, it does. And, you know, our, our airplanes use software and our trains do and, and our cars increasingly do, obviously.

Um, but I think, you know, I think, uh, I think it's, it's so we're not just, I don't believe we're one, you know, one model away from, um, Armageddon, any, any one point. I think there's so many controls that, that, that will be in place. The fact that some of the people building the models don't fully understand how they work is, is, is real.

Um, and that's genuine, and I think that's, um, that is a challenge, but they're working on it. And, um, you know, I think it's, it's, it's one of those things that once this space matures, as it as it does in all forms of software, you will have governance tools and trust, um, tools in, in place. And I think you, you, you have to have, I mean, there's, that's gonna be, that's gonna be a major opportunity for the software companies to build those things.

Um, but it's obviously a challenge for the enterprises, um, that wanna buy them and, and, and use ag, um, ag software. But I think, you know, I think we're a, um, you know, there's definitely a, there's definitely a kind of, there's a platform shift happening, and when platform shifts like this happen, um, you get a, you get a whole load of, um, some things become kind of features, you know, within, you know, the incumbent set of software and some things turn into companies. And I think one, the, um, the really, yeah, from my excuse, my, my point of view as an analyst is what's happened in the last five years has been incredible.

Well, you've actually now got pure play AI companies of, of massive scale like Open AI and anthropic, um, and, and the others, whereas you never had that before. You always had, um, you know, the same, more or less the same names, um, adding the features to what they had already. And I think that's, that's where it life is gonna get quite interesting, um, for everybody, both, both in terms of the vendors themselves, the investors in this, in this industry, but obviously mainly for the, for the enterprises.

And we'll be looking very closely at how that shakes out. So in other words, how, you know, how do you, how do the, the software as a service vendors, um, that are used to selling package applications and sell licenses, uh, on a subscription basis and things like that, um, to, to, on a number of seats used, um, that's gonna, we were already seeing, um, the overhaul of the pricing, for instance, of software, um, by, by, by, um, by the, by the promise of AgTech that is, that is already happening. Um, and we're now seeing flexibility being offered by, by the software companies that they always resisted doing, um, before, so, so flexible pricing models and, and, and things like that.

So there is a, um, that's, that's the kind of, they're looking, they're already looking at what they can think might happen in two, three years time, um, and, and having to adjust. So I think there's, there's definitely always, um, always a, a need for, um, people to be, you know, somewhat cautious as to, as to what they're doing. Um, but, um, you know, when you, when you think of some of the, you know, the kind of customer service issues, um, that companies have when they're using agent, um, ai, and that is, you know, they're not particularly, um, you know, that's, you know, people are usually not gonna get hurt as a result of those things.

Um, but obviously when we come to much more critical domains, um, such as transportation or, or weaponry and things like that, then, you know, that's where you have to have, um, some sort of, you know, governance structure in place. I don't mean just software, I mean, you know, legislation. And I think that's, you know, it's already happened in some parts of the world.

Um, sometimes you could see it's a bit too top down and a bit a bit crude. Um, but it will happen. Um, AI will be, um, a regulated industry.

Um, there's no two ways about that, um, because it is, it is very powerful and, uh, I think that's, that will, that will happen. So, um, you know, it's as much to do with, uh, trusting, you know, voting in the right people, I guess, to get legislation right. As, as, as, as anything else.

Yeah, it's, dealing with software is not easy. It's very difficult to, to, to validate software. I mean, at some point, coding was, or, or, or, or writing software, you needed a lot of knowledge from a hardware and a software perspective and expertise in order to build clean and useful applications that people understood.

Nowadays the software is being generated by software. I, I, you know, I have to ask, you know, what's, what's worse, you know, an agent, ai agent writing software and running it or a human vibe coding generating code and deploying it? Well, it's worse.

Um, I, I don't, I don't think one's worse than the other, the, the, the, I guess they're different. I mean, the vibe coding thing is interesting because, you know, it's opened up, um, you know, software development to so many different, you know, people with completely different skill sets. Um, you know, and then, you know, obviously agent, you know, um, generative AI writing code is obviously, you know, doing, doing something similar.

Um, you know, it's gonna affect people whose job is software development. There's absolutely no two ways about that. Um, and it already already has done.

Um, but I think there's, um, you know, there'll, there'll be so much, so much code needs to be written, um, because it's increasingly complex world and we can't manage everything, um, manually, that you're gonna need software to manage things that software currently doesn't manage. Um, so I think, uh, you know, I think you are gonna need both the ability for, um, um, you know, software to, to write its own code, but also the vibe coding stuff is, is interesting because obviously the, the potential for that is you are getting the domain experts directly into the process. Now, we've always had, as, as you know, in software development teams, there's always been this kind of challenge to get the, you know, the line of business people involved at the right time.

Um, that's why we went from, you know, wallfall to Agile, um, and, and all that kind of stuff. Um, and you know, that if, but if you had the actual domain expert being able to write their own little apps, um, then, you know, within a reasonable framework, a software development lifecycle framework that has, um, you know, testing QA and, and governance in place, um, then I think that's quite an exciting notion. Um, I think there's a lot of people who, um, would like to be able to, um, you know, build their own apps, um, to, to some extent.

Um, yeah, not everybody. Um, but, uh, so I think it's, um, yeah, it's an interesting, it's an interesting, you know, development. Another, another one that's, um, that's, that's very recent.

Uh, before we go, one more thing I wanted to hit on something you brought up right at the very beginning was the fact that, um, the, the, the world of ag agentic ai, uh, and processes and tools and so on, will not just be chatbots talking to chat bots, that we will be looking at additional types of tools in these tool chains. Some of them may be deterministic, conventional software platforms. Some of them may be, um, you know, data platforms and, and, and different ways of, of querying, um, structured and unstructured and, and even multimodal data.

Uh, others may be, you know, generative AI processes. Is, do you see an an entirely new, um, type of software industry emerging here to support ag agent tool chains? I don't think an entirely new industry, no.

Um, I, I suspect we're gonna get, um, um, you know, some startups that are, that are, you know, covering part of the process. Um, you know, part of the kind of, you know, the software development process, the governance process, um, the kind of agentic ops process. So we had it before, when you kind of go back to, um, there's a kind of category called application performance management that cropped up and this's got nothing to do with AI at all, really.

I mean, this is just how managed our applications are managed. Um, and you know, that that cropped up. And then you have, um, you know, when, when predictive, um, you know, when machine classic machine learning was started to come around, you had ML ops, machine learning, operat, operationalization, um, and then you, you know, then you, you have slight slightly different problems.

So when you've got a model, um, if you had a, a rules-based model that only does the thing that is it's programmed to do, that's fine. And obviously if, if, if the thing falls over, you can restart it and stuff like that. What the fundamental difference between, um, that kind of software and AI is obviously the model adapts, it learns, it decays, it drifts, it does all these things.

It's almost, you know, it's not, but it's almost organic in nature. And so that is where you have, um, the, the, you know, the challenge around the ML lops is kind of supposed to, um, create, and then we had a load of specialist vendors, um, and a lot of 'em are still around that, that cropped up to do that. I think we're gonna get the same thing, um, with ag ops, if that's, if that's what the phrase is gonna be.

Um, and, um, but some of those will get bought, some of those will survive, and some of those will go by the wayside. But you are also gonna get them all the application vendors of any scale. So, you know, Salesforce, Oracle, Microsoft, Workday, service Now, all these companies are obviously building out their own, um, tools, platforms, applications, uh, they all want to be, everybody wants to be the platform, but not everybody can be.

Um, and then you've got obviously the hyperscalers doing their thing and then sorts of other companies, um, offering, you know, ag agentic, um, tools. We are seeing a little bit of this bifurcation between, um, the companies that are going after the, as we said earlier, the creative, the probabilistic creative opportunities. So those aiming at marketing departments, um, have had quite a lot of recent of, of, of strong traction because they're solving a problem that really could not be solved until generative AI came to LLMs came around, it just didn't, it was just basically impossible.

Um, and then those that are dealing with more deterministic things sitting on top of relational databases, so C-R-M-E-R-P and all those kind of things. So we're seeing a little bit bifurcation there, but you're gonna get some of, some winners. There are that I suspect, um, that, um, that will, that will either survive or, or, or get or have a good exit, um, the financial exit or something like that.

So I don't, I don't think, um, but then again, as I mentioned earlier, this is the first time where we've had pure play AI companies of any scale and open AI is certainly off scale and it has ambitions from the device to applications and everything in between. Um, obviously it's known as a model provider, but you know, it's obviously trying to build chips, it's trying to build devices, it's, you know, however, how well it gets on with that, we dunno. Um, and you know, they're obviously now influencing, you know, where data centers get built, um, how much energy is used.

Um, this is, this is an incredible, um, you know, development in the space of a, of a decade from scratch. And so you are gonna get those kind of companies. I wouldn't, you know, you wouldn't necess say that's, that's peculiar to agentic, but that's peculiar, but it is peculiar to generative ai, which on Agentic is based on.

And so I think it's gonna be, um, really interesting to see how, you know, that company and one or two others of massive scale, um, uh, you know, influenced the way, the way the rest of the, um, the rest of the software industry goes. And I think, yeah, we're gonna have standards, we've got MCP servers and we've got, um, a two A protocols, and there's gonna be more. There has to be more.

Um, then it's the question of who becomes, you know, who's the platform, who's enabling layer within it that make it work properly, properly who are just trying to sell apps and they'll use anybody's a gen platform and yeah, and tools, vendors, um, and stuff like that. And then obviously the chips underneath. And so, you know, companies are always looking for companies in the sense of enterprises that buy this stuff are always looking for options and looking for diversity.

Um, and, you know, whether that be down the silicon level where at the moment there isn't much diversity, um, or right up the application level, although there obviously is. So I think it's gonna be, um, it's gonna be a fascinating, um, few years, uh, in the, uh, the agent AI space. Yeah, absolutely.

And, and I feel like, uh, like you said, that these, um, massive AI companies along with a lot of the traditional vendors, you know, companies like Oracle, um, Amazon, Google are angling to build an AI platform really. And, um, so we'll be definitely watching that. Um, we do have to wrap, uh, unfortunately, I think we could talk to you for, um, many, many hours.

Um, but unfortunately, uh, the, the timeframe for this episode is done. So thank you so much for joining us. Um, before we go, if everybody else wants to continue speaking with you, uh, where can they find you and where can they find your, uh, coverage?

com. Um, and you can also find me on, um, on Twitter X at, at nick patients and, um, and on LinkedIn. And yeah, I'd love to, uh, love to hear from anybody.

Absolutely. And, um, of course, uh, we will be continuing on as well with, uh, utilizing AI podcast series for futureum Group. Uh, so folks should look for that in their favorite podcast applications, uh, Frederick as well.

Uh, where can we continue this conversation? com. And as for me, uh, you'll find me, as I said on utilizing AI on the Techron gang on, uh, many other, uh, platforms, uh, here within Futurum Group, along with on social media as as FoST.

Thank you for listening to this episode of Utilizing Tech. Uh, you'll find this podcast in your favorite podcast application as well as on YouTube if you wanna see what we look like. If you enjoyed this discussion, please do leave us a comment, a rating, a review.

We'd love to hear from you. The podcast was brought to you by Tech Field Day, which is part of the Futurum Group. com, or find us on X Twitter, blue sky, and mastodon at utilizing Tech.

Thanks for listening, and we will catch you next week.