Utilizing Tech - Season 7: AI Data Infrastructure Presented by Solidigm - 09x03: Bringing Agentic AI Applications to Market with Brad Shimmin of The Futurum Group

Starting point is 00:00:00 From the very first episode of this podcast back in 2020, we've been focused on practical applications for AI technology. And we're starting to see these come to market with agentic tools. This episode of Utilizing Tech features Brad Schemin, VP, and Practice Lead for data and analytics at the Futurum Group discussing the ways AI is gaining autonomy. Welcome to Utilizing Tech, the podcast about emerging technology from Tech Field Day, part of the future in group. This brand new season focuses on practical applications for gentic AI and other related innovations in artificial intelligence. I'm your host

Starting point is 00:00:36 Stephen Foskett, organizer of the Tech Field Day events series, including AI Field Day. And joining me this week as my co-host is Mr. Frederick Van Herron. Frederick, welcome to the show. Well, thanks for having me again. So I'm Frederick Van Herron, the founder of High Fence, an HBC and AI Consulting and Services Company. You can You can find me on LinkedIn as Frederick V. Heron or on our website, highfence.com. And if you've been listening to utilizing tech or the previous utilizing AI seasons, you definitely recognize Frederick. You know, he and I have been talking about AI since well before all this generative AI and chat GPT hit the, hit the market. And one of the things, you know,

Starting point is 00:01:23 that I want to call attention to is the reason we called this utilizing AI way back in, I don't even remember what year that was, was because I was very interested in practical applications. How do we utilize this? How do we make this technology productive and useful in the enterprise? And Frederick, you've been working on that way longer than I have. Yeah, indeed. I mean, if we had a crystal ball, it would be a lot easier. I mean, I think the holy grail here is to have the machines do a little of the lifting for us, you know, like the mundane items.

Starting point is 00:02:00 And speech is the way we communicate with machines, right? And so I've seen the whole evolution going from CPU-centric to data-centric, and nowadays it has gone so far and so fast that Agentic AI is opening a door to new applications. And that's what we're really hoping for. In fact, at Futrum Group, one of the big focuses of the company is figuring out what's practical and what makes sense and what really has legs. And that's why I wanted to invite on the VP of the data and analytics team at Futurum Group to talk about some of those practical applications, some of the ways in which we're seeing AI coming to the enterprise. So let me introduce Brad Schemin, our guest this week. Brad, welcome to the show. Yeah, thank you, Stephen. And it's great to be on in the show. I appreciate it. So as Stephen mentioned, I'm VP and practice lead for our concern that focuses on data intelligence, analytics, and infrastructure.

Starting point is 00:03:07 So basically, everything that goes into building insight and gaining insight and taking action on insight within the enterprise. I've been an industry analyst for quite some time, but I've been a technology practitioner for far longer, and as you can tell for quite some time, going back to 1990 when I first started working with Fox Pro databases and Novell Network, if that gives you any insight into the depths of my suffering that I'm willing to endure with technology, because I adore it so much. But at any rate, I'm very glad to be here today to talk to you guys, both as an industry analyst watching this market and as a practitioner that is building agentic solutions within Future.

Starting point is 00:03:57 Yeah, so it's an interesting conversation. I mean, can you talk a little bit what is an agentic system for our audience? I love that because I recall it was about a year and a half ago. It was at a conference, and the vendor will shall rename nameless, but they like the color red. And on one of their slides was, these are the agenic processes that we support as a company, and we have built for you, our buyers. And it was a massive list, line by line by line. And when I looked at it closely, I noted that pretty much 95% of those were all a single transaction, like, you know, open the fridge door. check the weather, things like that. And I don't believe that's agenic.

Starting point is 00:04:45 I think that's transactional. That is something that anyone who's built software or works with software knows is you ask for something. It gives you something. And so when I think about agentic systems, I, as an analyst, define them as something that has a number of capacities and characteristics. And those are autonomy, first and foremost, the ability to act on its own without me saying, now shut the fridge door. The second would be the ability to reason and plan, which leads to said autonomy. So to be able to say, okay, the user has asked me to do something for them.

Starting point is 00:05:33 Well, what does that entail? What will I need to know and what will I need to do to achieve that? Make that plan to disambiguate sometimes what the user is actually asking for, turn that into some sort of actionable plan and then make it happen. And that comes to the third aspect, which is the ability to make use of tools and information to seek action on its own. And that is where I think there's been a lot of leeway made. I should say across all three, in terms of for the first, models have gotten much better at reasoning.

Starting point is 00:06:13 And as we see, many models now are just built in with inbuilt reasoning capabilities where they will go into think mode. The second models will be built with the ability to basically make a plan and to think about how they could execute it. Third, they will be able to make use of tools and information. And that last one is where we start to see all the technologies like MCP that I just knew we'd talk about today come into play and how popular that is right now in supporting agenetic solutions. But to summarize very quickly about that, you know, I see an agentic process is anything that a machine can do to basically, as Frederick mentioned earlier, to do.

Starting point is 00:07:01 take some of that lifting off the shoulders of a human to do that autonomously and to make that something that wasn't automatable, automated. Whether that is basically getting the weather and then booking a different seat for a stadium, let's say if it's going to rain for you or if it's to basically to put a hold on a stock that you know is going to respond to something happening in the market, doesn't matter. That's all, you know, just a matter of scale and complexity. But at the end of the day, it's just autonomous action taken by AI on our behalf. Right. It seems like agentic AI is kind of an evolution of generative AI. Now, from a practical standpoint, I mean, can you buy a genetic AI system? You know, how does that work? I mean, you talked

Starting point is 00:07:57 a little bit about MCP is how does MCP kind of is, how does it play an important role for people to build applications? Yeah. So to answer the first part of your question, yes, you can. We're seeing increasingly productized agentic solutions. And this is, you know, how the market evolves. It always starts with horizontal use cases that, you know, basically you have a set of tools. Like if you're a developer, you might have frameworks and libraries that would help you get to the end of that, you know, eugenic process. So over the last couple of years, I would have likely used Langecane and within that Lange graph to spec out how I wanted my agentic system to work and code that to work. But as time goes on, and as the marketplace always does, it leans toward building out

Starting point is 00:08:48 tools that and solutions, I should say, not just tools and not just resources. So that I I can basically as a consumer, whether I'm a consumer consumer or a business consumer, you know, turn on or open up a browser, let's say, and have at my disposal a complete agentic solution to do something, whatever that is. And right now, I think the market is predominantly delivering what I would say is reasonably consumable agentic processes, not so much in the specific get something done for everybody who's trying to book a dentist appointment, let's say,

Starting point is 00:09:33 but instead about how they might do common tasks. So if you go look at the vendors that I cover and look at quite a bit, you will see those that are focusing on And horizontal use cases like data integration, they are right now building out agentic solutions that are productized that go toward helping you, the data professional, basically stand up or find and then bring in data in a way that you can use for whatever use case you want. So if that means like authenticating to get the data, cleaning that data, making sure that it's not replicated with something else, making sure. making sure that it's harmonized, et cetera, and then standing it up for you to use. Or if you're a business user

Starting point is 00:10:20 and you're trying to answer a simple question, like, you know, what is the close going to be for sales this quarter in Chicago? Well, an agentic process can be built and is being built by a lot of these vendors that will walk you through that little basically without you having to write code or even build anything with a wizzy wig or drag-and-drop,

Starting point is 00:10:43 Just set you up to do that. And what's making that possible is to your second part of your question, this introduction of several protocols and tools that enable agenetic AI and this model context protocol with which Anthropic, a frontier model maker rolled out about a year and a half ago, is part and parcel to that or key to that because what it does is creates. a sort of lingua franca for how an agentic-based or agentic system that utilizes large language models, which is predominantly what we associate with agenic systems, allows a large language model to basically find out what is sitting behind this MCP server and what can I do with it.

Starting point is 00:11:37 Is it an MCP server that exposes capabilities like what can GitHub, for instance, do for me? What do I have access to? What can I see and do there? Or is it just a data source itself? Like if back to what I was talking about with having an agentic solution basically stand up a sort of what happened at the end of the quarter, what is going to happen at the end of the quarter, that can be an MCP experience, if you will, for an agentic solution. So the LLM basically works with that data through an MCP server. So it's becoming increasingly productized, increasingly abstracted away, which we all know in this industry is there are only two ways forward. One is, you know, if you want more performance, you basically cash everything. If you want more simplicity, you create another layer of abstraction until you basically don't have to worry about the performance or the complexity of what's underneath you. So, Brad, I'm trying to get my head around the market a little bit here, and maybe you can help me with that.

Starting point is 00:12:48 It seems like there's a bunch of different solutions that could all be labeled as agentic AI solutions. I heard you mention here just now, basically tools that are used for categorizing and harmonizing and massaging data for data professionals. I heard you talk about tools that would be used as part of an overall enterprise application stack. And I heard you talk about as well tools that serve the, really the needs of end users and business people, you know, essentially answer my question. You know, for example, you're mentioned there of GitHub. Another one that I know a lot of people are using is being able to query financial market data, you know, just public financial market data. data, being able to query the weather, you know, being able to query all sorts of data sources like that, even down to, you know, the sort of things that people use assistants like, you know,

Starting point is 00:13:54 the S word or the A word from, you know, or the, you know, the Google assistant, that kind of thing. And all of these, to me, they seem like they're agentic solutions, but they're all very, very different. You know, how would you break up the market? How would you categorize the world of applications in terms of what buckets would you put things in? Yeah. It's becoming much more complicated.

Starting point is 00:14:22 And I think that's fine, honestly, because if I'm a vendor and I'm serving a constituency in the enterprise, let's say, or I'm a vendor serving a consumer constituency. And each of those have, you know, tasks that they're trying to. to do. So back to the, you know, buying a ticket for a concert on a rainy day. You know, if I'm a consumer and I log into Ticketmaster and I say I want to buy a ticket for a concert and I've got three nights available to me and I want to only pay this much and I don't want an occluded view and I want to make sure it's on a night that has the best chance of not raining, let's say. That would be an agentic process that I would expect Ticketmaster.

Starting point is 00:15:08 to build for me such that my experience with Ticketmaster would not be, I wouldn't be opening up another agentic process. I would basically just be logging in to the Ticketmaster interface and saying, I want a ticket. And I might do that either by typing it out, but increasingly, I would just have my phone in my hand and I would be talking to my phone. Like I'm talking to you guys right now. And I just as I just said, I want a ticket that's on a night that's not going to rain and I want to have a good seat and I don't want to pay more than X. Those are the things that, you know, you as as a consumer would do, you know, on the phone, let's say, or in person, you know, in the last many decades to get something done. And agentics software is, I think, you know, the best route that we have

Starting point is 00:15:59 forward right now to at least approximate some of that. And the reason why it works is because the nature of agentic software, as I mentioned at the outset, something that has autonomy, something that can plan, something that can interact with and make use of tools and information, what that gives you is flexibility and the ability, most importantly, to respond to changes and unanticipated changes in situations. So if I was using like a system that had a pull-down menu that said, you know, what night would you like? How much are you willing to pay? And I hit the button to go. If something happens within that, that, you know, system, that workflow, let's say, that would basically kick that out and say that's not going to happen. Sorry, I would have to

Starting point is 00:16:53 start over with an agenic process and with, you know, the tools that we have at our disposal for asynchronous computing that we use in the consumer software space right now. So predominantly, it doesn't matter. This system could basically just sit there and wait for the tickets to open up that I want and then make the transaction for me. And the tool sets are becoming such that I would expect every vendor, whether they're selling to consumers or to the business, to then sell to consumers that would be building agentic processes,

Starting point is 00:17:29 into any use case and any workflow in which that sort of flexibility and adaptability to what would normally be a complex, hard-coded sort of problem, I think is going to be turned into agentic software and productized as such, even though to me the consumer, I might not ever know that that's really what's going on. So I'm sorry that's a bit of a long answer to your question, Stephen, It really, to me, says that we're going to see a market that looks like this, how I would describe it. You're going to have the underlying tools. So in the scenario we just laid out, you would have, as we've been talking about, our wonderful MCP protocol to allow the models to understand what tickets are available.

Starting point is 00:18:24 I would also have a what's an A2A, which is another protocol that the Google developed that works with MCP quite nicely to allow disparate agents. So the weather service might have its own agentic system with its own MCP servers that would deliver weather information. And Ticketmaster would have its own MCP servers and they would use A2A to a, and they would use A2A to. talk to one another so that the models can basically say, so what's the weather going to be? Is it changed? What's it look like now? And so you'll have these tools,

Starting point is 00:19:02 these underlying technologies and tools. You'll also have the model makers and providers, which are increasingly building more of a platform than just a model. So it used to be what we cared about were, what can a model do for me when it's responding to a query, but increasingly what we as consumers are paying for, are the attentive services that go along with that model.

Starting point is 00:19:28 So models look a lot more like a platform. So if you look at Anthropic, you look at Google is a great example with Gemini. You look at Mistral. All of these frontier model makers are building a very rich ecosystem of APIs, of supportive services for developers of software. So if I'm Ticketmaster, I'm going to probably take advantage of these growing platforms to speed my time to market in building an agentic system. And strangely, likewise, if I'm a consumer, and I'm just trying to do something for myself like doing some research on, you know, what's the best night to go to a concert in my area this year and who's playing, I could use the same tool set. The same tool set that, you know, Ticketmaster's using to build this, you know,

Starting point is 00:20:26 scalable, highly scalable solution, I would, as a user, as a consumer, use that the same way. And it wouldn't look any different to the back end, but it would look different to me because as I build it out. So, Brad, it's a lot of information you provided. So who are the companies we should keep an eye on around Agentic AI? Yeah. As I was mentioning a minute ago about, you know, the marketplace itself and how complex it is and how almost every, you know, company you interact with is going to be building and using agentic software. The same goes for those you might buy agentic technology from. So if I'm, you know, an AI practitioner and I'm building out AI solutions and I'm using a data robot or a data coup or, you know, any kind of, you know, AI platform like that, I'm going to have agentic tooling coming from those guys. They're building it right now into everything they have. If I am consuming models from the frontier model makers

Starting point is 00:21:35 via OpenAI from Microsoft on Azure, if I'm using Gemini from Google, or Cohere from Oracle and OCI or any of their own models. And the same goes, for example, with IBM and their granite family of models on top of the IBM Watson X platform, they're all building agentic tooling. They're all building agentic use cases. They're all, they're horizontal use of cases. And they're also working toward building actual productized solutions as we touched on briefly. So all of those folks are the ones, I guess,

Starting point is 00:22:14 I would watch out for it first because the ones who are making the underlying infrastructure that lets me run AI, they matter in this. The manufacturers of the models themselves really matter in this. And as I mentioned a minute ago, those models look more and more like platforms themselves than just a model that you download the weights for and run on your local machine. So I would say that to start with those, so start with the AI platform vendors, start with the model makers, and then branch out from there, depending upon what market you're in. If, for instance, you're doing sales enablement, management, you know, ERP, etc., obviously you can look to Salesforce with what they're building on top of data cloud and Einstein.

Starting point is 00:23:05 And you can, if you're a customer of SAP, you can look to what they're doing with Jewel on top of their business technology platform. And if you are a customer of Oracle, you can see what they're doing with their own stack of line of business software. All of that is, you know, seeing agentic processes bubble up through those software. And it's actually getting such that, and this is something I've had a little bit of hard time with because I think everyone who listens to this podcast has heard Satya Nadella from Microsoft mentioned through. three or four weeks ago, that he felt that software was itself going to collapse and that we would no longer, or soon, no longer, like, want to log into Microsoft Excel to use that, but instead might have a natural language interface that would be to an agentic process, which would see Excel as a tool to use, to get me the consumer what I wanted, instead of me having to open up a spreadsheet, type, put stuff in columns, et cetera. would basically, you know, take my question, disambiguate, turn it into a plan, and go get the data and use Excel to do whatever calculations it might need to give me what I want. So you're going to get it from whatever vendor that you interact with. So if you're an office user,

Starting point is 00:24:31 as I just mentioned, if you are a Google workplace user, you're going to get it from them. If you're a Salesforce user, you're going to get it from them. It's everywhere, in other words. Yeah, that was what I was going to say is, you know, Microsoft. I mean, you know, they're a primary provider of these tools for a lot of us. And I'm sure that many of us, you know, you mentioned Salesforce, Oracle. There are a lot of companies out there that are really trying to be the business CRM-Agentic provider. You know, how does somebody know who's got the best vision and who they should be talking to? Oh, us.

Starting point is 00:25:20 I think they should talk to us. Sorry, that wasn't supposed to be an ad. But, you know, I mean, basically, like, you know, if you're an end user and you've got, you know, an Office 365 subscription and a Google, you know, workspace and a sales force and so on. I mean, everybody's trying to show you their vision. who's got the good vision? Yeah, I think those that understand AI from a very pragmatic perspective instead of just trying to chase, you know, benchmarks and, you know, being looking popular and focusing on how cool the videos are you can create,

Starting point is 00:26:01 I think those that instead understand the necessities of performance, cost, security, and governability and transparency. and accountability, most importantly, that's the vendor you want to go with. So any vendor that emphasizes those aspects, what I call responsible AI, which is something we seem to have forgotten a little bit over the last year or so,

Starting point is 00:26:25 but before that was very important in the enterprise. But anyway, that's how I would separate the wheat from the chaff, honestly. And as I was mentioning, you should come to us because we're concerned, And consumers and users of these agentic solutions and building out what we call living comparative reviews of these spaces. And one of them just happens to be agentic platforms for sales and customer experience. And Keith Kerpatrick, my colleague here, just finished what we call a Signal Report, which is one of these living comparative reviews on those very products.

Starting point is 00:27:07 We have another one coming out by my colleagues, Diane Hinchcliffe and Nick Patience that's going to look at agentic AI platforms themselves. So I would say to anyone listening to this podcast, come back in, I think, two weeks' time. You should see a signal report specific to those. And these signal reports are actually built using agentic processes. We've built a mechanism that takes all. all of the data that we as an analyst firm aggregate and collect over time. So every conversation we have with vendors in the marketplace,

Starting point is 00:27:47 the briefings we take, the notes we make, all of the information that we gather, it couples that with the information that is out there, that the vendors are giving us and that they're publishing. It takes into account the voice of the customer and the actual experience of the customer through a partnership we have with G2, if you guys are familiar with them, and combines all of that into an automated living system that at any point, I can hit a button and it will generate a very detailed, forward-looking

Starting point is 00:28:22 analysis and assessment of that comparative, competitive marketplace using all of these resources. And so if you see an acquisition, for instance, like the one we like to use as a good example, is if Salesforce buys informatica, what will that do to the marketplace? That certainly would shift the power balance, shift the direction of the market itself. And so we would, as an analyst firm, want to be able to have our living report reflect those immediate impactful events. And that's why we built this as an agentic self-correcting, self-assessing, you know,

Starting point is 00:29:04 it basically just improves upon itself until it gets to the point where we as the builders of the system say, yep, that's it, you got it. And then we publish it. So it's, I'm excited about it. And it's built using this technology that we're talking about today. So what I think, what I find challenging today is agentic AI and the innovation goes so fast. You know, how do you, how do you keep your report fresh, right? How, and, and, and, And maybe a recommendation for people who want to learn more about Agentic AI and maybe protocols like MCP. I mean, MCP, what is it, like a year old or something like that? I mean, it's, it's, right, it's already all over the place, but it's only a year old. So I think one of the challenges is that it's interesting technology, but it goes so fast. Is that something you address with the report too? I mean, you talked a little about it being a living report.

Starting point is 00:30:02 Does that mean you're kind of absorb information as it becomes available and the report will spit out to write the right data? Yeah, exactly. So it could, we could run it 24-7 if we wanted to, and it could just drop 20, I'm sorry, they're actually extremely long. They're like really big reports. So this isn't like a one-page report that we're building here. This is like a deep assessment of 10 or 15 vendors, and we have five different metrics that we score for those vendors, everything from the business value index of them, like how their finances go and all that down to the capabilities they're building into those solutions. So looking at the release notes for a given product on a given day, looking at the financials that were published that same day or the day before. taking those both into account and then building out the report based on that, on that information.

Starting point is 00:31:07 And so that's why I say they're living entities in that, you know, instead of what typically happens with we analysts when we build a comparative report is you gather, gather, gather, and then spend months sometimes writing a report and working with the vendor to finalize that report. And in the meantime, the market has moved on to something. You know, it's like, okay, just the other day, we had a new protocol for agentic systems that lets you purchase. So let's the agents themselves make financial transactions.

Starting point is 00:31:44 So you have the A to A, and now you have A to P, which is agent to purchase by the same company. So Google set this up. And that's, you know, that, The market changed overnight because of that and how we build these systems out. And so if I'm doing an agentic signal on a, sorry, if I'm doing a signal on agentic AI platforms, I want A2P to be reflected in that. You know, who's adopting it?

Starting point is 00:32:12 What are they doing with it? What's the outlook look like for vendors who are adopting that A2P standard into their technology stack? And I guess the only way to make that happen is to use, these tools to help coalesce and sort and analyze that data because it is moving so quickly, as Frederick said. It's just an incredible area. And, you know, Brad, we'll definitely be keeping an eye on the whole space here on this podcast, utilizing tech, focused on a genetic AI.

Starting point is 00:32:46 Also, we're going to be doing a new podcast utilizing AI, which will be a weekly Futurum Group podcast. And of course, we've got our AI field day event coming up. So thank you so much for joining us today, Brad. Before we go, where can people continue the conversation? Because clearly you've got a lot to say on this topic. Where can they find you? Oh, they can find me on LinkedIn, Brad Schumann, all one word.

Starting point is 00:33:13 And you can find me on the Futuram's groups platform itself. So Futuram.com. We publish a lot of material actually outside of our, you know, customer, we have a customer area where we publish a lot of the deep research, like our forecasts and surveys, but a lot of data goes outside of that. And I would encourage you guys to check that out, because we do publish quite a bit at this company. Our analysts are very fast. We run quite quickly. So at any rate, that would be my recommendation. Find me on LinkedIn and find me on Futurum's platform itself.

Starting point is 00:33:53 Excellent. And Frederick, looking forward to seeing you at AI Field Day. Where else can we find you? Well, you can find me on LinkedIn as Frederick V. Heron. We now post multiple times a week, articles around AI, and on our website, hyphense.com. Excellent. And as I mentioned, you know, you'll see me on Tech Strong Gang most Tuesdays here on the Utilizing Tech Post podcast, as well as the forthcoming podcasts as well. So thank you so much both of you for joining us for this episode of Utilizing Tech and thank you audience for listening. We're very glad to have you here. You can find this podcast in your favorite podcast application

Starting point is 00:34:34 as well as on YouTube. Just search for utilizing tech. And if you enjoyed it, please give us a rating or review. We'd love to hear from you. This podcast is brought to you by Tech Field Day, which is part of the Futurum Group. For show notes and more episodes,

Starting point is 00:34:48 head over to our dedicated website, which is Utilizing Tech. or find us on ex-Twitter, Blue Sky or Mastodon at Utilizing Tech. Thanks for listening and we'll catch you next week. Thank you.

Utilizing Tech - Season 7: AI Data Infrastructure Presented by Solidigm - 09x03: Bringing Agentic AI Applications to Market with Brad Shimmin of The Futurum Group

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.