On Ep. 5 of Agentic Commerce, Simon Taylor, Head of Market Development @ Tempo, and Bam Azizi, CEO & Founder @ Mesh are joined by Pahal Patangia, Head of Global Industry Business Development and Payments @ NVIDIA to discuss open source models for financial services adoption, agentic workflows as intellectual property in commerce and more!
On Ep. 5 of Agentic Commerce, Simon Taylor, Head of Market Development @ Tempo, and Bam Azizi, CEO & Founder @ Mesh are joined by Pahal Patangia, Head of Global Industry Business Development and Payments @ NVIDIA to discuss open source models for financial services adoption, agentic workflows as intellectual property in commerce and more!
Timestamps:
Tokenized is sponsored by Visa
A world leader in digital payments, Visa is bridging the gap between traditional financial institutions and innovative blockchain networks, helping players in the payments ecosystem navigate the ever-evolving world of tokenized fiat currencies with confidence and ease. Learn more at visa.com/crypto.
Tokenized is also presented by Mesh
As the first global crypto payments network, Mesh connects over 300 wallets, exchanges and payments platforms, and enables anyone to pay and get paid instantly, anywhere, in any asset. Mesh makes digital transactions seamless, secure and universal, fuelling the next era of agentic commerce. Learn more at meshpay.com
***
We’d also like to remind you that the views or opinions of our contributors today are their own and do not necessarily reflect those of the companies they are representing. Nothing we say should be taken as tax, financial, investment or legal advice, do your own research!
Music by Henry McLean
Pahal Patangia 0:00
So what's happening right now is democratization. I see this as very much akin to the LLMs, the LLM journey that has been that we have seen in the last three years. There's a new open source model which Mistral puts out versus one which Deep Seek puts out versus one which Quent puts out versus Llama, and all of them led people to adopt this, and I anticipate that these protocols, as they are coming in, would bring in more and more people into this pre-poll developers, enterprises, users, whosoever you think of the persona, would to ultimately build upon, and that would also drive the whole aspect about interoperability, which ultimately would lead into the consolidation aspects of it.
Sy Taylor 0:59
Welcome to Tokenized. The show focused on stablecoins and the institutional adoption of tokenized real-world assets. My name's Simon Taylor. I am, of course, your host for today, author of FinTech Brain Food, and head of market dev at Tempo. And joining me for another episode of the Agentic Commerce series is the incredible Bam Azizi, CEO of Mesh. How you doing, Bam? I'm
Bam Azizi 1:21
doing good. Thanks, Simon, for having us again.
Sy Taylor 1:24
This series is really taking off. I think Agent Ecommerce is now the hottest topic in the world, and it's really captivating everybody's attention. And joining us is somebody from a company who's captivating a lot of attention in their own right-the world's largest company, in fact. But they are doing interesting things to support agentic commerce that most people didn't realize. So today we have Pahal Patangia, who is the head of global industry, business development, and payments at Nvidia. Pahal, how are you doing, sir?
Pahal Patangia 1:56
I'm doing great, Simon. Thanks for having me. It's a pleasure to be on the show, and what a great cry of the three of us with me! Looking forward to it.
Sy Taylor 2:03
Indeed, this is everything I love in one place. It's payments, it's Nvidia's heritage in video gaming, it's agentic commerce, it's stablecoins, it's it's all the good stuff. Before we start having fun, though, just to remind viewers and listeners that views and opinions of our contributors may be their own and might not reflect those of companies they represent, and please don't take anything we say as tax, legal, or financial advice, and always do your own research. All right, Paul, 30,000 foot view. What does agentic commerce really mean for somebody like an Nvidia? What does a GPU company, an acceleration company, an AI company, a hardware company do in payments and commerce
Pahal Patangia 2:42
absolutely seven, and I'm glad you started with a GPU company and a hardware and acceleration company because that is what the notion of NVIDIA has been over the years throughout its history over the past few decades, and that is constantly evolving. I would say pretty much over the last 20 years that NVIDIA has transformed into a full stack accelerated computing platform company, which is enabling AI applications for pretty much the entire gamut of the ecosystem. And what I mean by that is, before we get into the aspects of agentic commerce as it relates to NVIDIA AI, it's very important to understand where NVIDIA stands in terms of the platform and what it offers, which is ultimately enabling the AI boom, which you see day in and day out. And the way we describe it is the whole idea of a five-layer cake, which Nvidia offers for the ecosystem to build AI applications on. Now, what is this? What is this five-layer cake? Is the culmination of different ingredients which would make the idea of building AI applications, building AI factories today possible in a scalable fashion. So, at the bottom layer of that cake is the land, power, shell, and energy, which is obviously the common denominator to begin with, doing anything related to AI, and then on top of that are the chips, the hardware, the GPUs, the CPUs, the networking systems that come along with it, and on top of that, the systems at a data center level on how they are stitched up, what we think of them as different units, all of these chips coming together to ultimately build one big computer. So you would initially think of computer as a personal device from a childhood days, but now the data center is the computer, and that is the systems layer. That's the third layer. So when you when you have these systems on top of it is the foundation model layer that sits around. Now, these are the foundation models which encompass the knowledge and the domain and nuances of any topic or aspect. And there are great ecosystem partners: the OpenAI's of the world versus the metas of the world, the mistrals of the world. All of them building these foundation models, but these foundation models need to be compartmentalized into an industry use case, a domain, a domain problem, etc. And that's where the fifth layer comes in, and that's the application layer. And NVIDIA platform is embedded into all these five layers and brings together this five-layer cake. Ultimately, developers would use these five-layer cake platform to build applications for their own use case. One of the key use cases in the world of payments is agent e-commerce, and our pursuit is to embed our platform, both hardware, software, and and the models combined, into the ecosystem players, which are building these applications at scale, and that's why we are enabling the broader ecosystem to thrive when it comes to leveraging accelerated computing.
Sy Taylor 5:52
It's funny to me, Bam, as we've talked to so many people about regenic commerce. People just assume that there's all of this software running all of this hardware in the background, but you've been in the industry for a while. You you understand this stuff pretty well. What are your thoughts?
Bam Azizi 6:05
Yeah, it's it's funny that I have also a post on LinkedIn that went viral about the layered kick of stablecoin payment. It's very similar to what Panel said, but like generally speaking, even on the on that side, what I posted was basically the foundation where the distribution is, and then orchestration, and then the connection. And I argue that the connection is the most important one. Of course, selfishly, that's where mesh sits. I want to know from Powell's perspective what's the most important layer from Nvidia's perspective where you're investing most of your time and money.
Pahal Patangia 6:41
Yeah, no, 100% I think there are two key phenomenon happening in the payments industry from our POV, where we are infusing AI at scale. You know, one thing leads to another. I would say so. The very first phenomenon that is happening in industry is the whole idea of payments foundation models. Now, if you think about the agentic commerce funnel, that funnel is collapsed. The checkout process is collapsed. So, in the past world, where you as a human would have all the context on what shopping you want to do versus how you want to proceed to checkout, there was context being baked into you. But where would agents get that context from. Agents would get that context from learning about the insights about the customer, that persona, the liking, and whatever guardrails and you know limits you have defined for that transaction to be done, starting from the SKU to completing the transaction. Now, to do that, how can agents be armed with those insights? And that is where this new phenomenon, which is very much like an underground phenomenon, I would say, in the realm of ancient commerce, but getting super traction is the idea of payments foundation models. So, financial services and particularly the payments industry and banking industry has had all their data stored in tabular form, you know, structured format all this while. And what you had in the old days was exposing it to machine learning algorithms and building some propensity models out of it, and you would get the likelihood of you know how a consumer would transact or what they would ultimately buy next. But with the advent of these algorithms, particularly the transformer architecture, which has been the backbone of generative AI, there is a newer movement of exposing this tabular data to the transformer architecture, and that is the idea of payments foundation hall. And we have recently had a conference in media GTC where the likes of Revolut, Mastercard, Adien, all of them presented on the idea and the work which they are doing with respect to these payment formation models, and it's powerful because what happens as a part of these models is that they produce something called embeddings. Now, embeddings, to put it simply, are contextual representations of what Bahl would do. What are my dynamic likings from the recent past versus what have been my static likings from the way long behavior which is encoded in my personality. How do I bring all of this together and transform us to an amazing job of contextualizing and putting together those learnings into these embeddings? Now these embeddings would be fed into the agents, which would be ultimately doing actions in the agent e-commerce world, and that is where these two worlds are coming increasingly together, and they become the context layer for agents to execute effectively, to iterate effectively, to make sure that whatever action they are doing, they are within the limits, and ultimately learn and do it further next time. That's pretty much the trend, which is like feeding into this agent e-commerce world. And I want to double down on the agent tech commerce piece, where we are seeing a lot more work. Is obviously the if you look at the market maps on how agent tech commerce and payments are like evolving, but the biggest fruit which we are seeing, which is like. In today is if you break the funnel into search and payments is the search part. The search part, the problem has been broken down for so many years all this while, but now you have better algorithms to perform the search part, and that is what makes this new wave very sticky in terms of better, more personalized experiences, and we are doing a lot of work with the likes of PayPal in this realm, where PayPal wants to enable the agentic commerce capabilities to the ecosystem of 19 million merchants at the end of the day. Now, now these merchants, these now and pop shops and small and bigger merchants, whatever is happening is whatever is happening in the world of AI, they are very opaque to it. So what PayPal does is that PayPal comes in and tries to bring that goodness through their agent e-commerce platform to these merchants.
Pahal Patangia 10:56
The way they want to do this is fine-tune open-source models, which could be managed and controlled by the PayPal environment and the merchants by themselves, also very comfortable, you know, being a consumer and absorber of that stack, and that is the kind of effort which we are working very super closely with PayPal, and they're leveraging some of our open source network run models, fine tuning them to the domain and the use case, and ultimately making these small mom and pop shops agent tray.
Sy Taylor 11:31
I was going to say, Paul, you gave so much there. I want to try and say it back to you, just so that it resonates and lands, because I was like keeping up and keeping notes. So people forget that outside of the world of Anthropic and ChatGPT and the Gemini models, those open source models and NVIDIA is a leader with Nitrimon. I never know how to say Nitrimo
Speaker 1 11:55
is the yeah
Sy Taylor 11:56
that one with with small model performance, like consistently benchmark leading in being able to do that. PayPal, being an enterprise partner of yours, is bringing some of that value to merchants in payments. Value to merchants is the whole game. Merchants really run the world, and so unless you're serving them, you're serving nobody. They're the people selling the products, and as a payments company, they are your customer. They pay your bills, so you have to do goodness for them. I think Stripe launched one of the first payments foundation models. They had some good results on fraud, but what else could you do with a payments foundation model? Like, if I know have this incredibly rich, detailed AI set of 3d embeddings, multi dimensional embeddings about all of the linkages between all of these preferences for all of these customers. How might that help emergent sell more, serve their customers better, and do they want to take give all of that data to one of the big AI labs? Probably not. They want to use, but turns out open source is just six months behind the frontier, and even then, it's like by six months, performance-wise, day to day, most people are not going to see the difference. And these things are phenomenal compared to the free version of ChatGPT that most mom and pop shops are using already. So PayPal gets to give them something that feels incredible, and yet what they've done is package Nvidia products under the hood, and and that's something that I don't think people really realized about what you're doing. And I found it so interesting in the financial services survey that 65% of respondents are actively using AI, but there's 84% said open source models are important to their AI strategy. So, talk to me about why that is. Why is it that this open source model and the ability to use it is becoming so so important to the financial services industry in particular? Yeah, of course. The way the industry has been, it has always been a couple of steps behind when it comes to the latest and greatest entrepreneurship, and that that delta is the regulation. That delta is the ask for explainability. That delta is how much can I uncover in that black box model so that you know I can safely put into production. That's being the broader case for the industry. So where where financial institutions and even retail merchants, for that matter, are seeing more and more progress is that they
Pahal Patangia 14:30
want to see what's happening as a part of that mix of the solution which they are building. Obviously, want to have control of the weights and the ability to fine tune it at a greater scale. And Simon, you rightly pointed out, open source models are almost at par with the large shop foundation models out there, and this parallelism and accuracy then further drives the discussion onto other vectors about cost, about control, about maybe. Making sure that you know you check all the right boxes when it comes to the regulatory aspects of it, and obviously resiliency as well, because you may want to have different options when it comes to building these applications. And it's all to say, foundation model providers are our greatest customers and partners. But at the same time, when an enterprise needs more flexibility, which caters to their use case where open source fails better, the likes of Nemotron in the NVIDIA software stack with Nemo libraries, which help them fine tune these models, and that is like a big aspect in agent e-commerce down the line. All of that comes together as a part of the platform,
Sy Taylor 15:42
as somebody who's just been on vacation to a certain part of Orlando, why this didn't come to me sooner? But finding Nemotron must be something you talk about internally at Nvidia, and now I'm saying mine, mine, mine in my head. I can't stop it. Other other sort of movie franchises are available, but bam! How are you thinking about open source versus closed source in Agentic Commerce more broadly? Because this is kind of well, open weight versus closed weight. Really, it is something that, as a builder of a company in the stablecoin space and elsewhere, are you hearing about it from your customers?
Bam Azizi 16:15
Yeah, I think at the end of the day, customers they don't care about open source, closed source. I think the community cares. I think it's good for the community. It's good for the science. It's good for the technology that we're building together. So we need to push the open source as much as possible. The fact that OpenAI, Open Claw, they have the word "open" in it, but they might not be as open as we wanted to be. But I think it's good for the community for people want to build, but at the end of the day, customers wants the best of the best options in best in class practices to be able to run the show on on the front end. One thing that caught my attention, or when Paul was talking about it, was typically NVIDIA was like the Harvard layer, and then there is always like an intermediary or middle layer, like ChatGPT, like Cloud and others, and sitting on top. And then on top of it, we have the application layer. But you were talking like chatting with PayPal. PayPal is a great partner of Mesh as well. Does it mean that you're kind of getting rid of the middle layer and you're going directly to the partners? Does that mean like higher speed, faster, cheaper, better service for them. I'm just curious to know, like, does that raise an alarm for companies like OpenAI and Cloud?
Pahal Patangia 17:29
Not at all. Not at all. We, as a company, have a mantra of meeting developers where they are. So, if somebody wants to develop with our closest and biggest partners, which is the foundation model providers, we are more certainly than happy for them to do so, and would help push that effort into a direction where we get the best of the worst from the foundation models. At the same time, if somebody wants to build with open source, and this is governed by a lot of like business justification within the customer's organization. Then you have all these tools and libraries that could help you get there at ease using using this full layer of cake, combining software and hardware.
Bam Azizi 18:18
Makes sense.
Sy Taylor 18:18
I think that set of trade offs is fascinating, Paul. How do you coach a payments company who is thinking about serving their merchants through use cases? And what are you hearing from those payments companies? Like, I'm guessing with open source, it's going to be different, right? Like by use case, how do you how do you help people weigh up those trade offs?
Pahal Patangia 18:39
There are a lot of considerations, particularly when you are running so many complex models and models today, agents tomorrow, and will be like multi-agent systems in the very future. What matters is not just the accuracy of it, but once we have optimized for accuracy, what are the other drivers that help begin into this equation. So you could think about costs. So as you scale up to say in this example, 90 million merchants, you are making so many inference calls at the end of today. How do you make sure that those inference calls are best optimized from a cost perspective for your use case? That's one kind of consideration. The other consideration is latency, because nobody wants to be waiting with a snake game, whatever that runs on Google Chrome when you know Acton dice, and you need millisecond latency decisions. Millisecond latency response has to think through the model, has to think through, reason, and fetch information from different sources, apply all of that in the context of the question or the problem he asked, and take an action within the guardrails. To do all of that is a lot of tokens, a lot of decisions, a lot of flows, and have to be done very dynamically, smartly. And agents are capable of doing that if they are fine-tuned, if they are within the right guardrails. And then you do this one one pass, and then you have the feedback loop going in, and that's your data flywheel. So you have more data coming in from the previous work which you did on what was expected versus ideal. Then this feedback loop helps makes model better. And then when you take this analogy from one agent to multi-agent systems, thinking shifting the foundation to agent to commerce, think of multi-agent systems being one coming from the network side versus the shower side versus acquirer side. All those agents talk to each other versus a procurement agent sitting in an SAP talking to the inventory side of the equation versus going to different finance house within an organization, how does the system reason through and the residuals are better? This this becomes like an explosion of tokens. That's where the whole idea of like tokenomics comes together, which is not just you know grokking tokens out, but at the same time, how do you grok tokens out at in the most efficient way, cost wise, power wise, and obviously the latency and accuracy as a sit at a part of the equation.
Sy Taylor 21:15
Token quality output per kilowatt hour. There's some sort of an equation in there that's I think really, really essential for people to understand because you can burn a lot of dollars with API fees if you don't know what you're doing. Any enthusiast that's played with Open Claw will tell you you can very quickly run up $1,000 a month just connecting a few APIs, and suddenly you go down all kinds of rabbit holes. And for an enterprise that's meaningful, if what you were doing before was a Snowflake instance and CNNs are different types of machine learning, the economics here look wildly different for what is doing a loyalty product or a fraud screening product. And as you see, inside of a card network, that's going to be quite different to inside a merchant, inside an issuer, inside all of these actors have different internal roles that they want their agents to do, and all of those agents need to be different levels of quality. So they need different tokens. So there's actually a surprising amount of complexity in in setting that up, and not only setting it up for the economics, but then making sure it's getting better over time. That it's learning like a person would, you know, like, hey, you did that wrong. Don't do that again. Again, surprisingly hard if you've run an open claw to get the thing to do what you want it to do consistently. So to do that for enterprise with the experience of Nvidia is is fascinating. But let's zoom into e-commerce because we're sort of we're talking about agentic commerce specifically. How is this impacting commerce today? Are people feeling this at checkout? Are they seeing it? Like, where are you seeing kind of the the value coming through?
Pahal Patangia 22:48
Yeah, um, our pursuit as a company is to enable the players which are ultimately bringing the value to the end consumers, and that is where people want to work with the likes of the payment platforms like PayPal. At the same time, they want to work with large retailers so that they have a consumer-facing agent return on top of them. So we keep seeing, and this is you know beyond Nvidia, but the broader ecosystem, the way it's traversing, is you know how Mastercard is completing transactions into multiple countries, which are like purely Asian-fledged. I think those are some early indicators of success, which helps us think that these things will ultimately become mainstream. And obviously, there are a lot of considerations which we need to do. You know, is it yielding even a good checkout conversion at the end of the day, with all the recent newses that are coming around, and it'll take a while before I would say that it would require a lot more fine tuning and harder in these agents for them to be very capable of performing that specific task autonomously. We're seeing early signs with the Masterford example, which which I shared. But I think there's more work to be done, and the ecosystem needs to come together on that front.
Sy Taylor 24:10
I want to give a shout out to my friends at Sardine who've been doing a lot in the fraud space, where they have data analyst agents that can just do like fraud investigations. But you're benefiting from this sort of history of data, where they have 7 billion devices in their network, they built custom models for their entire life of their company's history, and they have agent performance logs of what worked, what didn't, and so it's it's that intellectual property applied to the agent, which is where the secret source was, it used to be about turning that into a SaaS workflow or an API workflow. Now, really turning that into an agentic workflow. Your agentic workflow is really where your IP sits in e-commerce, and I think that's such a such a crucial point. I think that's a great point, Pahol, to to just take a quick break while we hear from our sponsor. And I got to thank Bam and Matt. For making this show happen, because it's it's the most fun I get to have in my day job. It really is. So we'll just hear a quick word, and we'll be right back. This episode, if it's not obvious, is brought to you by our friends at Visa, a global leader in payments. Visa's tokenized assets platform VTap uses smart contracts and cryptography to help banks bring fiat currencies on chain. VTAP allows financial institutions to issue fiat-backed tokens, improving financial efficiency and enabling programmable finance. You can check out the links in this episode's description to express your interest in VTap. This episode is brought to you by Mesh, the first global crypto payments network. It connects over 300 wallets, exchanges, and payments platforms. Mesh enables anyone to pay and get paid instantly anywhere in any asset. Mesh makes digital transactions seamless, secure, and universal, fueling the next era of agentic commerce. Learn more at meshpay.com/ai. All right, thank you to Mesh, and thank you to all of our sponsors for making this show happen. Bam, I don't know about you, but I hear a lot about different protocols out there. There's almost too many to name. How are you talking to to your customers about some of the protocols at the moment? And like, what are your questions for Nvidia on that?
Bam Azizi 26:33
Yeah, I think the main question is: Are we going to see more consolidation or fragmentation? That's a billion-dollar question. If whoever has the answer can build a generational business here, I think it's hard to say. If I have a crystal ball, I would vote for more consolidation, similar to what we have we have seen on the internet. So there were a lot of different protocols, but we all landed on HTTP. There were a lot of protocols for device-to-device communication. We are all now using basically Wi-Fi and Bluetooth. So I would see we like even on the the charger level, we are seeing different type of chargers. Every phone had different charger or different socket, but all consolidated to one or two. I would say similar thing would happen, especially now that with the recent announcement of X402, that they were like taking X402 all the way to the Linux level, and they want Linux Foundation to be the host, and which is a neutral party, and then all the other companies like Stripe and Coinbase are going to support and put their brand out there. And my my background is authentication and security. We've seen similar type of consolidation on the authentication protocol. Now everyone is using Paske, but before that, we had like too many different options to pick. So that is my guess based on the knowledge I have, based on the conversations I had with a lot of different customers. But I'm really curious to understand what Paul thinks on on the on that side, especially that is there going to be a different protocol for human to agent interaction? Like agent is buying something on behalf of human and paying a merchant, which is a human, compared to agent to agent transaction, right? It would be different type of UI UX, different type of protocol. What are you seeing in the market right now?
Sy Taylor 28:22
And on that, Paul, I think I'm reminded of the XKCD comic. You know, there are there are 14 standards for authentication. We need one standard to unify them all. Sometime later, there are 15 standards on authentication. So, like you've been in the space for a while. How are you thinking about all of that?
Pahal Patangia 28:43
Yeah, I mean, Bam rightly pointed out. If I had a crystal ball, I believe would stop from there. But I think from RPOV, what's important here is that all of these protocols. What ultimately, this is my personal perspective, and agree with Bram's point that they would ultimately converge into few key names, whosoever it could be. It could be someone new totally coming in, but at the same time, what these protocols are doing right now is also kind of reinvigorating some pockets of developers who want to start building using one or the other protocol. So what's happening right now is democratization. I see this as very much akin to the LLMs, the LLM journey that has been that we have seen in the last three years. There's a new open source model which Mistral puts out versus one which DeepSeq puts out versus one which Quent boots out versus Llama, and all of them led people to adopt this. And I anticipate that these protocols, as they are coming in, would bring in more and more people into this pre-poll developers. Enterprises users, whosoever you think of the persona, were to ultimately build upon, and that would also drive the whole aspect about interoperability, which ultimately would lead into the consolidation aspects of it. So that is how I see it. And as these different developers, as these different users build these agents, build these capabilities. There's more and more appetite for running them securely, for running them in a in a guarded environment, and you know just everybody will be building the clause, but secure clause at the end of the day to consult to complete a transaction, etc. We are not something called Open Shell at GPC, which just concluded last month. Now, Open Shell is like a runtime, which it's an open source runtime. It's security hardened that sits between your agents and your infrastructure. It helps you create sandboxes to make sure that whatever prototyping you are doing is in a confined VM at the end of the day, so that well
Sy Taylor 31:06
you limit the blast radius when it's in a confined VM, right? And and it's it's these little idiosyncratic details that people don't think about, which is I'm building agents, I have my production infrastructure. How do I not put my agents, you know, like I've got containers and I've got Docker and I've got VMs in my production infrastructure. Hopefully, but not all companies do, especially in financial services. But to sort of limit the blast radius, I'm building these new agents. Am I just going to put them into that? Or wait, it would be really helpful if we had this nice little place to put them. And OpenShell sounds like a really nice way to do that. And Bam, it's interesting you talk about passkeys and authentication being kind of your experience. The one I was thinking about as Pahal was talking just then is what about the early days of mobile commerce with WAP? Do you remember WAP and like mobile phones pre the you know back in the feature phone era and people were trying to make payments through mobiles, and long before Apple Pay, and and I wonder how early we are in some of this stuff as well. Sometimes, so you know, I'm curious as to how you think about where do you spend your energy power at this point, because there's so much you could do, there's so much you could support, and really, like, what are you focusing on? You know, are you looking at, for instance, stablecoins and where do they fit? Are you looking at traditional human to agent, like agent for human, like traditional commerce flows? Are you looking at agent to agent? You trying to do a bit of all of it? Like, what? What's the? How do you focus? How?
Pahal Patangia 32:38
So, so good question. I would love your folks's perspective as well on that. You mentioned about stablecoins and obviously being at the heart and center of all things payments and media. It's about finding those key trends today. Those are payment foundation models and Asian tech commerce, but within there are like subsets of topics which keep coming in as emerging trends into the payments for it. We see stablecoins as just another wave that would bring upon the traditional fiat system, the the network rails, etc. And what that would do is ultimately, again, to the point of increasing adoption and bringing new kind of players. New generation would be more stablecoin friendly than being credit card friendly, for example. And there are already been great attempts and in bringing the two worlds together. So what I would do is the original AI use cases, the canonical AI use cases which have existed on the network rails, be it the fraud detection piece, be the authentication piece, be it personalization piece. All of them would continue to go on in the civic world. So, from my perspective, my pursuit is to is to continue to double down on that on the use cases which we have.
Sy Taylor 33:52
Yeah, it's the value added around the payments, isn't it? The rail may change, but whether using stablecoins or card networks, you're still going to have fraud. You're still going to have customer loyalty. You're still going to have all of these problems. So by focusing in in those areas, Bam, I'm interested in your view though as possibly the most connected network in the world in stablecoins. Like that's literally what you do. What are the main differences around agentic commerce and stablecoins versus the other rails.
Bam Azizi 34:22
I think agentic commerce can potentially use different rails, especially for what we're seeing on a day-to-day basis. That people are searching on their ChatGPT or on Anthropic or Perplexity, and they search for like a let's say pair of shoes or a T-shirt, and then the agent can go and pay on behalf of the user, and then the agent can use credit card, can use stablecoin. So they're kind of in par, maybe for cross-border payments, maybe for international shopping or international purchases. Stablecoin has an upper hand, but when it comes to agent to agent, I think nothing can beat. Stablecoin rails, and the reason is most of these transactions are microtransactions. How can you pay 0.0005 cents with Visa, or how how do you want to wire? Like it needs to be online, needs to be global, it needs to be right away. So that's where stablecoins are shining. The other part is the number of transaction that could happen. Right, so every human, in average, does two transaction per day. Agents might do 2000 transaction per day. So that TPS, that throughput, only supports on blockchain. Traditional payments are not built for agents, and they're going to fail. That's why I'm very bullish on a stablecoin and the the use of stablecoin for agentic commerce and for agentic in general, and I think the way to unlock it is basically consolidating on the protocol side, making sure at the protocol level it doesn't matter if it's a stablecoin or traditional payment things that like companies like Tempo are doing on the MPP side, so I think there is a lot that can be done there. We are at the beginning of that era, but I have a famous quote that what e-commerce did to commerce, which was like made commerce 10,000 times bigger, agentic commerce is going to do the same effect to e-commerce. So we're going to have like yeah yeah 10s of 1000s of transactions, millions of transactions happening in a second because of agent e-commerce.
Sy Taylor 36:25
It's an order of magnitude explosion, isn't it? I think there's something like 4 million emails per second on the internet, and that's just emails before you get into video. So how on earth is you know 60,000 transactions per second peak load going to be anywhere near enough for that kind of world, but bring us back to Earth, Paul. Like, where's the real traction? Where's the volume? Because I like to joke that there's more agented commerce protocols than there are payments these days. Certainly outside of the meme coins, but you probably have the best view out of anybody. You're like the infrastructure behind the infrastructure behind the infrastructure. So, where's the customer pull coming from? Where's the real volume? Where are the use cases?
Pahal Patangia 37:08
Yeah, there are two aspects to answer that. First is how the ecosystem is approaching it, and the way we divide it, as I was mentioning in the PayPal example, was the search and the payments part. Search part is done in, I would not say done and trusted, but it's a mature part of the equation. The payments part is where a lot of this sandboxing is happening, and that is where I'm very bullish on the likes of OpenShield, etc. Enabling the ecosystem to build those agents and have the capability to transact. That's what I see from the ecosystem part of it. If we overlay that in the longer term, I'm very bullish on the idea of multi-agent systems being part of this entire process on how these agents would interact with one another and how will we play a role in making these systems better at the end of the day. That's through the feedback loop. That's through the security runtimes. That's through the cart railing. Obviously, you'll need a lot of fine tuning to make these agents do what they have to do and not deviate. I think those are the aspects which we are looking forward to help the ecosystem develop.
Sy Taylor 38:18
I think there's so much to help out on. I think some of the themes today are really around those tokenomics, which you know Bam and I both had a little smile about because in the stablecoin land we think about tokenomics in in a different way.
Bam Azizi 38:30
Different tokens,
Sy Taylor 38:31
different tokens, but it's tokens all the way down. There was a great Ribbit piece about it's tokens all the way down, and it's something I've been talking about for many years. Whether it's identity and passkeys, you know something you guys know well, and the tokens in the cybersecurity world, Visa, Mastercard's network tokens. Then there's kind of the open banking tokens, the stablecoin tokens, and of course there's the token economy, the token factories, all very different technically in terms of what they do. But we call them all tokens. Token is the most frustrating word in the English language, it literally means a stand-in for something. It could mean anything, so it's kind of kind of a frustrating word in that. At
Pahal Patangia 39:09
least the doggone could have.
Sy Taylor 39:11
Yeah, but but but you have to know your economics from it, and I think if you know your economics and you can think about throughput and speed, which is what delivers customer experience, or whether it's stablecoins or any other network, all AI experience, speed and price always seem to be the thing that brings us back to it. So Paul, I want to thank you so much for this. As a long-time fan of the big green monster that is Nvidia, and as a fan of payments, this has been a real joy. And the fact that you're naming everything after a small fish that lost its way from home. Nima Cole, I'm using it. There's there's just so much to like about you guys. But Paul, if people want to find out more about you or what you're doing in payments at Nvidia, where do they go to do that?
Pahal Patangia 39:53
Absolutely, please reach out to me personally on LinkedIn, and I'm available on email at people. AI at NVIDIA.com, and if you want to learn about all things financial services and media, there's a dedicated industry page on the India website where we highlight all the good work which we are doing, not just in the world of payments, but banking and capital markets as well. And would love to sprinkle that goodness of AI, as I said at the beginning, to everyone that we causes so, and happy to be a partner.
Sy Taylor 40:25
Thank you so much for how and Bam, of course. If people want to get connected to the Mesh network itself, how do they do that, and how do they find you?
Bam Azizi 40:32
Yeah, they can go to Meshpay.com or on Twitter or LinkedIn Meshpay, or for me, like they can search Bam as easy Mesh on Telegram and Twitter,
Sy Taylor 40:41
and you'll find me at Sy Taylor on all of the socials. Screaming into the void at fintechbrainfood.com. Just wrote a piece all about invisible commerce, how I think agentic commerce might be broken, and so I have hot takes too. And of course, you'll find this show if you subscribe to it and you hit the like button and you do all of those things that podcast hosts make you do whenever they're closing out the show because it actually does help others find it. And for conversations like this, it's really, really worth it. So if you enjoyed it, tell a friend, and we'll catch you next time.