Hi everyone, welcome to Payments TV. I’m very excited about this very special episode we’re gonna have today. My name is John Gaffney. I’m the Chief Content Officer here. We are gonna talk about AI and AgenTik AI. And it’s for our What’s Next in Payment series, The Rise of Digital Labor, Exploring AgenTik AI and Banking and Fintech. You know, AgenTik AI has emerged as a pivotal focus across the sector. It signals a new competitive battleground for companies in AI. Analysts are noting that the technology’s potential extends far beyond traditional AI applications. And we’re going to talk about that in just a second. Unlike conventional AI, which operates within predefined constraints, AgenTeq AI possesses the capability to autonomously assess, plan and execute tasks. How’s it going to play in the payment and financial services space? We’ve already had that discussion with Mark Sunt, CTO at Stax Payments, and I’m looking forward to another one today. Mark, welcome. Hey John, it’s great to see you again. Great to see you. You know, we’ve talked in the past, we had a great conversation about how AI plays at Stax payments. I understand that there’s been some developments over the last month, month and a half that you’re very excited about. Want you to tell us about those. Yeah, hey John, thanks for having me. Last time we talked, we talked about AgenTeq and we talked about cloud architectures and how these things were all coming together. And I think we kind of relived some of our past history. It’s like, hey, remember when computers had monolithic applications and we moved to multiple applications on a computer and they needed to talk and there needed to be this inter process communication and IPC was all the rage and it was named pipes and shared memory. Then we moved to distributed communication using something called DCOM and COM, and then there was CORBA. There’s all these ways that applications used to be able to communicate, and I made the analogy a little bit, and we did, talking about how I kind of feel like I’m reliving this in this age of AI because we’re building these huge monolithic large language models and we all figured out it’s like, wait a minute, it’s a mile wide, it’s an inch deep, it answers two or three questions, it’s like, is that all there is? So we knew that we had to move to more specialized models. I think that was kind of the obvious conclusion of the discussion we had in November. But there’s kind of this gaping hole of like how do these things communicate? How do you get information context between the large language models and how do you orchestrate what’s going to happen? And the analogy has always been like, well, it’s a little bit like distributed computing. We have these objects and they communicate. We had these elaborate ESBs and we would sequence things to get done. Those are interesting architectures, but they tend to be brittle. They tend to be of if and else based, they’re not very dynamic. Well, what’s been released, and it was released November, right when we were talking about philanthropic, and it’s actually called the Model Context Protocol or MPC. Let me show you what I’m talking about here real quick because this is absolutely game changing. And this is anthropic, Mark? It’s an open source anthropic announcement. It happened at the end of November of last year, but it’s the model context protocol, and it’s really getting a lot of headwind or wind in its sails, if you will, because people have figured out, it’s like, hey, I love these multiple models, but they’ve got to be able to communicate between themselves. So the MPC model is an open source solution that was released by AgenTik, And there are a ton of people who are adopting this, comes in kind of a client flavor where you use the models and there’s a server flavor where you’re really exposing functionality from applications you’ve written or from other large models. And they’ve got SDKs across a whole bunch of different languages. But what really gets me jazzed about this is that it’s going to give us the capability of having a hierarchy of models. So doing things like a model can go discover the capability of another model. I can call it as needed. It’ll have this distributed capability of being able to have hierarchy. The analogy I use is we had these ESPs that were kind of you do this step, you do this step, you do this step. In the financial services space, you had to have pixel perfection with the OCR that would read the PDFs you’re looking at. But now I can literally have a model that’ll take a look at a PDF. I do this all the time with my cell phone bill and it looks at it and it doesn’t have to know. I can ask it like, what kind of bill is this? I don’t have to have the cell phone. And it will tell you. Yeah, it’s like it just reads it and then what’s great is I can say, what is this? And it’ll tell me, hey, it’s a cell phone bill. And then I ask it like, hey, what number was called the most? And luckily it’s my wife and my NCAA buddies. But it does that and you can have models that are actually just looking at information around customers. We have this thing called know your customer. It’s a big deal in financial services. And I can build more generic solutions now for vetting customers because it’ll just look at the documentation, it’ll figure out what it is, and then it’ll follow the appropriate path. And this MCP protocol is going to let us do that. So it’s super, super exciting. And I’m really jazzed about the solutions that I’ve been able to create already using some of the libraries that are out there. So one of the things we talked about Mark was Confident AI. Are you confident that Agenetic AI can talk amongst themselves? Yeah, I think this is, it’ll be the Luna Franco of being able for these things to be able to communicate back and forth and be able to do discovery and to be able to expand their capabilities on the fly. And this might be the first iteration of this inter process communication, but it is absolutely game changing. This is like I’m losing sleep over it kind of thing because it’s that important, frankly. So are you losing sleep because of all the opportunities, or are you losing sleep for the challenges? Well, probably a little bit of both, but I do think the opportunities here are really, really good because we talked a lot about creating the kitchen sink LLM and it doesn’t scale and they’re super expensive. We didn’t have the customization that we needed to support business functionality. So it really is kind of this iteration of doing business process and you hear robotic business process automation has been on for a long time. And it’s how do we extend that with AI? And that’s what’s really getting me jazzed because I think this is just the beginning of all the opportunity we could think about here. It is tremendously exciting. It’s good to talk to somebody like you that really gets it and sees the opportunities. So what are the things that spooked me in getting ready for this interview, Mark, was the pricing. Think OpenAI actually announced $20,000 per agent. I mean, the pricing scare you? Do you think it’s something that’s kind of temporary right now? It’s a great, great question around pricing. I think there’s a lot of variables that have to go into this. I literally am building AI models to do the pricing analysis, right? I mean, I can now do a deep research with these models and actually build out. The thing that kind of the unanticipated costs are how much data are you moving? Is there transactional costs? What is the step function costs? Or is an entire workflow a cost? So there’s a lot of pricing information out there. I think you have to really look at the unanticipated costs and data movement and inference costs are going to be one of those things you got to model and you got to try to figure that out or you’re going to be in for a big surprise and you build these things and you figure out that you’ve got a flow of 50 different steps and every step kicks off a cost counter where there are other technologies out there where workflows are end to end and a workflow is what triggers a cost. Just you’ve got to look at the costing details and model these things out end to end or it’ll be really expensive really quickly. But I would say with regard to the level of pricing, the $200 per month ChatGPT Pro is literally giving me grad student level research. I’ve been applying to it that I don’t have time to do anymore and I don’t have any interns right now. So it’s one of these things that I can throw it into research mode and can do a deep research and give me back pretty credible findings with references that I can use to make sure that it’s actually a defensible outcome. I think $20,000 is cheaper than a grad student for a year, so that might be a cost benefit because there might be some opportunity there, but I think you have to explore the models. Use the free version first, do some research, and then step into that model as you need to, as you find utility. But we’ve had some incredible outcomes from some of the deep research capabilities of ChatGPT. But Mark, tell me about some of those. I mean, technically a year down the road, if this all nets out the way it fits, what’s some of the outcomes we’re looking at? Yeah, so for me, one of my favorite ones is we acquired a new company and they went through a PCI audit. I had to sign that audit and I was kind of interested in seeing how thorough the audit was. So I created a model with the PCI specification 4.0.1, and I ran our findings and our outcomes through that model. And I was able to identify very quickly that there was a few things within the compliance space that were tested, but maybe not with the rigor that I had hoped. So it literally validated a PCI compliance outcome that I really wanted to know much more deeply since I was the person signing that certificate, that attestation actually. But it was absolutely like a super productive opportunity for me to vet an audit that we’ve never done before in a brand new company that we just acquired. So I’m interested to know how the interactions with the rest of the C suite at Stacks work. So let’s say you’re talking to the CEO, you’re talking to one of your colleagues, and you’re telling him about all this, What’s the reaction? I mean, are they saying, I’m glad it’s your problem as CTO, or are they saying, come me when we’re done? I work for an incredibly open organization that is so welcoming to new information, new ideas. And in fact, I did a CEO presentation about AI and our adoption of AI at Stacks, and everybody wanted to know about what we were doing, how we’re using a technology, how it’s actually benefiting the business, and they’re very open to it. In fact, I kind of have a little bit of coaching. Said, I think during some of our executive calls, say, what did I learn from AI this week?’ type of lessons learned and I’m hoping people adopt the technology. And in fact, I talk to people in our org all the time that are using some of the tools, NotebookLM being one of them, that is just an incredible tool for building RAG models from your desktop. And that’s something that I used to write a ton of code for and host a site. And I literally put a bunch of documents in a notebook LM model and I get pretty much the same outcome. It’s incredible. We’re adopting it, and we’re adopting it with guardrails, so that’s still the thing that we have to think about. What’s the context of what you’re putting in there? You’ve to make sure that the vendors you are using have privacy controls in place that you feel good about. There have been vendors who have some really loose controls. I. E. CI can’t do it. So we’ve had to say no to some vendors that have been out there. But I think that the tools we’ve adopted have pretty rigorous security and privacy policies in place, I feel comfortable using them. So NotebookLM is the Cinderella of AI. It’s amazing. If you haven’t loaded it, load it right now. Right now. Absolutely. Unbelievable. I mean, I use it for vetting candidates. I literally put their job resume in, I put in our job description, and I say find the gaps and then give me questions to talk about those gaps. I use it for contract vetting. It’s like what are the contracts I should be, or what points within this contract should I be concerned about? Not that I’m trying to give my legal team advice, they’re smart people, but I want to at least understand like, Hey, where are the red lines going to happen in this contract so I can prepare the vendor for the conversation? I use it almost every single day for some functionality, and it’s actually built into our Google subscription. I feel good about the privacy and we use single sign on, so it’s control. On the same note, I use Copilot for Microsoft documents, I can take a spreadsheet. It’ll do literally linear equations. I’m asking it to solve problems within a spreadsheet by giving it, I don’t know, eighty-forty parameters and saying solve this for x. I mean, this is stuff that I would have had to written a program for a while ago and now it’s just a matter of routine. I’m using English to query spreadsheets. It’s amazing. Wow, that’s amazing. So Mark, let me ask you this for my final question. Do you see this as a combination play in terms of who gets to supervise it? So for example, the reason I asked you about the C suite, what if a CMO comes to you and says, here’s the problem I have with digital marketing. I’ve read that AgenTeq AI can help out. Does that then go to you and how do we align the organizational objectives with this incredibly exciting development? Yeah, I do think that getting the product organization on board with understanding the opportunities is huge, but it kind of comes down to what’s the business business need, and what are the expected outcomes, and then how do we apply this technology? So I always think of, as a consulting manager, a scope approach team and timeline, but let’s come up with a scope, let’s come up with an outcome, let’s come up with an approach, what’s the best approach, and then I’ll figure out the team and timeline to actually recognize the benefits of the technology. So I think you have to have a systematic approach to it. I think Dave kind of just messing around with it or over it. I think now everybody’s looking for what’s the expected outcome here, and how is it going to actually move a needle for the business. It’s kind of the formality of just software development life cycles that we’ve had, or just a company has as they mature their product life cycles and product delivery capabilities. And I think AI is just following that same trajectory, but we’ve got to rationalize it now. For tinkering’s over. Make this thing real, like what does it mean to be real? Mark, I need you to talk about confident AI, that’s our phrase. AI? Well, I’m pretty confident. It’s coming, that’s for sure. And I do think, yeah, these changes like MCP are absolutely inflection points that we need to understand. And it’s interesting the technology was released November, and now you go to YouTube now and do a search on MCP, and there are dozens and dozens of videos. So it’s interesting to see the groundswell of people adopting this technology because they finally, including myself, had came to the realization like, hey, these things have to communicate. How are they going to do it? Like holy smokes. Here’s how. That’s amazing. All right, another great conversation with Mark Sumpt, who is the Chief Technology Officer at Stacks. Mark, thank you so much again for joining us. Yeah, thanks John. It’s always a pleasure. Thank you so much. Have a great day.
