On Ep. 1 of Agentic Commerce, Simon Taylor, GTM @ Tempo, and Bam Azizi, CEO & Founder @ Mesh are joined by Jeff Weinstein, Product Lead @ Stripe and Steve Kaliski, Principal Software Engineer @ Stripe to discuss the role of AI agents in the future economy and more!
On Ep. 1 of Agentic Commerce, Simon Taylor, GTM @ Tempo, and Bam Azizi, CEO & Founder @ Mesh are joined by Jeff Weinstein, Product Lead @ Stripe and Steve Kaliski, Principal Software Engineer @ Stripe to discuss the role of AI agents in the future economy 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
Speaker 1 00:00
Then about a year ago, we started working with open AI to help them think through an idea around how would you be able to purchase directly within chatgpt. You're already using it to prompt and to discover, and if you go to any retailer, or even the stripe business, and you look at the refer traffic, where are my customers coming from? They are dramatically increasingly coming from these AI platforms. And so it's natural to try to move the execution of a purchase closer to its intention, and the intention is starting to move to these AI platforms.
Sy Taylor 00:42
So welcome to tokenized. The show focused on stable coins and the institutional adoption of tokenized real world assets. My name is Simon Taylor. I am your host for today, author at FinTech brain food, and head of market dev over at tempo. And we've got something a little bit different. We're going to jump into all things agentic commerce and agentic finance. And to do that in this mini series, is the CEO and founder of mesh and a returning guest now my co host, BAM Azizi, how are you, sir?
Bam Azizi 01:13
Good, good. It's a pleasure being with you, Simon. I appreciate it. Now.
Sy Taylor 01:17
Appreciate you wanting to do the agentic commerce space. I think at some point we should get into your history and why you find that so interesting. You've got a strong background there, but we have a lot to explore in this series. Can you tell us a little bit about why you wanted to get into agentic commerce given your history, you know, best known now for mesh and stable coins, yeah, 100%
Bam Azizi 01:38
I think the agentic commerce is going to be the future of commerce, and even bigger, is going to be the future of finance. The reason is we will have limited number of people connected to internet, like maybe around 5 billion. We're going to have 500 billion agents interacting on behalf of us and on behalf of themselves. So they're going to accelerate the future economy 1,000x faster than Internet. Say, think you're the beginning of that, and we're just scratching the surface. And it worth the time to share our insights, and also bringing experts, like always, you do the best of the best to share their insights so people can know what's going to happen the next five to 10
Sy Taylor 02:19
years. Well, and as a primer, we do have possibly the perfect guest for today's show. First up is Jeff Weinstein, product lead at stripe, and prolific poster on x. How are you, sir? Good.
Speaker 1 02:32
Yeah. My my Twitter profile, somewhat is smarter than I am, so we'll see how it's tested today.
Sy Taylor 02:38
Thank you so much for joining us, Jeff, and of course, joining you as your buddy cop duo, as always, is Steve kalisky, who's principal software engineer at stripes. Good evening. How are you, sir, good thanks for having us. All right. Just before I get into the show, I've got to remind everybody that views and opinions of contributors today might be their own and might not reflect those of the companies they represent, and please don't take anything we say is tax, legal, financial, or any other form of advice. And of course, to thank mesh for making this series possible and our good friends over at visa. So I'm going to start us off with just a zoom all the way out primer. What does agentic commerce mean? Because I could ask 10 different people and get 10 different answers, but Jeff, you've probably been wrestling with this as a practitioner for longer than most. So what do you view as agentic commerce in terms of the primary payments flows and use cases? What does it look like to a customer? Yeah,
Speaker 1 03:38
so at stripe, we're building economic infrastructure for the internet. And if you've used your computer in the last two years, a lot of the internet is sort of AI filled, or AI mediated, or AI assisted in ways that felt like science fiction 10 years ago. And so if you instead of using a search engine to see 10 blue links, you might ask a question. And so the use of AI in an applied way is just been deployed on society, for consumers and business at an incredibly rapid rate. And we see that in in all the stripe data. And so then, you know, we were starting to move from people using it as a read only tool, so I ask a question, I receive an answer. I have a little bit of follow up text comes back in a chat box to increasingly write activities. And you're starting to see that, obviously, hugely in software engineering, where AI can execute code, write code, deploy code, test code on your behalf. And we're starting to see it in more domains. And a domain that is of great interest to us in stripe is commerce, and how will AI be able to help people and businesses find what they need, be they products or services, actually be able to execute them more efficiently and have the foundations to build new types of shopping and business experiences that would have felt again like science fiction years ago? So that's sort of where we're coming from, and we can get into all the nitty gritty of agentic commerce and how it applies in each use case, but that's sort of like the high level. What got us excited about it
Sy Taylor 05:09
is moving from those 10 blue links and going to a website, creating a cart, hitting the Checkout button yourself. And there are so many different variations of that experience popping up in so many different ways and so many things being imagined. But Jeff, what I always enjoy about your ex posts is you're showing this in production, and I think that will be what surprises people, is that these payments flows are happening today. So just to double click on that, could you give me an example of something that's in production today.
Speaker 1 05:41
Yeah, so you know, when we started really thinking about this and building prototypes, 18 months ago, we really weren't sure what would be some of the first use cases. And then about a year ago, we started working with open AI to help them think through an idea around, how would you be able to purchase directly within chatgpt, you're already using it to prompt and to discover, and if you go to any retailer, or even the stripe business, and you look at the refer traffic, where are my customers coming from? They are dramatically increasingly coming from these AI platforms. And so it's natural to try to move the execution of a purchase closer to its intention. And the intention and the intention is starting to move to these AI platforms. So we set off to find a way to be able to have people buy within chat, GBT, and we got to pair with the great team open AI to do so, obviously. And there were a couple big problems to solve. One problem to solve was, how does a business express their checkout, right? You could imagine agent commerce being sort of adversarial towards a merchant or welcomed by a merchant. You know, there are good reasons for sort of both types of use cases, right? A browser automation could be very welcomed, where also an API call could also be welcomed, and we made the decision with them that we wanted to make this fantastic for merchants to be able to directly express their checkouts in a way that AI agents could buy sort of very directly. So that was one sort of main tenant of the design center. And then another one was, how do you make sure that any payment credential that a consumer wants to use is available inside of these AI platforms, but can be executed directly by the seller. So you're not actually buying directly from chat TBT. You're still buying from the merchant, but you're sort of mediated. You're assisted by the AI agent. And lastly, how do we make this something that can go from an idea to really scaling, and we want any type of merchant to be able to publish their checkout and their product catalog and all of their policies in a way that any AI agent is able to access with everyone's permission. And so we, again, with open AI co developed this agentic commerce protocol, which is an open standard for businesses to be able to express their checkouts, as well as a bunch of principles for how payments and credentials can work. And so those are some of the things we've tackled. And we got to launch that with chat GBT a few weeks ago, and Etsy was the first merchant and is already live. You can go on chatgpt and buy from a US Etsy merchant today. Shopify is rolling out the next few weeks, and Walmart just announced that they'll be selling through traffic GB as well. So we've gone from the whiteboard to really like in the browser, on your on your phone, much more quickly than I would have ever imagined. And while it's still early days, there is actually code working and transaction slowing.
Sy Taylor 08:37
You know, it's funny. Before we came to record this, I was coming back from London, and I was sitting on the Elizabeth line, on new high speed rail that gets me home. Thank goodness for the infrastructure. And next to me, I just happened upon, and I wasn't intending to peer over the shoulder of somebody, but somebody was using chat GPT to compare hair dice. And this is the standard consumer behavior now is like I go there first I have that natural language conversation. So of course, the checkout should be expressed there, but what you've done is you've moved the checkout to the experience, but without losing that merchant link, and bam, I know that point about being able to accept any payment type is very core to what mesh is about. How do you think about the different pros and cons from a merchant standpoint, in this brave new world,
Bam Azizi 09:27
merchants are like mostly are going to welcome because it increases the revenue for them, and they will enable their end users or customers to pay with one new type of payments. Our mission at mash is basically enabling anyone to pay with anything that they have. We're just facilitating that. We happen to feel like stable coin is going to be the future, because it's global. It doesn't have any border. It's easier to access. Anyone with the internet can have access to now another like financial service. Or financial tool. And I think agent tech would basically accelerate that need, because now all of a sudden, agents are interacting on behalf of us. Maybe you want to buy something from a merchant that only accepts a local currency in Philippines. How do you want to pay that? You want to pay with Visa card? You want to wire? I think stable coin has a lot of use cases there. I think at the end of the day, same as crypto agent e commerce, would have probably some pushback, but eventually all the merchants would enable that through, as Jeff mentioned, browser automation or API, and the one that explored API would benefit the most
Sy Taylor 10:40
explore away, for sure. Steve, talk to me about like, how this differs from some of the things you've had in the past. I know stripe has a number of products that had enabled virtual cards and things of that nature, as you were sort of finding your way to this. How is this different, mechanically, under the hood to some of those other instances?
Speaker 2 10:58
Yeah, sure. So I'll give you a real example of stripe isring. So we tend to think of you or me having credit cards that we use to buy our normal stuff, but virtual cards actually power a lot of B to B infrastructure. Imagine you're booking a flight, and you have a portal where you're picking your flight in a hotel, and you're paying there, and now, ironically, the travel agent, but maybe not real agent, is going off and booking the flight with a virtual card, and you're seeing a singular invoice across all the video of individual charges on the card that works well in these like very bounded, deterministic environments where there's one platform you're interacting with, a very finite list of hotels or airlines I'm interacting with, and then also you're in this context of you're already using a credit card, maybe, or some other payment method, but the default is that the seller wants to receive a credit card. I think what Jeff alluded to, that's important here is in this model where you have a buyer, an agent and a seller, everyone has their own independent preferences, and you can make the discovery really great of showing a product to purchase, but if the payment method I prefer, whether that's just a personal preference or a geographic requirement to preference, isn't available, that doesn't help with the conversion, right? And when you're the seller who's checkout, even if you've tuned it and written your own API for it being expressed in that application doesn't convert. Well, that doesn't help you, right? So being able to make sure that the payment methods that your buyers, maybe on your site prefer, and certainly the payment methods that the buyers at the agent application prefer, are available,
Speaker 1 12:36
is hugely important. I think it's especially important in agent Tai commerce that the plethora of the world's payment methods is available for all merchants, because in traditional sort of search experiences, you would get, you know, the first 10 blue links, and then there's a nice, kind reminder at the bottom that there are 50,000 more pages for you to review. And actually, maybe the thing that is most personal to you. What you actually want is on page 38 but you, you know, you just don't get you just don't get there. You sort of get the top of the stack of sort of paid and SEO results, but in a very customized LLM, you might get the product that is actually on page 38 and that might be sold from Japan or Buenos Aires, or anywhere, and that merchant might be very happy to sell you the good, especially if it's a digital good, right? Happily sell it to you over IP addresses, but they might not accept the payment method you have. And so again, the sort of cross of consumers, agents and businesses is much larger than what we're used to, sort of traditionally, going to a single site that's like somewhat geographically preferred. I think that agenda commerce might kind of connect the world faster, and then you're going to need all the payment methods and all of the currencies available to make those connections
Sy Taylor 13:56
actually work. I want to replay that point as well. There's something Steve that you mentioned about. There's three different actors in this payment flow. There's the ultimate human buyer, there's and whether they're represented by an agent or not, it's kind of a there's the agent middle layer, and then there's the merchant, and potentially there's even payments processes behind those merchants. And so they all have a series of different preferences, of like, Who do I want to buy from? What do I want to get paid with, or pay with? How do I like to display this? And it's ultimately, the merchant who is going to decide, and they care so much about conversion. I've gone to all of this effort to try and get a customer to consider my product, and if they're just going to walk out the door. All of that effort was wasted, so I want them to convert. So now, if the agent is making up new checkout interfaces, then maybe that's not going to convert. If the agent can't offer all of the payments types in the world, maybe that's not going to convert. Maybe it's going to break if the merchant can't make a decision about. Hey, I think this is fraud. Then maybe it's gonna So, though, you have so many points at which that payment can fail when you've changed it, because it's so simple, when I land on a website, but if it's if this check is my checkout, but somewhere else. So how did you think about like, what are the failure modes here, and how you start to solve some of those?
Speaker 2 15:19
I think that's a great point you just brought up, which is, in a regular payment flow, I'm going to the merchant's website. They control what I see there. They control the payment method collection, they control the checkout and cart state. They have all the signals that they're seeing right, your IP address, your location behavior on the page. And then they can, you know, on their own or with vendors as well, make a decision to approve or deny that transaction. And in this case, we've, like, sort of ripped it in two, right? The seller still maintains their payment processing, their cart state and their back end. Almost all the interactions the human buyer is having are in a totally different application. So you both have this problem of, well, how do I make sure that when I pick a shipping option, or I, you know, say I want two or three, that that's actually what ends up happening with the seller? There's a problem of, I put in my credit card info or in my payment method preference, how do I know it got there and it's going to be used in the way I expect it to be used? And then third for all parties, how do we make sure that the signals that are being collected in that agent's application that would typically inform a risk or fraud decisioning get over there as well, so that the seller can have the data that they need to proceed? So for each of those different parts, we had to figure out how that communication would occur and what the searchable guardrails would be. So at the top is check out. So that's a big part of the agentic commerce protocol. Is just like, how do I say I want this shipping option, or I want one more, or this is my buyer info, so on and so forth. And if we can both speak the same language on the agent side and the seller side, then that can be deterministic, right? So great, check. Then there's the payment part, right, which is, how do I make sure that I can properly collect payment credentials within the agent tech application, secure them in some way to send over to the seller and then let the seller use it up to the amounts that the buyer and other allowances that the buyer agreed upon up front. And then even in there, have complex flows like three Ds or step ups, tunnel back and forth. And then the third part, the risk side, something we spent a ton of time with stripe radar working on, is how do we make sure that the signals that are being collected by the agent can be sent over as part of that payment credential for the seller? Then to sort of glue their side of risk insights together with to properly process payment. So every step has to have the same language that's being communicated with, because it's it can't all just occur in one account context. You've sort of split
Sy Taylor 17:54
the experience in two. I know marketing teams will hate this, but I really loved that phrase you used once of like quantum entangled checkout.
Speaker 3 18:01
I'm glad you said it. I've been looking for someone else to say that one, but it
Sy Taylor 18:05
really does just make sure that the state is the same at both sides, right? Like, and there are so many failure modes in payments. This is why my favorite line is payments are easy. It's the edge cases that are hard. Like, on the surface, payments are just like, oh, I want to buy this thing for this price, go process the payment. But it's all of the things that can go wrong before it, during it, after it, that are just kind of unbelievable to kind of drill into. So bam. I'm interested in your thoughts on what you've heard so far.
Bam Azizi 18:35
Yeah, payments are working until they aren't right. So the checkout as we know the E commerce as we know it is dead. We're going to live in a new world that basically everything will be different. Everything will be optimized around agent tech, because we won't go shopping to shopping anymore as a human, our agents will do and they will prioritize shops that are easier to shop or easier to pay for them based on the payment methods they have access to. And they don't have any geolocation restriction. They're like global. They go and shop at page 38 as Jeff mentioned, right? So they would go and find the best for you as a human who is running those agents, and they would pay with whatever it's possible to pay. And I believe that's where, like the stable coin, as a borderless type of payment comes to the picture. But more importantly, not only e commerce, as we know, is that is going to be the new type of commerce. There would be merchants that they're selling a fraction of the service. There would be merchants that they're selling compute. There would be merchants that are selling like it's limited access to certain APIs that they will charge you, like half a cent. How you want to pay half a cent with Visa? Or do you want to wire it, like to a local currency? You have to find a new method that enables those type of transactions, those type of behaviors. I think that's where the stable coins are coming to the picture. Or crypto payment as a method you. Comes to the picture again, the consumer can still pay with the Visa card or gives you the Visa card or top up their wallet, but at the end of the day, the settlement happens through something that is enabling global, borderless and also instant type of payments.
Sy Taylor 20:15
We should speak to the Cloudflare guys, because I know they're doing something around x4 or two, and the idea there is that the API key is dead, and that actually you could get something that's more micro payments and where agents are accessing a digital resource, then actually, then being able to communicate and coordinate at a much greater scale is a fascinating idea as we get into that sort of space, but to the point like the broader like commerce space, of like the rest of GDP is buying stuff. It could be your groceries, it could be a flight, it could be, could be kind of just about anything that you can buy on the internet. It's just about anything you can buy. But if it goes increasingly borderless, it gets interesting. And Jeff, I'm interested in what patterns you're observing in, like the human behavior, the agent behavior from your sort of early volumes in this space, you know, is there a particular type of merchant that this is working with? Is there a particular type of payments flow that you're seeing adoption in and why you pick this space?
Speaker 1 21:16
I think what we're hearing is that all businesses are now facing new channels, and each of the channels is going to have its opportunities, right? An agent can find you anywhere on the internet, but also its challenges. You know? How do I know that it is a good agent that I have permission, versus a sort of spam bot that is now looking for access? And so, how do you sort of cut the space of all the world's businesses? Sort of a deep question. But I think you generally see, you know, are you a direct to consumer business with a high AOV average order value, like good that you're physically shipping? And in those cases, you know, we've seen some businesses really thrive with very direct new channels, right? There's an entire world of Shopify businesses that find their customers on Instagram that you know, you just could not have had an efficient way to get those types of D to C customers 10 years ago, and now you can absolutely be a single person business who sells globally and is able to find their customers in a very niche ways. So I think that we've sort of shown that with every new type of technology and distribution that there's some new opportunity for businesses, but then there's also a risk maybe your business upsold a lot of subscriptions at checkout. Well, how do you do that? Through the agent? And so these are the reasons where we think that the objective commerce protocol is going to evolve to support new business types, subscriptions, usage based, because we think that all businesses are sort of starting to merge types of business models. You might be sort of a digital content business, doing news, but you also host events. You might sell hats, but also you want a subscription so that you get access to the newest hat. All businesses are going to have to face this plethora of business models, and so they're going to need new tools. And then when we talk to these businesses, they are asking questions, just like, how do I physically model my business such that I can digitize my inventory? How do I make sure that my product feed is available for LMS in a perfectly defined type of way. Even the semantics of how APIs work in the past, where you sort of look up one record at a time, doesn't quite work for LMS, where they want sort of all of your information in context to provide the best solution. So I think you'll increasingly see stripes supporting making it easier for any type of business to sort of model what they offer, be it consumer or business, be it SaaS or e com, and then have one click style choices about how they're going to publish and price and merchandise that across the board. And as BAM said, too, it really depends on what you're selling, because if you're a API business and you're selling something that's relatively cheap, you know that's mostly just compute, maybe all traffic is good traffic, right? Like, I sort of don't care who called my API, I'm sort of happy to do so in an almost permissionless type of way. But increasingly, we have digital businesses that have extremely high cost on a variable basis because of the LM fees, and so there's really no like, one size fits all. And I think that you're going to want some type of platform and open tools that let you pick and choose these different type of business model and connect most safely for each of your business lines.
Sy Taylor 24:34
Yeah, it's interesting to see whether or not like chat GPT, which is currently the default consumer interface, ends up seeing like a long tail of consumer agents that are also out in that space. And yes, a lot of people, 60, 70% of people, might use chat, GPT as their daily driver, but maybe they've got a personal trainer agent, and maybe they've got like a personal shopper, and maybe they've got something else that was crafted by somebody else. And how do you know that that was a good agent? Maybe it is a good agent. And you know that because you've seen the agent before. But how do you know that the person that's using that agent is a good human and not a fraudster? I call it the Weinstein matrix. A little while ago, that you've got this new two by two of like, good human, bad human, good agent, bad agent. And how do you think about like, the kind of the flows that are going there and where they go so interested in, like, what you've seen in that space. And maybe it's too early to say, but stripe out of anybody is probably positioned to see the emerging behaviors because of how your clients index.
Speaker 1 25:34
I mean, good agent, bad agent is really in the eye of merchant beholder. One person's bot is another person's, you know, customer. So if you're a content business, you might have completely thrived with search index being the implicit way that an explicit way that you you track traffic by letting bots on your site for indexing, and then you receive customers or eyeballs later. And that's, you know, a sufficient sort of exchange of value that is well balanced, and other businesses see that as against their terms of service, the same exact behavior. And so again, you need to have flexible tools that let individual businesses pick their business model and then be able to sustainably, actually execute that on the internet, those tools haven't really existed at scale, and you're seeing lots of different new types of ways that they can be implemented. And one of my favorite examples is a startup that was helping in the last few weeks in which, you know, you tell me if this is an agent or not, but they build software that helps hospitals reorder all of their supplies. There's, you know, hospital needs huge number of SKUs to run the hospital. It's incredibly laboriously complicated to make sure you have the right SKUs at the rice hospital the right time. Obviously very important to have the equilibrium correct there. But you know, they are dealing with hundreds of vendors, some of which are fax machines, phone calls, APIs, vendor portals, and they need a large team of people to even recon, order, process, reorder, make sure everything is in the correct place. Is it agentic commerce to have agents that automate all of the ordering and processing and checking of all those SKUs to the vendors that they're receiving a robot phone call or a robot initiated fax or robot initiated API. It looks like the regular traffic that they were expecting, but it is being superhumanly charged by by AI on the hospital, sort of technology side. And so I think again, adentacommerce is a fun sort of two word nomenclature for this entire set of problems. But I think that the specifics of how each each of these business models are implemented require their own solutions. And so we think you should expect stripe to have many new ways that agentic commerce is implemented and and because we see businesses, they change their business model each week, each month, as they're as they're finding product market fit.
Sy Taylor 28:01
I love that idea that every complex back office spreadsheet is actually going to be an agent, but it might not look like it to the merchant, and being able to see and know that it could be, could be kind of powerful. Steve, any thoughts on things Jeff had said so far? So you came off the mute key.
Speaker 2 28:16
I think sharp in general, looks at what developers are doing as often early signal, but might become normal for others. Tonight, I reflect just on like my own usage of these kinds of tools, and I think what's interesting is basically, in the course of a year, I went from using a text editor that didn't have billing in it to a text editor that does have billing in it, and that my interactions with that product have where I have an outcome I'm trying to achieve, right? Which is some text blob being dumped has a monetary side effect, right? And that I'm no longer thinking in like, $19 a month. I'm thinking like, how many tokens of this particular model did I burn, and was the outcome worth it? And if you look at the early more consumer agent to commerce use cases today, it's like I still have an outcome that I'm trying to achieve, which is like buying a gift for my friend. It looks really normal compared to like, you know, my entire JIRA ticket being completed in one go. But you can imagine, over time, consumers become more used to the idea that prompts that they're giving an agent have a monetary side effect, right? Which could either be literally buying the thing, which seems like the base case, versus I need additional data. Back to Jeff's point of Read and Write APIs here, I need additional data, or I need to affect something in the real world to achieve the outcome. And those have implicit cost in them, whether that's the tokens that are being consumed in the act of doing it, or the actual third parties that are being interacted with to purchase or transact. So it's starting to feel very normal for me in my tech center spending money as I work. And I imagine it will start to feel more normal for consumers.
Sy Taylor 29:55
Too. Fascinating. Yeah, the per second, Per. Token labor cost of, hey, I'm having a dinner in a few weeks. Can my personal chef and go out and design a menu, get the food ordered, get it delivered to my usual space. Actually, that's burning tokens at the personal chef. It might be burning tokens somewhere else. And that shift in mentality of like, okay, the work before, the work is now almost like, you know, giving the teenager the $5 to run down to the store. It's kind of that mental model, but applied. I want to thank our sponsors for a second when we come back, Steve, maybe you could start to make some of this real for us. Maybe show us what it looks like in action. That'd be really, really cool. So before we move on, let's hear from our sponsors that make this possible. 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 back 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 mesh, pay.com forward slash AI. Thank you so much to our sponsors, Jeff, we were talking about token burning and that being a new mental model for consumers.
Speaker 1 31:59
Yeah, there's just a new type of utility online, which is token usage. And compute used to be on the order of free, but tokens are not. And so we we're just hearing from these super fast growing AI companies that a major portion of their cost of goods sold is the tokens themselves. And it introduces a new set of problems, which is, you know, a small portion of their user base is very token hungry. And so, you know, maybe, point 1% of the user base might use 60% of the tokens, or some kind of extremely lopsided
Sy Taylor 32:33
mentioning no names, but looking directly at Steve, yes,
Speaker 1 32:36
you know, we have our we have our large whales at the poker table of tokens. And you know, if you just think of it, a subscription only business with tiers, you're trying to kind of come up with the best kind of common denominator, tier package for the distribution of of your usage. And now it is much, much, much more lopsided than it used to be. And so we, we hear businesses asking us for sort of token based billing in addition to usage or outcome base, which is, I want to charge my customers X dollars per inference, or I want to make y percent over top of the raw cost of the tokens. And of course, the token costs are changing rapidly too. I mean, you can just watching Twitter is like updates from the mod providers, be they new models or new prices almost twice a day or something like this. So I think again, we're going to see all the world's business models sort of get more complicated, and then you're going to want some platforms that help you digitize this in some kind of way. And then again, the same with fraud. One of the things that's so hard to sell online is gift cards to a retailer, because they can easily be stolen and resold, and they kind of have no owners, and you can, you can use them online or offline. Tokens are also a really fun thing to steal online, because they have an incredible resale market. They're just as good as cash. They're very difficult to trace. And so businesses that are in some ways GPT wrappers, which we mean in the kindest, most possible way. Sometimes that's said in most derogatory fashion. But we think a lot of the world's businesses will be reinvented as sort of GPT wrappers. Those GPT wrappers, one of their core costs are these tokens, and they have to make sure that they're sort of protecting them and only selling them to real customers. And so we're seeing all kinds of new fraud patterns that we're trying to address. So tokens themselves is not exactly agentic commerce, but is a new kind of input in the supply chain to agentic commerce, which needs these new types of solutions.
Sy Taylor 34:31
It's definitely its own payments flow with its own nuances, a bit like gift cards. I love that. I feel like I need to deep dive into that actually and explain it, because that feels very FinTech notary, but bam, I know you had a question as well.
Bam Azizi 34:42
Yes, my question is like, little bit outside of a topic of token, but in general, we know that the payment providers, they compete with each other all the time, so whomever is the first in the list gets the most of the traction. It's limited real estate on every website. But agents are kind of blind. They don't. Care where you're sitting on the on that the stack. So what if there are multiple agent e commerce payment providers like PayPal offers the same thing, Amazon offers the same thing. I want to kind of pick Jeff's brain in terms of, like, how you guys seeing the competition emerge, and is that confusing for agents like, what would be the preference there, if everyone offers the same thing.
Speaker 1 35:21
Our core design audience is Internet businesses. And internet businesses, they like to accept all the world's payment methods, and they want to blindly optimize the black box algorithm of conversion and fraud and cost and so stripe sort of exists to make that choice available to businesses. And we think that applies directly to agent of commerce as well, where we want to help businesses sell through new channels of AI agents, and they need to reach all the consumers. And the consumers are gonna have lots of different payment methods, some of which are payment methods, again, that the business themselves not natively accept. If I'm based in New Jersey and someone is paying with the third most popular South Korean payment method like maybe that sale just doesn't happen because I don't know how to accept it from the United States. So maybe Steve can give us a little bit more of a deep dive technically on it, but one of the core principles was to, one, make sure that everything stripe builds makes all the world's pay methods work, and two, by default, make our offerings payment processor agnostic. And so our what we've launched is a stripe shared payment token is a new API that lets agents be able to securely transfer credentials to businesses to process on their preferred payment processor, and that does not have to be stripe. And so a poignant example is Etsy, which is the first merchant that went live in chat GBT. They do not process cards with stripe. So the first launch of agenda commerce with chat GBT is with a payment processor that isn't a stripe. And so we really taken it as a core design principle to make this open to all payment processors. And we've recently seen PayPal world pay and a few others announce their support of the agenda commerce protocol, because it is, again, as a sort of core design payment processor agnostic. And we can sort of, I think, even jump into a demo to see exactly where that formal, sort of the formal, I think elegance, in some way of the design is really sort of at the software payments layer. And then you can pick whatever payment processor you like.
Sy Taylor 37:36
And I just want to stick an exclamation point on the end of that. That That blew my mind when I found out that Etsy was not a stripe customer, and that actually this open source protocol is truly open source, and all of your competition can use it, which really does look like genuine open source. One might argue that somebody like Stripe is well positioned, if this thing exists, to capture market share. I can see that there's a sensible rationale there, but also that this thing needs to exist in order to capture any of that market share. And there's a recognition that if unless it exists, it doesn't exist.
Speaker 1 38:10
It's a networked problem, right? It's a network of consumers and businesses, a network of agents. And when you want a network to exist, it needs multiple parties and ecosystem. And so we wanted, again, it was sort of not even a debate internally. It's like for this to exist, it's gonna have to work for everyone, and so that's how we designed it. I'll just give a
Speaker 2 38:27
quick example of where I think this exists today, which is, when I create any website, I my meta tags that show the description and the title and the images, and then any application that, whether it's x or other social platforms renders the link. They all know how to render it, right? So the worst case scenario is we have, like, 100 different ways to communicate that's not friendly to anyone, and the thing that we are trying to do is just render something, and the equivalent is here, which is we're just trying to make it possible for any seller, independent of their circumstance, or the vendors they use to make their checkout available and process payments, because that's just beneficial for everyone. And then we can build higher order products, you know, on top of that, to serve our users directly. So the demo I want to do is is actually talking about one of those higher order products, which is what we're calling the stripe shared payment token. So I'm going to just do a screen share quick,
Sy Taylor 39:19
and for listeners, we'll do our best to try and describe this to you.
Speaker 2 39:23
We're in my tech center. This is where all of the aforementioned tokens get burned. In this case, I will be doing it manually, so no tokens were harmed in the recording of this
Sy Taylor 39:32
demo, and this is why you need to check us out on YouTube if you are listening to the podcast.
Speaker 2 39:37
Okay, so what we're looking at right now is actually a normal sort of pre agentic stripe integration. So, you know, I have my seller Stripe account, and I just want to collect $10 so I create a stripe payment intent. I'm going to do sort of a just generic card here, and I just want to collect and capture those $10 so just as the base case. So we're going to run that. So this is back to that earlier point around how as the seller, I have the checkout page, I'm collecting the credential, I'm processing payments, I'm responding with success. So we can see in the stripe payment tent for $10 it was received. Everything worked out as expected. Now, if I'm building an agent, and I'm collecting payment info in my application. How do I get this payment method sent over? Right? I don't want to just like, send a credit card number across the internet, and even if I could, what would I do about the long tail of other payment methods? Right? Not everything is just encoded as a 16 digit number that I can pass around. And on top of that, how do I make sure that I'm not just sending a crunch over that could be freely used, but could be used up to the limits that the agent and the buyer sort of agreed upon together? So I'm gonna cheat. This would be where the the agent would come in and write the code for me, but we're just gonna add in a little code snippet here, which does two things. First, we're going to introduce the agents Stripe account, right? So I'm in the AI chat application. I've decided to buy something. I'm presented with payment UI to pick payment methods so that collection is occurring in the context of the agents account, right, not the seller's account. And what I'll do is I will mint a shared payment token that attaches a card or any other payment method I collected from the buyer. In my agent's application, I'm going to say that it can only be used up to $10 for some duration of time, right? So imagine that's the cart total. So now I provision that I'm going to send that over to the seller, and now the seller is going to process that instead of the payment method that they would have collected otherwise, right? So we'll run that now. So we'll just go through this quickly. So I've minted the shared payment token. It has some risk details associated with it. It's scoped to a particular user. So in this case, my stripe test merchant, it has a limit of $10 and it hasn't been used yet. Now as the seller, I can look and learn a little bit about that share payment token. So I can know that the brand is visa, in this case, when it expires the funding method, but I don't actually have access to the raw credential, and I can see that I can use it up to $10 so in our example earlier, I ran that payment attempt against the $10 and I could capture it. So what we're seeing here is basically, in just a few lines of code, I can provision for any payment method a shareable token from one Stripe account to another Stripe account and include that with just one line of code change to process into my existing payment stack. Now let's assume that you know, I applied that $10 limit, but I want to charge $20 now. So some miscommunication occurred, or the seller wanted to maybe charge more than was authorized. So we'll run that one more time, so we can see there's a failure here. The amount is greater than the remaining amount capturable again, the seller didn't have to change anything besides that one line for their payment processing, and Stripe will help enforce the limits that the agent and buyer agreed to upfront. Again, that applies to other payment methods as well. So what's interesting here is, like we've made it really simple to do that transport of payment credentials for to work across a variety of different payment methods, because it's just a object ID that's being shared around with limits rather than the actual credential itself. And then stripe is sort of acting as an arbiter here to enforce what the buyer and agent and seller initially agreed upon, you know, to make sure that the payment that was expected is the one
Speaker 1 43:32
that's going through to bam's point earlier, you know, with stable coins being a new way that both consumers are holding funds around the world, and a growing set of businesses who want to transact the shared payment token is, or not, payment method specific, and that the stable coin acceptance that stripe has launched over the over the last year, and increasingly, the ability to hold stable coins works with exactly the same shared payment token, because, Again, as a principle, it's both payment method agnostic and payment processor agnostic. It's a way of, sort of transmuting the intention to buy from from the execution of buying. And again, we've for for stripe customers who are using our payment intense API. It is that one line of code change and you're no longer collecting the payment method directly. You're receiving it from the agent with the consumer permission. And so we think it's a huge unlock to take businesses who have already modeled how they want the world to work, and they just now have a new channel to receive transactions
Sy Taylor 44:34
from. And I think this comes to bam's point a minute ago, which is to the competitive dynamic. You've made it really easy to do on stripe. You've demoed how to do it on stripe, but there's no reason why somebody couldn't do that with a with another payments processor. So that's a really interesting point. Bam. I know stable coins are near and dear to your heart.
Bam Azizi 44:51
Yes, the other part that the stable coin can help is that part that the last part that the logic stopped the payment to go from $10 to an. Dollars that also can be done on the smart contracts, right? So do you see in future like Stripe moves toward like a more decentralized way of doing it? So there is no trust, there is a smart contract running that part of logic as well.
Speaker 2 45:14
I would expect that as stripe wants to support as many payment methods and payment flows as possible, obviously, stable coin is a great example with smart contracts of how we would contracts of how we would encode that. I imagine other payment methods are going to build similar technology to do that enforcement on their own right, whether that's at the network level or whoever's administering the payment method. So for us, we want to make it really easy for agents and sellers alike to take advantage of all of the different payment method technology that's being built, and I think stable coins and smart contracts is a great example, and other payment methods will
Speaker 1 45:46
certainly have something like that too. And we just rolled out recently, the ability for businesses, for last year or two, have been able to accept stable coin for one time payments with stripe, and now they can accept subscription stable coin payments. And the way that's implemented is through a smart contract where we're being able to pool funds on whatever the sort of cron regular basis is of the subscription. And you can, you know, go and manage your subscription on the blockchain the same way you sort of could go manage it in a more centralized way. But again, I think businesses are facing this proliferation of new ways of accepting money and customers, and they have all these different types of preferences. What's sort of their command central, where they can work in Fiat and stable coins. Sell directly on their own business through agents, do their own merchandising, do their own pricing. And is that 48 different independent systems, or is it a set of modular APIs that sort of work across all of these new payment systems? And that's really our focus. Is like making these modular APIs that solve each of these problems in the stack, and then businesses agents, people can sort of pick and choose where they want to use very hosted stripe services, where you do very little work yourself, or use our raw APIs and then kind of embed it into your application however you like.
Sy Taylor 47:02
I think that explosion of new payments flows, that explosion of new ways to get paid, to pay, then this new intermediary, it creates this interesting set of challenges for merchants. And yeah, it just gets harder and harder to get paid and to pay. We need to try and simplify that for any of it to work. But your closing thoughts for today,
Bam Azizi 47:23
yeah, it's just want to say super exciting, Jeff and Steve, you guys are building the future of commerce one line of code at a time, brick by brick. I appreciate you guys coming to the show, sharing it with the audience.
Sy Taylor 47:35
Yeah, no, I was just going to echo that, Jeff, if people want to find out more about you or the agentic commerce protocol. How
Speaker 1 47:41
do they go about doing that? Well, the agentic commerce protocol is available at agentic commerce dot Dev, which happened to be available as a domain name a week before the chat GPD launch. So no, it's good fortune. And then I'm available on Twitter. And I'm Jeff underscore Weinstein, very easy to find on the internet, and
Speaker 2 47:56
Steve just at Steve kalisky, so people want to follow me? I guess we could do that too.
Sy Taylor 48:01
Yeah, there's plenty of good demo videos going on over there. And bam. How about you and mesh?
Bam Azizi 48:06
Sure you can find at mesh pay, or go to Mesh pay.com, my handout Twitter is bam, as easy with two eyes at the end. So changing that to make it easier, but for the time being,
Speaker 1 48:18
yeah, do you remember the extra i Yeah, thank you for having us. Simon BIM, Simon, we've been chatting over a glass year or two on this topics, and every time we get to chat, I sort of fine tune my thinking on the topic a little bit more. So this is really quite a pleasure
Sy Taylor 48:32
any excuse to riff with FinTech nerds. I'm absolutely here for and we'll have many more episodes like this coming up in the not too distant future with some other agentic commerce, folks, but we started this one off strong. If you want to keep listening and watching, then go ahead and subscribe. Hit the like button on everything that you've got if you are watching or listening. And if you want to find me, you'll find me at sy Taylor on all of the socials, and@fintechbrainfood.com screaming into the void about agentic commerce, stablecoins and everything else. Thank you so much for watching or listening, and We will catch you next time. Bye for now.