Simon Taylor on Where Agentic Spend is Going

Will stablecoins replace cards — or is that narrative missing how payments actually work?

In this episode, Reggie Young sits down with Simon Taylor, founder of Fintech Brainfood and Head of Market Development at Tempo, to unpack one of the most debated topics in fintech: agentic commerce. Simon challenges the idea that stablecoins will displace card networks, explaining why cards don’t move money, how stablecoins actually fit into the stack, and why virtual cards may be the first real payment primitive for AI agents. 

The conversation explores machine-native payments, the rise of agents as customers, and Tempo’s Machine Payments Protocol — a new attempt to rebuild payments infrastructure for the internet era. Simon also shares where adoption will come from, why most narratives get the timing wrong, and what the next battleground in fintech will be.

Transcript

Reggie Young: Welcome back to Fintech Layer Cake, where we uncover secret recipes and practical insights from fintech leaders and experts. I’m your host Reggie Young, chief of staff at Lithic.

Reggie Young: On today’s episode, I chat with Simon Taylor, the founder of Fintech Brainfood and the head of market development at Tempo. In March, Simon put out two awesome essays on the topic of how agentic commerce will develop. The first is called “Agentic Commerce Will Use Cards First, Then Stablecoins,” and the second is called “The Intention Layer.” They’re some of the best syntheses I’ve heard on how agents and stablecoins will fit together and where cards come in.

Reggie Young: Fintech Layer Cake is powered by the card issuing platform Lithic. We provide financial infrastructure that enables teams to build better payments products for consumers and businesses. Nothing in this podcast should be construed as legal or financial advice. If you’ve enjoyed this podcast, please give us a review on Apple, Spotify, or wherever you listen. Awesome. So let’s dig in. A lot to talk about today.

Excited you’re part of an exclusive group, third-time Fintech Layer Cake podcast guest. I think it’s just you and Matthew Goldman. I think this is the only three-timers.

Simon Taylor: That is an illustrious and very difficult act to follow. Genuine cards industry legends. So in rare, thank you for letting me spam you once again, my friend.

Reggie Young: No, always great to have you on. You’ve been putting out a bunch of awesome, very thoughtful stuff lately, so a lot I want to cover. A good place to start is there’s a recent Citrine Research piece that argued stablecoins are going to disintermediate Visa and MasterCard. That subsequently tanked the networks’ stock. You pushed back on that research pretty directly. What do you think they got right, and where is there an area to break down?

Simon Taylor: So they definitely saw something in the market, which is there is this like incentive mechanism for using the word AI to do job cuts. And AI is making people a lot more efficient. And then the other thesis is that spreads beyond SaaS and goes into other industries, where they got it wrong in financial services is Gell-Mann amnesia effect when they talk about your industry, like, well, I think directionally, you’re right, but you’re wrong about the timeline. And you’re wrong about the rationale and the reasons. So for those who didn’t read it, the thesis was, agents will prefer stablecoins to Visa and MasterCard and Amex, but Visa will be a little bit better off because they’re pretty smart in stablecoins, which I do a podcast with Kai Sheffield. And I was like, do you realize you’ve probably helped the share price by about half there just by existing and doing the work you’re doing. But that aside, I think it misses where the real opportunity is in stablecoins, and agentic commerce more generally, it’s a really good narrative, but it wasn’t the truth. And hey, same with Figma being down, I don’t know if you saw the news today, they’re down like 80% since their IPO because Google launched some agentic design tool, really cool tool, probably does challenge Figma a little bit, but Figma is using AI really well. And they have 136% net revenue retention rates. So like, the fundamentals don’t match the narrative just yet. Maybe they do, Claude can do amazing things. But let’s see.

Reggie Young: Yeah, I think a little overblown fear from that research piece. That big meaty topic that I want to talk about you with is just like cards are emerging as a sort of like a stablecoin, there is a key player in stablecoin agentic payments. You have some interesting thoughts in the piece you wrote on X, about the different roles that cards and stablecoins play in payments. So I’d love you to break that down for listeners.

Simon Taylor: Yeah, sure. Let me pull that up so I can remember what I said. The speed of fintech makes a month seem like years. Yeah, but you’re talking about agents will use cards first, then stablecoins, which is a piece I wrote, because there’s a lot of companies out there who are big in stablecoins, saying that stablecoins and agentic commerce are the obvious answer. They’re in stablecoins are internet native. But I think first of all, people confuse what cards actually do. Cards don’t move money, right? They authorize the movement of money. A card is like an identifier that links a human person to some account number somewhere in some pot of money, and then lets you walk into a store, swipe and authenticates that like, hey, look, this person is good for the money, they can probably walk out of the store. Reggie has got some money in his bank out somewhere, we’ll figure all of that out later. You’re good. Just keep going. Stablecoins move money. The other thing that moves money is banks, through the correspondent banking network internationally, and with the Fed domestically. And the problem with the Fed is it’s not 24 by 7 yet. And the problem with correspondent banking is it’s definitely not 24 by 7. So this is my first observation, which is stablecoins don’t solve cards, necessarily, because cards aren’t an issue for the consumer, they work, and they’re extremely overpowered, because they’re accepted everywhere. Stablecoins, I don’t know if you’ve tried to buy a salad with stablecoins lately, pretty hard. They’re not accepted everywhere. So these two things combined actually make a lot of sense. And yeah, the other argument I said is I think agents definitely need their own payment method. Definitely. Like your agent goes, you type in your favorite LLM, them, buy me a sandwich, please. And it goes, cool, here’s the website, click pay for me, will you? And it’s like, well, that’s not very agentic, is it? And then everybody’s been trying to put the checkout in the agent. And then they’ve gone, No, nobody use that. That didn’t really work. But why don’t we just give them a payment credential. And this is where everybody goes, Oh, well, surely the payment credential they’re going to want is a stablecoin. And again, if nobody accepts that stablecoin, what was the point in giving them the stablecoin? So my unpopular opinion was agents will use virtual cards first. Yes, virtual cards are amazing. Heck, privacy.com you might have heard of them do a pretty good job with virtual cards, Stripe issuing ramp breaks. They’re all built around this idea that when you get a new account, here comes this virtual card, but like the expense management software, you can set budgets, you can set limits, you can say you can only spend at this particular merchant and you can spend $7.77 at lego.com. Like you’d be weirdly specific about a virtual card. So if you’re worried about an agent, just like hallucinating, it wants to YOLO, go purchase a bunch of random Pokemon cards for you that you didn’t ask for, then you can stop that with a virtual card, which is pretty amazing. So I think that’s going to be the first disruptive thing that happens. And then like all of the x four twos and all of the marketing, especially the memes, not necessarily Coinbase, but the memes around x four or two, like this is going to replace everything for agent to agent commerce. Problem is it’s not in fact, the volumes on x four or two are plummeting. It was really mean coins. So where’s the volume for agent to agent going to come from? And I think actually, where’s the volume for stablecoins going to come in? Firstly, the obvious answer is better settlement for cards. Visa already does this. If you are a stablecoin backed card today, you can settle directly with the card network and acquirers can settle directly with the card network. So you’re not waiting three to five days from a neobank in Brazil to get paid in the US. All of that can happen 24 seven with a stablecoin from both of the major card networks at the moment. That’s huge. Now think about an agent Reggie, an agent like sells tokens, really. So classic merchant, I have a bunch of widgets, I want to sell them. But it’s going to take me three days to get paid. And then when I get paid, I can go buy a bunch of new inventory to sell a bunch of my like Pokemon cards or whatever it is that I’m trying to sell. So that’s the classic gap that lending always used to close with tokens. That’s times 100. Let’s say Reggie, you go build Reggie AI, and everybody wants Reggie AI tokens, and it’s like exploding. But in order for you to get paid, you got to wait three days for that to land in your bank account. So you’ve got to close your service for three days in the middle of hyper growth, and then maybe open it again and try and continue. So I think that instant settlement is the superpower. It’s not a very sexy use case, but it’s the key. And then where I think stablecoins actually come in later on is the more complex use cases because as powerful as virtual cards are, you can just be a lot more fine grained with stablecoins like they’re way more programmable, there’s way more dimensions, and you can have this like inheritance structure that’s really nice. So you can have a company level wallet, and each of the agents underneath that could have their own sub wallets, and they can have employee wallets. And so when you get to that sort of level of discipline, when you think about a finance infrastructure, where I’ve got 100 finance agents or multiple employee agents, wallets start to come into their own a little bit more. And at the end of it, they all might have virtual cards attached to those wallets, right? Because they still need to buy things in the real world. But overall, as I guess you can tell, my take is, there’s a lot of nuance missing in this debate. So if you go to brainfood.com, look for agents will use cards first and stablecoins. I ramble on about this for about 3000 words. So how about it, guys?

Reggie Young: No, I read it. I immediately emailed Simon was like, we got to talk about some podcasts. It’s a great piece that folks should go listen to for sure. Yeah, it’s in gmetroprivacy.com we’re seeing a ton of like agentic use cases there the virtual card it is the most common misconception I encounter is this idea that card networks move money to your original point. It’s so funny. It’s like they are a messaging network at the end of the day. And you think about them as moving money because your card, you pay with money, but it’s not that all the money moves to the actual banking system. So you can easily swap that out for stablecoins for the settlement piece in the background. So I love it when I do you mentioned in that piece that jumped out to me is the idea of an agent as the customer and I think you hinted about this a little bit already, but I’d love to double click into that a little more walk me through what that means. And what you think the implications of treating the agent as a customer are?

Simon Taylor: Okay, one of the fun anecdotes that Noam Levine from Andreessen brought up recently was that in the early days of Stripe, the Collison brothers were struggling to get investment because people were saying like, who’s your customer? And they said, oh, our customer doesn’t exist yet. They said that in a much more charming Irish voice, of course, but I’ll say it’s not a good pitch. And I think what they’d realized is that the developer was the customer of tomorrow and the developer needed better documentation, clean APIs, they needed to be able to get moving quickly. And that speed, that time to first payment, that time to hello world had to be as small as possible. That is now widely understood. I think everybody in tech tries to deliver that, but in 2010, that was a wild idea. There’s an argument that says, well, the vibe coder is a new type of merchant. Like they’re building their first agent and maybe even Stripe’s a bit complex for them. I was like, no, Stripe’s pretty great for them for the most part. Agents are pretty good at setting up Stripe and doing all that kind of stuff. But what about an agent? How is an agent going to buy your product? And how do you think about them building products, selling products, buying products? If an agent is buying from you, how are they going to buy from you? How are they going to look at your website? Because maybe all of those graphics you have and all of those things that you use to trick a human into pushing purchase, maybe that’s just wasted tokens to the agent. What they want is just like a really nice clean markdown file, like helloagent.md. Here’s where you click buy from and here’s some really clean. So thinking about the customer of tomorrow, agents will sell their own products. They need to buy products. They can’t see a website like a human would. They’re going to prefer the markdown. They will also then prefer internet native payment methods. And that’s where Stablecoin’s Instant 24x7 programmable. In that broad idea, Citrini is right, but on the timeline, maybe not so much.

Reggie Young: No, Citrini, we’ve been talking and thinking about this a lot at Lithic for a while, like where all the legacy infra was built for human spend, which is just like a funny, it’s like a fish and water problem where like you never thought about like all payments are human payments until now. And so you look at like the sort of transaction control layers that existing infra options have built and they’re just like pretty static. They’re pretty simple. They don’t allow a lot of the rich context. We’ve been talking about this as authorization intelligence, like this idea that you need smart authorization controls, like the sort of static legacy controls on transactions don’t work anymore now because you need like smart, dynamic card controls that can take in context. And does this agent have permissions and like all that sort of fun new stuff that, yeah, it’s a brave new world. So about card products, I was just thinking of this, but I’d be curious for any other thoughts you have on what needs to change to make a card product truly agent native. And maybe this is a potential jumping off point to talk about a recent Tempo announcement, the merchant payments protocol.

Simon Taylor: Yeah, it is. But I think, look, the card networks have been doing a pretty good job here with Visa’s intelligent commerce and MasterCard’s agent pay. They’re trying to make sure that at the network level, the intent of the user can be received by the merchant and validated on both sides. And there’s some tokenization happening there. And that works really well for their networks. If everybody in the ecosystem can use those tokens. But there’s still like a cleanliness of being able to make that payment problem upfront, right? So to set aside, there’s a whole bunch of authentication and security challenges that need to be solved lower down at the network level. There’s an agent experience problem. Like your checkout isn’t optimized for the merchant. And then what we’ve tried to do is take that checkout, place it inside the merchant themselves. Every time they tried to use browser use, it’s a bit useless. You need to really rethink how commerce works to be agent first. And that’s exactly what we’re talking about with the machine payments protocol. How do you make it a part of the internet infrastructure itself? The internet has protocols for email and HTTP and HTTPS and networking and TCP IP. We always had HTTP four or two. We always had the ability to make payments on the internet. But the irony is if you’ve ever heard of HTTP 418, 418 just responds I’m a teapot. And this is something that almost every developer does when they’re playing around with HTTP for the first time. So HTTP 418 has done far more transactions than four or two ever has on the internet. And I think part of the reason for that is we’ve been in an attention economy. And the way you monetize that was like I need to get as many eyeballs as possible and aggregate them. This is aggregation theory. Then I need to get as many impressions onto my website and get clicks. And then I might be able to drive them towards an intent and a purchase. The human web is built for that. And the payments system and architecture is built for that. But agents show up with a mandate. The humans already told them what they want. They’re just like go buy it in the most efficient way possible. So how do you give them the most direct path from the mandate they already have? have to payment is completed. And that’s what the machine payment protocol is kind of about. So how do I get them there? I need something that works the way the internet already does. So same HTTP semantics that www.authenticate, all of that authorization needs to be HTTP native, not card network native. And I think that’s a different place. And some of the existing standards have invented their own versions of that. MPP is an IETF grid standard. It’s been fully submitted to IETF, it could potentially find its way into the fabric of the internet itself. Then it needs to be payment method agnostic, because it could be a card, but it could be UPI, it could be PICS, it could be a stable point, it could be anything else. And then machines can authenticate fast and then figure out settlement later. And the throughput needs to be extremely fast. Did you know, I was doing some research, there’s 4 million email sent per second. I don’t know what payments infrastructure can do 4 million transactions per second. In fact, I don’t think we should even try. So what the machine payments protocol is fundamentally, is an attempt to build something for agents operating at machine speed, or the speed that tokens stream through the internet. So we invented this idea of sessions, specifically, Brendan and Georgios and GK, they built this with Stripe. If you ever been to a gas station, you swipe your card, you fill up and you see the numbers ring up. Now you as a payments nerd know exactly what’s happening there. The pre-authors been taken, it’s collecting the numbers and then the capture happens once you hang the receiver, the fuel nozzle back to the hook, right? What sessions are, right? So I’m an agent, I want to get a bunch of tokens for something. And maybe I just want to generate an image this one time from X, maybe I want to do a search of X, like things that I would never go buy a subscription to have access to, but I want to do it this one time because I’ve got this one task. That’s the perfect thing for a machine payments protocol session, which is, I don’t know how many tokens I’m going to use, but I just want to do this one thing, start the session, then close out, capture, and then settle as a card transaction, as a stablecoin, settle on lightning network. And that’s what the machine payment protocol is in a nutshell. And that being payment method agnostic is something that we’ve built and put out into the world. We think stablecoins will be a really good solution for that, but they won’t be the only solution by any stretch.

Reggie Young: I love it. How do you see a sort of path towards adoption you’re talking about? One of the things I love that you included is the XKCD. Folks should go check this out. The intention layer is a phenomenal read. I love the XKCD panel you included about how like, we have 14 standards. We should have a standard that applies across all of them. Okay, now we have 15 standards. How do you think we’re actually going to get to a sort of unified, because it’ll drive so much value once everybody’s kind of operating in one key layer. How do you think that adoption curve is practically going to happen?

Simon Taylor: I think it’s just got to add value and get some traction, as stupid as it sounds. And I think that’s a big reason why I was really keen to have Visa’s involvement and Codd’s involvement in it, because it was about serious grown-up things. And if you just over-index into stablecoins, then you tend to draw the wrong crowd, frankly. I am the disappointed dad of crypto. I’ve been in this game far too long, and it’s just, they did meme coins again. And we need to get away from that. The thing that’s really surprised me, though, is the people using MPP that have played with it have been like, I wouldn’t have used that API, and I found myself doing it. And I wouldn’t have used that service, but I found myself doing it, because it was just so easy. It was just available in the command line. And so it’s when a developer says, oh, this was useful to me, ah, we’re getting something. We’re starting to get something. So long may that continue. We don’t need a 15 standard, but like, hey, the internet’s not short of a few standards, like CSS, JSON, like, oh my God, there’s plenty of them. To a non-technical person, the amount of acronyms in finance is one thing, but in the internet stack, holy crap, there’s plenty of them. I thought I did not a comp science degree, but a diploma in my teens. I thought I understood the basics of the internet, but like, they just keep adding new ones and new ones. So these are out there all the time. There will be more standards and more protocols, but ultimately it shouldn’t matter to you eventually, unless you’re really operating with it. It should just be a really great use case. Another one that came out of the hackathon that blew my mind was somebody saying, like, we can probably do something where we wrap all of the payments for local parking in a city and you would no longer have to download the individual app. Like if we can wrap them all in the proxies for making payments, then we can get the GPS off your device. And then you just walk out of the car and then your agent would figure out how to pay for you. So there’s a lot of little schleps that you can build. Once your agent can pay, you just. have to close the gap as much as possible between the agent knows what its job is, and that’s how it pays. That’s what MPP is about.

Reggie Young: I would absolutely sign up for that parking app. Always a pain every time I park different app, you got to download. So 100% every city is different.

Simon Taylor: All of the apps are categorically awful software, they’ve probably got spyware in them.

Reggie Young: Yep. It’s a nightmare. Yep. Awesome. Any other things? I have my standard wrap up question for you. But then curious, is there any other last things you want to cover about obviously Tempo has been making news on this merchant payments protocol this week, folks should go look it up if you haven’t read about it. But any other things you want to share?

Simon Taylor: The other thing I’d say is that the Tempo mainnet is live. We buried the lead in that. We were seven months from first line of code to mainnet, which the team just absolutely pulled out. Honestly, I sit in this team, and I’m just blown away by what I see in Slack every day. I’m like, goodness me. What am I doing here? These guys are amazing. I’m just a guy with a British accent and a blog. What the heck? But it’s really nice to be around such a team that’s shipping incredibly. And we’re gonna have some good stories coming up. But there’s some really lovely features in there. So if you go to wallet.tempo.xyz, sign up for a wallet, the flow is seamless, you can pull funds from any of the chain quite easily will even give you $1 if you connect your x account to it. And then try sending people money. It is aggressively frictionless what you can do when you design all the way through the stack down to the chain. And the chain can do amazing things like it has native support for pass keys, like what payment system in the world has native passkey support available to you. And it’s just this developer primitive that you can play with and build interesting things. So the amount of stuff we’ve been vibe coding on the test net internally that we wanted to build in our previous banker jobs is unbelievable. So go have a play with it. Like honestly, I was speaking to a developer at one of the global central banks who was just like, I’m obsessed. I love tempo. For a long time, they were using another chain to demonstrate stuff internally. And then they’re like, nah, this is the one because it’s got all of the pieces in there. So yeah, shout out to the team. They absolutely crushed.

Reggie Young: Aggressively frictionless. I love that phrase. I think about shop pay. First time I ever used that product. I was like, Whoa, you know, it’s like somebody rushed past you. And there’s a breeze in front of you. It’s like, I was gonna click the button, but you did it for me. And it already showed up at my house. What? Slow down. You’re not even sure I bought something. It’s here. I love it. Awesome. For my standard wrap up. What’s something you’re thinking a lot about that you think people in fintech aren’t talking about enough, which I feel like is a hard question for you. Because you think a lot about a lot of things.

Simon Taylor: Here’s a medley. I think it’s interesting that calcium is leaning into being the anti insider trading regulated position. I did a piece on it in brain food last week called the everywhere insider. So fintech brain food.com find the everywhere insider. And I just don’t think people are really paying attention to some of the weak signals coming out there. And then lastly, the one question I get from every VC is Simon, have you seen like the agentic version of Chime yet? Like, where’s the cash app, but that’s agent first. And I’m like, "Well, no, I haven’t." I think we’re about six months away from that. I think it’s coming post open claw and post on device. And with some of the pieces, I can see how that could start to come. And I think we’re a little way. So what I’m working on now is wallet wars, part four, the battle to be your personal agent and guardian. And I think that is going to be the most interesting battleground in all of tech, because like, it’s not clear who gets to win. Is it going to be like Apple? Is it going to be an AI lab? Is it going to be the banks? Is it going to be fintech companies? Is it going to be something open source? Is it going to be like some startup? So we are heading into the most interesting time in fintech I can remember.

Reggie Young: I am very excited to read all your thoughts on that. So look forward to it. Awesome. So this has been a great conversation. Again, folks probably already subscribed to fintech brain food, but if you don’t go sign up for it and tempo XYZ, go get your wallets and test out the aggressively frictionless experience.

Simon Taylor: Wallet dot tempo dot XYZ. And if you’re curious about MPP, MPP dot dev, we got a little demo on that where you can try using different API’s that you haven’t from a command line interface that are already wired in and you can generate an image with some new model that you’ve never used before. And if you want to demo it to somebody, that’s the penny drop moment. Have fun with the guys and let me know what you build. Just email me I’m fintech brain food at gmail.com.

Reggie Young: I love it. I go clear my calendar to go play around that stuff. Thanks. You’re welcome, sir.