Speaker 3: Well, thank you everyone for joining us today. We have such a great panel discussion. We have three folks up here. We have 30 minutes, so we are gonna dive right into it. So, agenda commerce, about a year ago, I'm not sure if many of us knew what the term meant. I think it's probably all that we can really think about right now, at least on this panel. Maybe some of you are even searching what it means right now. But just to kick things off, Jason, I'll look to you for this first one. What are we talking about when it comes to agenda commerce? What do we mean? How is it different than the AI assisted commerce that maybe we would have been familiar with for the past couple months.
Speaker 2: Yeah, so I think agentic commerce can have a broad definition. I actually starting by defining even what an agent is in the AI world. An agent is very simple. It's an AI model that uses tools and has guardrails, instructions, and it achieves something in the real world. That's really all that definition means. And this could be a variety of different tasks. When we think about in the commerce world, there's this end state that we imagine where agents armed with context, Able to navigate the world, we'll be able to handle every piece of the shopping chain, of the commerce chain, I should say. That includes discovery, it includes evaluation, the checkout, and potentially post-checkout with returns and other components there, other steps there. But you can imagine a world where agents are almost an avatar of each individual. They're able to act in a way that helps accomplish commerce works. What I will say is that that is that's the end state of where all this might lead. I do think there's this huge gray area middle ground, which is every incremental step towards that. And so that's discovery powered by AI, but humans are making the final decision. And and and we may have different words for that, but I do think in our vernacular, in my vernacular, think of that all as agentic commerce. There's just different pieces of how we'll how we'll accomplish that. And there's different commercial motions that will still take take different approaches.
Speaker 3: Michael, is there anything that you might want to add from the merchant network data side?
Speaker 1: Well For context, right now, probably less than 10 basis points of all transactions are done agentic. Now, our definition would be a transaction that happens end-to-end without any human involvement, right? So if it's just discovery and then you have to complete it, we wouldn't define it as agentic. So maybe for that, in some verticals, you may go up to 10 or 20% in terms of discoverability. but for actual transactions happening, it's a very, very slim percentage of transactions, and it's probably growing about 20% months over months. And it's definitely top of mind of Pretty much any merchant we're talking to. I actually checked right this moment. We're having a webinar with dozens of merchants about this. it's it's top of mind across different functions at large companies, right? Because it will impact how their business is gonna run, right? Where are they spending marketing dollars, where are they s what do they need to do in their tech stack to advance it, and probably a lot of other things we'll cover in the twenty-two minutes we have.
Speaker 3: So I mean I think there's a a little bit here which is a a question around is this something that's happening right now that we need to be thinking about or is this something that's gonna play out for years in the future? So you kind of spoke to it a little bit, but where is your head at right now? This is very much a moment in time we need to capture, or maybe this will take a a couple years to play out fully and it needs some time to mature.
Speaker 1: I think it's gonna happen substantially faster than people anticipated. a lot of the customers we talked to expect it to happen. And part of the problem here is if you're not ready as a merchant, it'll just happen over you and you'll be it'll be very, very hard to correct. I did an exercise maybe in July. I needed a new Whatever, not an AC because I live in London, but like some cooling mechanism. And I was working with Chat2PT trying to understand what I can get. A lot of the merchants were blocking all agency traffic into them. So just the discoverability side, right? Because they're all have multiple bot detection technologies. And when the information came back, only two sent the like w were able to send this the model, this is the price, this is what you want. And I had to pick between them. Then I met the next day, I met the CEO of one of those. companies like they you're blocking all the genetic traffic, you're not gonna be discoverable. It took them about a week and they fixed it all. Like it's it's happening right now and and people know they need to get on it.
Speaker 4: I would like to add that forget commerce, fundamentally procurement and shopping are quite different things. because procurement is typically more logical than shopping. Shopping experience involves a lot of emotional decision. It's not like you know, always just optimize for like you know the best outcome from a bunch of parameters. While procurement can be much more logical on one hand, and So on the other hand, the data environment is better usually than like you know Shopping experience, because shopping experience, a lot of data actually exists in the open web, sometimes even just social media, right? So people's kind of a comments, feedback about a certain product. While like you know, procurement data can be much more structured, exist in the ERP, exist like you know in a bunch of APIs, machine to machine. So I believe like if talking about which might come first, like no matter for B2B or B2C use cases, when it comes comes to the category of procurement, because for B2C you can also have like those grocery procurement. For example, someone some agent or AI monitoring your like you know refrigerator, whatever, like you know, decide like on whether you need to backfill something, it might be easier. But when it comes to shopping experiences, not only like the logical part will be different, it's also about how to solve for the emotional part.
Speaker 3: It's helpful to see where you think there might be early traction here. In the panel description, we cited a market stat that said this could be a $200 billion market by 2030. Jason, in your perspective, is this something, you know, we're seeing traction in some places right now more so than others. Is this cannibalizing existing e-commerce? Is this net new value? How are you thinking about some of the early parallels that we might have with agenda commerce to e-commerce? kind of building off of the traction point that Kai just mentioned.
Speaker 2: I mean, I think there's a lot of similarities in that it's a brand new way to shop, brand new way to interact with commerce. And if you look at how the world has changed as e-commerce rose, I think people still go to retail stores. And the over I don't think anybody would would contest that the overall market of both retail and e-commerce together is much larger than when it was just retail. And we would expect the same here in that. We'd expect the overall pie to grow. Some of that will be net news, some of that will be camelization, but I think the overall trajectory will definitely be.
Speaker 1: Yeah. I think the sta think you're quoting Morgan Stanley. If you believe McKinsey, they quote five trillion of a gentic by 2030. I'm probably more in their camp. I think it's gonna be some more substantially larger. And to your point of growing the total pie, like we and I'm sure there's some panel here that talks about the impact of AI on coding and what is it does to software and s what does it do to software and so on. It replaces other functions that you will purchase, back to your point about around procurement. That currently we don't think about them and we have people do them. And then when an agent will pay either another agent or a human, right? If you remember rent a human just from a couple weeks ago, that will also count because these transactions will happen autonomously and that'll think that'll grow the payment and the gentic commerce as we would define it substantially.
Speaker 3: Okay, so let's assume that we've got this huge market, somewhere in the range of two hundred billion to five trillion that we're talking about. Where does value accrue in the stock here?
Speaker 1: well if you go back to the nineties as e-commerce started to appear, right in the conversations where I I still remember in maybe not in the nineties, when 2015 or something, I were talking to large retailers and they would treat e-commerce as, yeah, yeah, we have this dot-com thing because everybody had need needed dot com, but we don't really care about this. And now everybody cares about this. It's part of your strategy. I think agentic is gonna be part of the strategy of pretty much every business that wants to sell anything. And where value I think that similar ways where value accrues today, so discoverability, whoever owns the consumer's way to initiate transactions and whoever owns the rails, I also think the biggest value accrual will be on the consumers. You'll get better product, better ROI, right, not necessarily cheaper, and substantially more convenient.
Speaker 4: I would say also in the it's still a bit early to tell what will be the pivotal part in the ecosystem for the future agenda commerce. You know, in e-commerce world, apparently the winner was the marketplace for e-commerce. They are like the most critical part in the e-commerce world. And in agenda commerce, whether in the future it becomes like, you know, the foundation model provider to provide the agents, et cetera, like you know, A lot of the decision making is going to be held by the agents, not the human beings. Or, like, you know, it is going to be the merchants, or it can be like those protocol holders to decide like how the agents and the merchants, merchants' agents can like you know work with each other. but one trend I can see very clearly is like a lot of those ecosystem participants they are moving, trying to move quite fast. From an ERP perspective, you already see a lot of AI native ERP trying to normally From startups or like you know existing giants that they are working towards those AI native protocols. And then also for a lot of merchants, I just talked with Michael before the stage, like you know, a lot of merchants they are also trying to get them ready to become a genetic commerce native. And then also for financial infrastructure. As a financial entertainment service provider, or like a bunch of card schemes like Visa MasterCard, they are all like playing in the in this direction. And I guess needless to mention, like you know, OpenAI, like a lot of those foundation models, they are pushing that as well. So we see like a lot of those ecosystem partners, they are moving to this direction. There might be a single singular point where like you know all those things click, and then you know it really creates 10 times better user experience for existing. consumers or like B2B buyers and then they will switch to this new experience.
Speaker 3: I think one of the the undercurrents for all of this is also an aspect of trust that needs to happen for the maturation of the industry for the sector. I mean, Michael, this is kind of I'm gonna look directly at you for the for this one. How do you think about trust in a world of agentic commerce? How do you be are you able to discern good agents from bad agents and malicious actors?
Speaker 1: So trust is a fundamental component of every e-commerce transaction, right? And if you go back to the small villages, you knew who you were buying from. Now we have a big global village, or at least a few of them. And you don't necessarily need to know the person to be able to trust them, but you need some establishment of trust for that to happen. And now you will likely have an agent or a chain of agents needing to trust another agent on the other side. And trust is not just a matter of identity, right? So identity and trust maybe two. two are two separate things are potentially needed for one is potentially needed for the other, but not necessarily. You need to understand currently, right, where we have businesses transacting with consumers, right? Or in your case, like businesses transacting with Businesses but humans that are initiating that. And for regulatory purposes, you need to know who that is. And I think that will shift first for the identity question. Is it a customer that I already know? Do I want to give them the same, I don't know, loyal use the same loyalty account, even if they're using through an agent, right? So that's an identity question. And a trust question is maybe it's a person I know and it's a loyal customer, but they're coming through an agent that I don't want transacting on my behalf. Let's say it's a competitor's agent that in it wants to use all of their competitors. As an inventory and retain the customer over there for whatever purposes. And there are multiple types of definitions of who you want to let in for what purpose. Or I think when we're talking to our customers or just to our partners in the industry, first question is, let's make sure that it works, because currently a lot of it doesn't. Second question is, let's make sure I don't go bankrupt because something bad happens. So a stat, I checked the stats on our system this morning. over 12% of all transactions since midnight till well, 9 a.m. or so when I checked, were fraudulent. Right? So to your question of bad actors and not. And out of all of e-commerce, one to three percent are fraudulent depending on on the vertical. So you're seeing substantially more currently, substantially more agentic transactions that are fraudulent because fraudsters are earlier adopters of technology. It's super scalable, so they can go and do it, and consumers are laggards, or a lot of them. I think early on. When did you you launched GenTIC in some January, right? And we started monitoring monitoring it. It was 100% open AI. Then I think Perplexity launched second, so you and they launched like a slightly better experience, so it's shifted to 90% perplexity, 10% open AI, every new protocol that shifts. And then you start that seeing fraudsters pick up and be over fifty percent of volume So it's a very incredibly critical thing to get done because you g if you get it wrong, you lose all of your money. And that's true by the way for the banks as well. I don't know if you see it in the B2B side, but banks are very afraid because it's not it's not clear who who owns the liability. Do they go after OpenAI if they sent them the wrong transaction, or do they go after the consumer who they can't even know whether they done the transaction at all or initiated a transaction at all?
Speaker 3: Two good points. Does it work and how do I make sure that it doesn't bankrupt me? I like that. Guiding principles. Jason, what would you add to the conversation here from your program?
Speaker 2: I, you know, this question about how quickly agentic commerce will be adopted, I think trust is actually what the real bottleneck is gonna be. So from a technology standpoint, we're not too far off from being able to accomplish this effectively. As a merchant, all you really need is a product feed. You need an MCP server that allows an agent to connect to that and to see what products are on sale or on display. And the models will get better at managing context, they'll get better at navigating and making decisions, and those capabilities will play out across all kinds of tasks, not just shopping. So definitely a focus for every major AI lab. Once you have the technology in place, I do think this element of trust, exactly what was described, is gonna be the limiting factor or the deciding factor in terms of how quickly this is advanced.
Speaker 4: I I quite agree to to Jason. Actually, if you think about the trust framework, it exists quite well already in the current commercial world. so I I just want to add two points maybe. So one is like you need to allow those error correction mechanisms to exist. So when agents make an error Or something you don't want to happen, you are able to correct it, then you know that fortify the trust because you are it's not like you know a one-way decision, it is irreversible. So, for example, in a lot of payments world, you have chargeback. So means like if agents like you know, create a transaction you don't want to happen, you can charge back and reverse the transaction. And the other point is you need to have some hard gate or like you know, deterministic guard rate. So, for example, like if you use a wallet as the financial infrastructure for a genetic payments, then basically you can set the hard gate, like you know, the budget. For example, like this wallet cannot exceed a certain budget of the payments or of a certain kind of transfer transaction, then you can control like you know how much damage, if some error really happened, that also helps you to build the trust in this system. Or you can be very specific about like you for this. Wallet or like you know for this a wallet belongs to a certain agent, it can only spend on certain category of the merchants. This is also already supported by the current technology. And by applying those things into the genet commerce, I believe that yeah, trust shouldn't be a main issue if we just work on this direction.
Speaker 3: I think stable coins get brought into the conversation a lot here too. Do you have any thoughts on how stable coins fit into that overall piece of the puzzle for what we need from a financial instrument perspective?
Speaker 4: I I think like stable coin don't really provide fundamental I think three factors that I mentioned about wallet. One is like you know the tryback still doesn't exist. We make a transfer to like you know crypto, basically it is immutable. You are not able to reverse it very easily. And two is actually the regulatory compliance. So still there it points a lot of I would say a challenge from the regulator for stable coin or crypto for the commerce world. It's my not not to not to mention a genetic commerce, even just for the current e-commerce, the penetration of stable coin is still pretty low. but it has a lot of value in exotic countries. Means like in a lot of countries where their currency is unstable, stable coin definitely plays a very important role there in the transactions. And thirdly, it's basically you are very difficult to set Those hard gates or like you know, hard guard rails for a stable coin. As like you know, what the technology already allowed. In the current kind of a transaction infrastructure or payments infrastructure. but actually, you know, now there is more compatibility. So there are also like stable coin funded cars exist. So actually it can you know have a lot of middle ground if the purpose is simply to support stablecoin. You can also have like you know, stable coin funded wallet, it can serve like you know similar purposes. It's only like you know, if you look at the fundamental transaction media. I will still have some question about stablecoin as like you know the most fundamental infrastructure. I would say like you know wallet so far as I can see is might be easier to accomplish and better use experience.
Speaker 3: It's helpful. All right. I I'm gonna share a little bit about me personally, which is I think I've experienced two years worth of AI change with within one month. I don't know if anyone else feels that way. So Jason, I'd like to get your perspective on where the future of model development is heading because I think part of this is if we're thinking about agenda commerce tied to model development, what are those changes that we can expect to see in the next six to twelve months or even one to two months maybe at this rate that we're experiencing that we should be prepared for with this future.
Speaker 2: I I think regarding agentic commerce, the models are actually pretty close to being able to accomplish a lot of what's required. what I will say is that the res release cycle for new AI models has shortened dramatically. So as a result of all the capital in the space, that attracts a ton of talent. It attracts a ton resources. so this year, OpenAI has released two new AI models. It's GPT-5.3 and GPT-5.4. It's been two months into the year. For context, it used to take us two and a half years to to release one new version of this. So if you look at GPT-3 to GPT-3.5, that was a two and a half year wait. GPT 3.5 to GPT-4, I believe, was 18 months on the order of that timeframe. And now we're seeing these improvements happen much quicker, right? Essentially a four to six week release cycle for open AI. So there's a bunch of capabilities that will play into a GentaCommerce. multimodal would be one, right? Being able to understand images and video. But the overall sentiment I would present is that the capability of these models is increasing quickly. and we'd expect that on on almost a monthly, four to four to six week cadence as opposed to what we had before, which was a year.
Speaker 3: We've mentioned discoverability, I think, at various points throughout the conversation too. I'd like to double click into it a bit because I don't know if anyone really knows how you optimize for an agent just yet, and that's likely gonna change quite a bit over time. I'll anyone on the panel have thoughts about this, but how should we be thinking about optimizing for agents?
Speaker 1: Well, I'll I'll start. I think one of the biggest reasons, in my view, that the consumer will be a big beneficiary of that is they'll be able to process substantially more information. A lot of the time when we buy today, we go to brands because we ha we put a lot of trust in this brand. Right? If I buy from this company, it's probably a higher quality product. We don't necessarily vet this. Humans are very bad decision at decision making. And I think agents will make it substantially better decision given the information. So the first thing that brands need to do or businesses need to do is making sure they have as much information out there accessible to agents to process. Consumer experience changes, right? If when you buy something online, you want it as quickly as possible. I want this product, click here, ship it to me, don't want to deal with you anymore. And from an agent perspective, you can ask for more information. Yeah, I don't know, if you're buying, it would be LA specific, you're buying a beauty product, you want given the science behind it, not just the pretty pictures. And I want to compare it to I don't know, my Medical test results from before, and it can take 20 seconds or 20 minutes, it doesn't really matter. The more information that's gonna be out there, an agent will be able to make a better decision, more personalized decision with higher value. And it's not how we're set up now. A lot of the discoverability is picture driven and kind of Google SEO-driven. and I think that's gonna be completely changed.
Speaker 2: Yeah, I think what I'd add to that, those are great points. I'm not enough of an expert to understand the specifics around how discoverability will be addressed, but I think the overall sentiment is just what an exciting opportunity. I mean if you think about shopping transactions today, most of those transactions occur because an ad is served. Maybe I go on Google, I do a search, I click on one of the top three links, those are likely sponsored. and it takes me to a new web page. And now there's gonna be this whole new intermediary. Almost like a sis an assistant for me, who is doing all this research on my behalf. It completely changes every chain of how that shopping decision is made. And so without getting into specifics of how SEO might change, I think the sentiment I would have is just what an interesting opportunity to rethink how to interact with agents at every level of the commerce experience.
Speaker 4: Yeah, I have an interesting story to to share. I so I basically vibe code an agent, like you know, the supply chain so-called supply chain is actually a procurement agent to help me buy some buy some wines basically. And then also on the other hand, like you know, we have of those white merchants there. And I really see like my agent like an overnight, just talking to a merchant like you know for kind of a $20 discount overnight, like you know, trying to bargain this and that. So what I'm trying to say is and because like my prompt or like you know mark out real to him to to the to the agent is like you know to find me the best deal instead like you know strike the deal. So what I want what what I try to the point that I'm trying to make is in The genet commerce world, I guess the autonomous reasoning is very, very important. So if there is going to be a lot of human intervention in the experience, so far I see it might not be as as fun, especially for a shopping experience as like you know the current e-commerce, honestly. But like if we think about a world, like you know, it is a really you delegate a lot of authority to the agent, and then you know, it is a purely kind of an autonomous. autonomous reasoning, to Michael's point, like from end to end, I guess like things, the way we think about commerce might fundamentally change.
Speaker 3: This has been a fantastic panel. I have one more question for you all. I think there might be a lot of folks here who are wondering, okay, this is great and all, but what should I do about it? So if you had to leave one piece of advice to say, here's your action for the next 90 days, here's how I'd be thinking about it, what would it be?
Speaker 1: I think most of the audience here are investors, so they'll need to understand like I I think it will dramatically change who are the winners and losers in a broader ecosystem. so you'll probably need to revisit your thesis on things. If there are businesses here that are trying to sell stuff, making sure that your just your products are discoverable, your stacks ready, who are part your partners to be able to help you be there right now because it's happening now.
Speaker 2: Yeah, I agree. I mean it for the merchants, make sure your products are discoverable and understand what that means and what that would require.
Speaker 4: Yeah, I would just suggest like you know merchants to be very open minded and really rethink the user experience. It's not like you know something just slightly changed to the current experience. It's going to be a fundamental revolution.
Speaker 3: Amazing. Well thank you so much for this wonderful panel.
Speaker 1: Thank you.
Speaker 2: Thank you so much.
Speaker 2: Yeah, so I think agentic commerce can have a broad definition. I actually starting by defining even what an agent is in the AI world. An agent is very simple. It's an AI model that uses tools and has guardrails, instructions, and it achieves something in the real world. That's really all that definition means. And this could be a variety of different tasks. When we think about in the commerce world, there's this end state that we imagine where agents armed with context, Able to navigate the world, we'll be able to handle every piece of the shopping chain, of the commerce chain, I should say. That includes discovery, it includes evaluation, the checkout, and potentially post-checkout with returns and other components there, other steps there. But you can imagine a world where agents are almost an avatar of each individual. They're able to act in a way that helps accomplish commerce works. What I will say is that that is that's the end state of where all this might lead. I do think there's this huge gray area middle ground, which is every incremental step towards that. And so that's discovery powered by AI, but humans are making the final decision. And and and we may have different words for that, but I do think in our vernacular, in my vernacular, think of that all as agentic commerce. There's just different pieces of how we'll how we'll accomplish that. And there's different commercial motions that will still take take different approaches.
Speaker 3: Michael, is there anything that you might want to add from the merchant network data side?
Speaker 1: Well For context, right now, probably less than 10 basis points of all transactions are done agentic. Now, our definition would be a transaction that happens end-to-end without any human involvement, right? So if it's just discovery and then you have to complete it, we wouldn't define it as agentic. So maybe for that, in some verticals, you may go up to 10 or 20% in terms of discoverability. but for actual transactions happening, it's a very, very slim percentage of transactions, and it's probably growing about 20% months over months. And it's definitely top of mind of Pretty much any merchant we're talking to. I actually checked right this moment. We're having a webinar with dozens of merchants about this. it's it's top of mind across different functions at large companies, right? Because it will impact how their business is gonna run, right? Where are they spending marketing dollars, where are they s what do they need to do in their tech stack to advance it, and probably a lot of other things we'll cover in the twenty-two minutes we have.
Speaker 3: So I mean I think there's a a little bit here which is a a question around is this something that's happening right now that we need to be thinking about or is this something that's gonna play out for years in the future? So you kind of spoke to it a little bit, but where is your head at right now? This is very much a moment in time we need to capture, or maybe this will take a a couple years to play out fully and it needs some time to mature.
Speaker 1: I think it's gonna happen substantially faster than people anticipated. a lot of the customers we talked to expect it to happen. And part of the problem here is if you're not ready as a merchant, it'll just happen over you and you'll be it'll be very, very hard to correct. I did an exercise maybe in July. I needed a new Whatever, not an AC because I live in London, but like some cooling mechanism. And I was working with Chat2PT trying to understand what I can get. A lot of the merchants were blocking all agency traffic into them. So just the discoverability side, right? Because they're all have multiple bot detection technologies. And when the information came back, only two sent the like w were able to send this the model, this is the price, this is what you want. And I had to pick between them. Then I met the next day, I met the CEO of one of those. companies like they you're blocking all the genetic traffic, you're not gonna be discoverable. It took them about a week and they fixed it all. Like it's it's happening right now and and people know they need to get on it.
Speaker 4: I would like to add that forget commerce, fundamentally procurement and shopping are quite different things. because procurement is typically more logical than shopping. Shopping experience involves a lot of emotional decision. It's not like you know, always just optimize for like you know the best outcome from a bunch of parameters. While procurement can be much more logical on one hand, and So on the other hand, the data environment is better usually than like you know Shopping experience, because shopping experience, a lot of data actually exists in the open web, sometimes even just social media, right? So people's kind of a comments, feedback about a certain product. While like you know, procurement data can be much more structured, exist in the ERP, exist like you know in a bunch of APIs, machine to machine. So I believe like if talking about which might come first, like no matter for B2B or B2C use cases, when it comes comes to the category of procurement, because for B2C you can also have like those grocery procurement. For example, someone some agent or AI monitoring your like you know refrigerator, whatever, like you know, decide like on whether you need to backfill something, it might be easier. But when it comes to shopping experiences, not only like the logical part will be different, it's also about how to solve for the emotional part.
Speaker 3: It's helpful to see where you think there might be early traction here. In the panel description, we cited a market stat that said this could be a $200 billion market by 2030. Jason, in your perspective, is this something, you know, we're seeing traction in some places right now more so than others. Is this cannibalizing existing e-commerce? Is this net new value? How are you thinking about some of the early parallels that we might have with agenda commerce to e-commerce? kind of building off of the traction point that Kai just mentioned.
Speaker 2: I mean, I think there's a lot of similarities in that it's a brand new way to shop, brand new way to interact with commerce. And if you look at how the world has changed as e-commerce rose, I think people still go to retail stores. And the over I don't think anybody would would contest that the overall market of both retail and e-commerce together is much larger than when it was just retail. And we would expect the same here in that. We'd expect the overall pie to grow. Some of that will be net news, some of that will be camelization, but I think the overall trajectory will definitely be.
Speaker 1: Yeah. I think the sta think you're quoting Morgan Stanley. If you believe McKinsey, they quote five trillion of a gentic by 2030. I'm probably more in their camp. I think it's gonna be some more substantially larger. And to your point of growing the total pie, like we and I'm sure there's some panel here that talks about the impact of AI on coding and what is it does to software and s what does it do to software and so on. It replaces other functions that you will purchase, back to your point about around procurement. That currently we don't think about them and we have people do them. And then when an agent will pay either another agent or a human, right? If you remember rent a human just from a couple weeks ago, that will also count because these transactions will happen autonomously and that'll think that'll grow the payment and the gentic commerce as we would define it substantially.
Speaker 3: Okay, so let's assume that we've got this huge market, somewhere in the range of two hundred billion to five trillion that we're talking about. Where does value accrue in the stock here?
Speaker 1: well if you go back to the nineties as e-commerce started to appear, right in the conversations where I I still remember in maybe not in the nineties, when 2015 or something, I were talking to large retailers and they would treat e-commerce as, yeah, yeah, we have this dot-com thing because everybody had need needed dot com, but we don't really care about this. And now everybody cares about this. It's part of your strategy. I think agentic is gonna be part of the strategy of pretty much every business that wants to sell anything. And where value I think that similar ways where value accrues today, so discoverability, whoever owns the consumer's way to initiate transactions and whoever owns the rails, I also think the biggest value accrual will be on the consumers. You'll get better product, better ROI, right, not necessarily cheaper, and substantially more convenient.
Speaker 4: I would say also in the it's still a bit early to tell what will be the pivotal part in the ecosystem for the future agenda commerce. You know, in e-commerce world, apparently the winner was the marketplace for e-commerce. They are like the most critical part in the e-commerce world. And in agenda commerce, whether in the future it becomes like, you know, the foundation model provider to provide the agents, et cetera, like you know, A lot of the decision making is going to be held by the agents, not the human beings. Or, like, you know, it is going to be the merchants, or it can be like those protocol holders to decide like how the agents and the merchants, merchants' agents can like you know work with each other. but one trend I can see very clearly is like a lot of those ecosystem participants they are moving, trying to move quite fast. From an ERP perspective, you already see a lot of AI native ERP trying to normally From startups or like you know existing giants that they are working towards those AI native protocols. And then also for a lot of merchants, I just talked with Michael before the stage, like you know, a lot of merchants they are also trying to get them ready to become a genetic commerce native. And then also for financial infrastructure. As a financial entertainment service provider, or like a bunch of card schemes like Visa MasterCard, they are all like playing in the in this direction. And I guess needless to mention, like you know, OpenAI, like a lot of those foundation models, they are pushing that as well. So we see like a lot of those ecosystem partners, they are moving to this direction. There might be a single singular point where like you know all those things click, and then you know it really creates 10 times better user experience for existing. consumers or like B2B buyers and then they will switch to this new experience.
Speaker 3: I think one of the the undercurrents for all of this is also an aspect of trust that needs to happen for the maturation of the industry for the sector. I mean, Michael, this is kind of I'm gonna look directly at you for the for this one. How do you think about trust in a world of agentic commerce? How do you be are you able to discern good agents from bad agents and malicious actors?
Speaker 1: So trust is a fundamental component of every e-commerce transaction, right? And if you go back to the small villages, you knew who you were buying from. Now we have a big global village, or at least a few of them. And you don't necessarily need to know the person to be able to trust them, but you need some establishment of trust for that to happen. And now you will likely have an agent or a chain of agents needing to trust another agent on the other side. And trust is not just a matter of identity, right? So identity and trust maybe two. two are two separate things are potentially needed for one is potentially needed for the other, but not necessarily. You need to understand currently, right, where we have businesses transacting with consumers, right? Or in your case, like businesses transacting with Businesses but humans that are initiating that. And for regulatory purposes, you need to know who that is. And I think that will shift first for the identity question. Is it a customer that I already know? Do I want to give them the same, I don't know, loyal use the same loyalty account, even if they're using through an agent, right? So that's an identity question. And a trust question is maybe it's a person I know and it's a loyal customer, but they're coming through an agent that I don't want transacting on my behalf. Let's say it's a competitor's agent that in it wants to use all of their competitors. As an inventory and retain the customer over there for whatever purposes. And there are multiple types of definitions of who you want to let in for what purpose. Or I think when we're talking to our customers or just to our partners in the industry, first question is, let's make sure that it works, because currently a lot of it doesn't. Second question is, let's make sure I don't go bankrupt because something bad happens. So a stat, I checked the stats on our system this morning. over 12% of all transactions since midnight till well, 9 a.m. or so when I checked, were fraudulent. Right? So to your question of bad actors and not. And out of all of e-commerce, one to three percent are fraudulent depending on on the vertical. So you're seeing substantially more currently, substantially more agentic transactions that are fraudulent because fraudsters are earlier adopters of technology. It's super scalable, so they can go and do it, and consumers are laggards, or a lot of them. I think early on. When did you you launched GenTIC in some January, right? And we started monitoring monitoring it. It was 100% open AI. Then I think Perplexity launched second, so you and they launched like a slightly better experience, so it's shifted to 90% perplexity, 10% open AI, every new protocol that shifts. And then you start that seeing fraudsters pick up and be over fifty percent of volume So it's a very incredibly critical thing to get done because you g if you get it wrong, you lose all of your money. And that's true by the way for the banks as well. I don't know if you see it in the B2B side, but banks are very afraid because it's not it's not clear who who owns the liability. Do they go after OpenAI if they sent them the wrong transaction, or do they go after the consumer who they can't even know whether they done the transaction at all or initiated a transaction at all?
Speaker 3: Two good points. Does it work and how do I make sure that it doesn't bankrupt me? I like that. Guiding principles. Jason, what would you add to the conversation here from your program?
Speaker 2: I, you know, this question about how quickly agentic commerce will be adopted, I think trust is actually what the real bottleneck is gonna be. So from a technology standpoint, we're not too far off from being able to accomplish this effectively. As a merchant, all you really need is a product feed. You need an MCP server that allows an agent to connect to that and to see what products are on sale or on display. And the models will get better at managing context, they'll get better at navigating and making decisions, and those capabilities will play out across all kinds of tasks, not just shopping. So definitely a focus for every major AI lab. Once you have the technology in place, I do think this element of trust, exactly what was described, is gonna be the limiting factor or the deciding factor in terms of how quickly this is advanced.
Speaker 4: I I quite agree to to Jason. Actually, if you think about the trust framework, it exists quite well already in the current commercial world. so I I just want to add two points maybe. So one is like you need to allow those error correction mechanisms to exist. So when agents make an error Or something you don't want to happen, you are able to correct it, then you know that fortify the trust because you are it's not like you know a one-way decision, it is irreversible. So, for example, in a lot of payments world, you have chargeback. So means like if agents like you know, create a transaction you don't want to happen, you can charge back and reverse the transaction. And the other point is you need to have some hard gate or like you know, deterministic guard rate. So, for example, like if you use a wallet as the financial infrastructure for a genetic payments, then basically you can set the hard gate, like you know, the budget. For example, like this wallet cannot exceed a certain budget of the payments or of a certain kind of transfer transaction, then you can control like you know how much damage, if some error really happened, that also helps you to build the trust in this system. Or you can be very specific about like you for this. Wallet or like you know for this a wallet belongs to a certain agent, it can only spend on certain category of the merchants. This is also already supported by the current technology. And by applying those things into the genet commerce, I believe that yeah, trust shouldn't be a main issue if we just work on this direction.
Speaker 3: I think stable coins get brought into the conversation a lot here too. Do you have any thoughts on how stable coins fit into that overall piece of the puzzle for what we need from a financial instrument perspective?
Speaker 4: I I think like stable coin don't really provide fundamental I think three factors that I mentioned about wallet. One is like you know the tryback still doesn't exist. We make a transfer to like you know crypto, basically it is immutable. You are not able to reverse it very easily. And two is actually the regulatory compliance. So still there it points a lot of I would say a challenge from the regulator for stable coin or crypto for the commerce world. It's my not not to not to mention a genetic commerce, even just for the current e-commerce, the penetration of stable coin is still pretty low. but it has a lot of value in exotic countries. Means like in a lot of countries where their currency is unstable, stable coin definitely plays a very important role there in the transactions. And thirdly, it's basically you are very difficult to set Those hard gates or like you know, hard guard rails for a stable coin. As like you know, what the technology already allowed. In the current kind of a transaction infrastructure or payments infrastructure. but actually, you know, now there is more compatibility. So there are also like stable coin funded cars exist. So actually it can you know have a lot of middle ground if the purpose is simply to support stablecoin. You can also have like you know, stable coin funded wallet, it can serve like you know similar purposes. It's only like you know, if you look at the fundamental transaction media. I will still have some question about stablecoin as like you know the most fundamental infrastructure. I would say like you know wallet so far as I can see is might be easier to accomplish and better use experience.
Speaker 3: It's helpful. All right. I I'm gonna share a little bit about me personally, which is I think I've experienced two years worth of AI change with within one month. I don't know if anyone else feels that way. So Jason, I'd like to get your perspective on where the future of model development is heading because I think part of this is if we're thinking about agenda commerce tied to model development, what are those changes that we can expect to see in the next six to twelve months or even one to two months maybe at this rate that we're experiencing that we should be prepared for with this future.
Speaker 2: I I think regarding agentic commerce, the models are actually pretty close to being able to accomplish a lot of what's required. what I will say is that the res release cycle for new AI models has shortened dramatically. So as a result of all the capital in the space, that attracts a ton of talent. It attracts a ton resources. so this year, OpenAI has released two new AI models. It's GPT-5.3 and GPT-5.4. It's been two months into the year. For context, it used to take us two and a half years to to release one new version of this. So if you look at GPT-3 to GPT-3.5, that was a two and a half year wait. GPT 3.5 to GPT-4, I believe, was 18 months on the order of that timeframe. And now we're seeing these improvements happen much quicker, right? Essentially a four to six week release cycle for open AI. So there's a bunch of capabilities that will play into a GentaCommerce. multimodal would be one, right? Being able to understand images and video. But the overall sentiment I would present is that the capability of these models is increasing quickly. and we'd expect that on on almost a monthly, four to four to six week cadence as opposed to what we had before, which was a year.
Speaker 3: We've mentioned discoverability, I think, at various points throughout the conversation too. I'd like to double click into it a bit because I don't know if anyone really knows how you optimize for an agent just yet, and that's likely gonna change quite a bit over time. I'll anyone on the panel have thoughts about this, but how should we be thinking about optimizing for agents?
Speaker 1: Well, I'll I'll start. I think one of the biggest reasons, in my view, that the consumer will be a big beneficiary of that is they'll be able to process substantially more information. A lot of the time when we buy today, we go to brands because we ha we put a lot of trust in this brand. Right? If I buy from this company, it's probably a higher quality product. We don't necessarily vet this. Humans are very bad decision at decision making. And I think agents will make it substantially better decision given the information. So the first thing that brands need to do or businesses need to do is making sure they have as much information out there accessible to agents to process. Consumer experience changes, right? If when you buy something online, you want it as quickly as possible. I want this product, click here, ship it to me, don't want to deal with you anymore. And from an agent perspective, you can ask for more information. Yeah, I don't know, if you're buying, it would be LA specific, you're buying a beauty product, you want given the science behind it, not just the pretty pictures. And I want to compare it to I don't know, my Medical test results from before, and it can take 20 seconds or 20 minutes, it doesn't really matter. The more information that's gonna be out there, an agent will be able to make a better decision, more personalized decision with higher value. And it's not how we're set up now. A lot of the discoverability is picture driven and kind of Google SEO-driven. and I think that's gonna be completely changed.
Speaker 2: Yeah, I think what I'd add to that, those are great points. I'm not enough of an expert to understand the specifics around how discoverability will be addressed, but I think the overall sentiment is just what an exciting opportunity. I mean if you think about shopping transactions today, most of those transactions occur because an ad is served. Maybe I go on Google, I do a search, I click on one of the top three links, those are likely sponsored. and it takes me to a new web page. And now there's gonna be this whole new intermediary. Almost like a sis an assistant for me, who is doing all this research on my behalf. It completely changes every chain of how that shopping decision is made. And so without getting into specifics of how SEO might change, I think the sentiment I would have is just what an interesting opportunity to rethink how to interact with agents at every level of the commerce experience.
Speaker 4: Yeah, I have an interesting story to to share. I so I basically vibe code an agent, like you know, the supply chain so-called supply chain is actually a procurement agent to help me buy some buy some wines basically. And then also on the other hand, like you know, we have of those white merchants there. And I really see like my agent like an overnight, just talking to a merchant like you know for kind of a $20 discount overnight, like you know, trying to bargain this and that. So what I'm trying to say is and because like my prompt or like you know mark out real to him to to the to the agent is like you know to find me the best deal instead like you know strike the deal. So what I want what what I try to the point that I'm trying to make is in The genet commerce world, I guess the autonomous reasoning is very, very important. So if there is going to be a lot of human intervention in the experience, so far I see it might not be as as fun, especially for a shopping experience as like you know the current e-commerce, honestly. But like if we think about a world, like you know, it is a really you delegate a lot of authority to the agent, and then you know, it is a purely kind of an autonomous. autonomous reasoning, to Michael's point, like from end to end, I guess like things, the way we think about commerce might fundamentally change.
Speaker 3: This has been a fantastic panel. I have one more question for you all. I think there might be a lot of folks here who are wondering, okay, this is great and all, but what should I do about it? So if you had to leave one piece of advice to say, here's your action for the next 90 days, here's how I'd be thinking about it, what would it be?
Speaker 1: I think most of the audience here are investors, so they'll need to understand like I I think it will dramatically change who are the winners and losers in a broader ecosystem. so you'll probably need to revisit your thesis on things. If there are businesses here that are trying to sell stuff, making sure that your just your products are discoverable, your stacks ready, who are part your partners to be able to help you be there right now because it's happening now.
Speaker 2: Yeah, I agree. I mean it for the merchants, make sure your products are discoverable and understand what that means and what that would require.
Speaker 4: Yeah, I would just suggest like you know merchants to be very open minded and really rethink the user experience. It's not like you know something just slightly changed to the current experience. It's going to be a fundamental revolution.
Speaker 3: Amazing. Well thank you so much for this wonderful panel.
Speaker 1: Thank you.
Speaker 2: Thank you so much.
