Why 2 Crypto VCs Envision a Future of 'Headless Merchants' and Autonomous AI Agents

Transcript

Hi everyone. Welcome to Unchained, your no-hype resource for all things crypto. I'm your host, Laura Shin. Thanks for joining this livestream. Before we get started, a quick reminder, nothing you hear on Unchained is investment advice. This show is for informational and entertainment purposes only, and my guest and I may hold assets discussed on the show. For more disclosures, visit unchainedcrypto.com. Introducing Nexo, the premier digital wealth platform. Receive interest on your digital assets. Borrow against them without selling. Trade a variety of cryptocurrencies, all in one platform. Now available in the US. Get started today at nexo.com/unchained.

Today's topic is the future of agentic commerce. Here to discuss are Noah Levine, partner at a16z, and Robbie Peterson, junior partner at Dragonfly. Welcome, Noah and Robbie. Thanks for having us. Yeah, super excited to be here. So, the agentic commerce space has seen a number of new technical offerings in recent weeks, and both of you wrote pieces this week on what that future looks like. It was pretty interesting, actually, cuz your visions differ a bit. Um but Noah, why don't we start with you? Describe where you think this is all headed.

Yeah, for sure. Well, I mean, I think it's it's helpful to start by saying that we've seen this massive catalyst in new developers by virtue of, you know, platforms like cloud code. And as a result of that, I think what we're seeing is that there's this whole new demand for new services that these developers want to use. And so, with the advent of new protocols like MPP and XPRL 2, what we're seeing is this there's this new class of merchants, which I kind of titled headless merchants, which effectively the developer has no insight or has no idea who they necessarily are. But when they're going in and instructing their agent to go and purchase something, that agent is going out, discovering them, and then paying on a per transaction basis to to effectively get those services. And I think that this is going to create a new wave of sort of B2B

commerce, where developers are looking for these new tools, and rather than them necessarily dictating who they are or going and and finding them via website, they're going to be able to see them a more natural discovery through cloud code itself. And Robbie, what about you? Why don't you describe the vision that you laid out in your little essay on X? Sure. Yeah, I would say broadly, I I agree with Noah on maybe on the direction of travel. I would say maybe where we disagree is probably the magnitude of of how how big agentic commerce is, especially in the near term. Um I think my vision of, you know, how how things going to play out is kind of broadly fit into three categories. So, one is commercial agents, and intuitively, these are agents that are deployed kind of within

businesses. Um two, I would say you have kind of consumer agents. Uh so, these would be agents that sort of augment, you know, our personal lives, or um you know, helping us buy buy kind of more consumer items. Um and then last one I would say is these kind of bottom-up agents, and this is more consistent with kind of the open claw phenomena. And these are agents that are actually actually autonomous and actually transacting in the real world. And I think 95% plus of of kind of, you know, most most agentic um you know, activities is ultimately going to be these commercial agents. And I think intuitively, that's consistent with how, you know, you know, if you look at like the SaaS market today, um you know, it's 95% plus of of software is probably deployed within businesses, right? And governments. Um and I think the same thing will be true,

you know, for agents. And um I think maybe the nuance that that everyone seems to be missing is is that, you know, an agent fundamentally, what it does is it it automates. Um but that automation doesn't necessarily mean that it has to actually spend. Uh and I think people have kind of conflated the two. Um and then they just, you you know, sort of drawn the conclusion that agentic commerce is going to be massive. Um so, I think that's on the on the sort of um the commercial side. And then I think on the consumer side, uh I'm also very skeptical that agents are actually going to be transacting on behalf of, you know, individuals autonomously. And I think, you know, the common sort of, you know, bull case for agentic commerce that people love to cite is, "Hey, I'm going to tell this agent, you know, go

book me a book me a trip to Tokyo, let's say." Um and it's going to, you know, it's going to find the hotels, it's going to scrape a bunch of sort of data, it's going to say, you know, it's going to read reviews, look at photos. Um maybe it'll have my calendar and all my preferences as well. And it'll ultimately say, "Okay, you know, I booked this thing. You know, you never had to kind of look up from what you were doing, and that's it." Um and I think that's kind of that rests on a a bunch of implicit assumptions. Um one of them being that, you know, we will be willingly, you know, you know, we will actually want to outsource all of that to an agent. And I think the more more realistic path is, "Hey, I'm going to ask this agent uh to, you know, I'm

going to tell it I want to go, you know, book a trip to Tokyo." Um and then it's actually going to give me options. And then it's going to ask follow-up questions. And it may say, "Hey, you know, where where where in Tokyo do you want to stay?" And um "Hey, here are some, you know, reviews, and this is what they're saying about this hotel versus that hotel." And so on. And I think the other nuance people are missing is, you know, preferences aren't a static thing. I think they're ultimately revealed in the process of discovery itself. Um and I think that's not just true for, you know, whether it's booking a trip. I think that's also true for, you know, whether you're buying groceries. And it certainly drove you buying clothes. You know, you actually want to go see what the what the clothes look like, and so on. So, I

think just most consumer decisions, um the agent won't have the full context. Um and I think, you know, again, I think there's all these qualitative inputs as well um that you don't actually discover until you go through that process. Uh and, you know, an agent doing all that on on your behalf never actually captured that that kind of nuance that the consumer ultimately wants. Um so, that would be that would be on the uh consumer side. And then uh to Noah's point, I I do think there is this emergent sort of category. And I I would call these like bottom-up agents. Um and again, this was sort of catalyzed by, you know, the open claw phenomena to kind of Noah's point. Um I do think this category will grow. I think that said, it is still extremely nascent. Um and if you just look at the data, there aren't

that many, you know, people who are actually using these agents to transact. It's more, you know, a lot of it's been sort of speculative, whether it's XPRL 2 or MPTP. Um so, I just think it's it's very early, and I think the market's just generally, or at least at a minimum, social sentiment's got kind of got ahead of itself here. Yeah, I mean, honestly, when I was reading both of your pieces, I was thinking that this will probably roll out in stages. And so, even though the visions that you laid out were a little bit different, like I didn't think either was kind of right or wrong, because I was like, well, they're probably both going to be right at different times. [laughter] Um so, maybe let's talk about just this moment right now, where things are kind of getting getting built out, because like something that was interesting to

me was that in both of your pieces, um or or just in various other things that you wrote, um I saw that you both referenced how stable or sorry, um how credit cards will be uh fitting in with stable coins. And um you know, I think you both have ideas on like how much agents will use credit cards versus stable coins. Um so, yeah, can you talk about like how you think that will be decided, or like how um you know, the trajectory of both of those types of payments will go um in this early phase. And either one of you can go first.

>> Yeah. Yeah, I can I can touch on that. No, I mean, I think, look, like if you, you know, I think it's very similar to the discourse that was happening a few years ago when people were talking about, well, stable coins are going to replace all of consumer-to-merchant payments. And you know, it turns out that the the networks beyond, you know, a lot of the the additional value-added services they provide, there is a huge uh you know, sort of Lindy effect, where you have consumers that have been using cards for a very long time. You have merchants who've been accepting them for a very long time. And so, I think, you know, it's very natural that, for example, stable coin and cards ultimately became the preferred form for consumer-to-merchant payments. And I think a similar thing is kind of happening in in agentic commerce, where,

you know, people are coming out and saying, "Look, you know, there's there's all these benefits of using stable coins versus cards." But, you know, what they don't maybe realize is that for most the the average consumer, they're much more, you know, familiar with using a card, and they would prefer to use one. And I also think there's been a lot of arguments that have said, you know, uh agents can't hold cards, and this technology doesn't work. But I think if you look under the hood, you can see things like, you know, the same technology that powers Apple Pay with tokens is the exact same technology that uh the networks like Visa and Mastercard are using to power agentic commerce. And so, I think from a starting point, you know, it's going to take longer for the networks to um you know, to establish their standards and to make cards

conducive with this new form of commerce. But I think there's nothing from a technical standpoint that is going to stop cards from working. The what I kind of talked about is, you know, again, I think kind of to Robbie's point, like for consumer for a traditional consumer-to-merchant transaction, like if I'm looking to book a flight, you know, people are going to have preferences, they're going to want to use cards to get things like points. There's a, you know, an element of chargebacks and fraud, which is helpful. I think that is probably the more boring side of agentic commerce, and is likely to be going to be one largely by the card networks. Um but that being said, I think again, this whole new sort of developer ecosystem that's forming, what we're seeing is that there's all these new merchants that are being established that, you know, are maybe a a couple of

buddies in their basement that are building a merchant in a couple of days via vibe coding. For those people, it's going to be somewhat, you know, challenging for traditional acquirers and processors to underwrite them, so they can accept cards. And I think stable coins are kind of emerging as this, you know, new digital cash, similar to how, you know, you'll have Macy's will accept cards, but you'll have street vendors on the, you know, on the corner who are accepting cash. I think that for that group, stable coins are actually a very uh useful solution. And and actually, just a quick question, who would be the equivalent of that in this world that would, you know, be the equivalent of the street vendor accepting cash?

Yeah, so I mean, a great example is, you know, a couple of or the the report I did or the article I did before, I was looking at XPRL 2 volume. And, you know, there's this this really great data provider, Aleph, who has an XPRL 2 server. I didn't have an account in Aleph, so I just wanted to get a couple of data points. I was able to, you know, spend I think it was 30 or 40 cents to build the entire report. No card, was very easy. You know, I think for for something like that, that's, you know, data is a great use case for that. And Robbie? Yeah, the only thing I would add is I think yeah, and I think Noah makes a bunch of great points there. And I thought his piece was actually the best piece that anyone's sort of written on,

you know, how this how how things evolve at least with respect to card networks versus blockchains. I think the reality is like today both are insufficient and they're insufficient in their own ways. So if you look at, you know, card networks, what they've solved, which is actually very very difficult to solve and I don't think a lot of people fully understand this is is risk scoring and off. So they're actually like authenticating each of these, you know, transactions and it's not as simple as just saying, "Hey, this transaction's fine. Let's let's, you know, let it through." It's hey, we've like all these, you know, sophisticated data points that we're that we're looking at and then we're assigning some confidence interval to like, "Okay, could this be fraudulent? Could it not be?"

So they've solved that. Where they're maybe a little more insufficient is is on the settlement piece, right? Because settlement is not, you know, instant and there's a bunch of structural reasons, you know, for that which we can unpack, but that is where they are insufficient. Now conversely, if you look at blockchains, you know, they have this instant instant settlement, you know, intuitively this is a global, you know, asset ledger that syncs in real time, you know, which is very conducive for agentic transactions. What they have not solved conversely is you know, the actual risk scoring that the card networks have solved. So I think they're insufficient in their own ways and I think they both have a a lot of work to do and I think each of them respectively will kind of kind of continue to converge and build upon the same vision. I think the reason at least

with respect to these kind of bottom-up agents as I would call them, as Noah alluded to, you know, this could be, you know, a bunch of, you know, people just hacky devs kind of spinning up, you know, different different agents kind of, you know, in their garage or whatever. I think with respect to those use cases, I think blockchains win for a different reason and a non-technical reason, which is they're open and they're permissionless and they're unencumbered by sort of any regulation, right? And maybe said differently, there's just way less friction building on on blockchains versus building on traditional, you know, rails. And, you know, Visa and, you know, Mastercard, these are public companies that have, you know, that have shareholders, that have institutional counterparties and they they will inherently not be able to take the same amount of risk that, you know, like a blockchain could because

blockchain doesn't have, you know, shareholders, they don't have this risk. So I think for that reason most of this experimentation probably actually does happen on on chain, but maybe that's not for the reason that most people are are thinking intuitively. Okay. So um one thing that I did want to ask about was um you know, for like um the way so the way that this goes with, you know, what you call the bottom-up agents and then um I I don't remember what you call the other ones, >> [laughter] >> but I was just curious like you know, for the bottom-up agents, they would still need to have some back and forth with like their human, right? Or like how how is it that they are more autonomous?

Yeah, so the way the way I would frame this is, you know, if you're looking at kind of these the other side is commercial agents and then kind of kind of consumer agents. And the reason I would say these are more top-down is because I think they will ultimately, you know, be distributed and there's they're going to be distributed via, you know, some software vendor probably at least on the commercial side. And then on the consumer side maybe it's, you know, Open AI, you know, distributing this via, you know, ChatGPT or Anthropic, you know, through Claude. But there will be some human loop in in both of those contexts, right? Whereas I think with respect to these bottom-up agents, they actually are, you know, agentic. Like I think semantically you can actually use, you know, the word word agentic in that context, right? And they will be fully autonomous and they

will be procuring kind of their own resources and, you know, calling these APIs and and so on. And there won't be a human necessarily in the loop. Like what I'm confused about is who will create them. Will like how do they create them It almost sounds like you're saying they create themselves, but I don't understand that. No, I think like they'll still be created, you know, by a human and they'll still, you know, fundamentally be serving, you know, on behalf of a human. Maybe at some point you have agents spinning up other agents and so on, right? So yes, they will there will be human in the loop insofar as they will be serving a human, but the guardrails will be much wider and I think that's kind of the nuance is is, you know, on these commercial and consumer agents the guardrails are very narrow and if

anything they aren't even, you know, transacting, it's more, "Hey, I'm a research assistant that's going to help you figure out what hotel to book, but I'm not actually going to book the hotel." Like at the end of the day it's going to be you that actually authorizes that transaction and maybe you already have your card details, you know, plugged in. So it's just like an Apple Pay transaction, right? And again, on the commercial side I don't, you know, it's it's hey, it's a Salesforce agent, right? Which is going out and doing automated, you know, sales outreach. That does not necessitate the the agent actually transacting, right? And then similarly, maybe it's a, you know, finance agent that's, you know, automating finance functions. So none of that necessitates an agent actually transacting. Whereas conversely, I think with these with with the bottom-up agents, you know, they actually are

autonomous and they're actually going out and and doing things. And maybe they're interfacing not just with, you know, with you know, these APIs and procuring resources from humans, but they're probably also going to be interfacing with other agents. And I think that's a qualitatively different category. Yeah, I forget which one of you said this, but I guess that example of the data that you're procuring for like a really small amount of money would probably be a good example of that. Um so one other thing that >> [laughter] >> I think both of might have been both of your pieces or or even in this conversation that you both referenced was a lot of times payments actually cannot be instantaneous because of the possibility of fraud. So how do you think that gets factored into agentic commerce and how this is all built out?

Yeah. Noah, do you want to take this? Yeah, for sure. Well, I mean I think, you know, again, if you take back to an example like booking a a flight, you know, I think again, if you go through traditional card networks, you go through the traditional rails, I think you get the same benefit, you know, fraud and chargeback benefits that you get, you know, with traditional e-commerce. I think the again, going back to kind of my my thinking of there will be a whole new class of merchants that are going to get paid on a, you know, per transaction basis. If I'm buying a service for fractions of a cent or a few cents, this notion of fraud I think is a lot less relevant because frankly, you know, if I use a service and maybe it's a scam or it doesn't work, it's frustrating for a

little bit of time, but ultimately I can be okay with, you know, wasting a few cents and, you know, can say, "Look, don't use the server service again." I think where this gets problematic is when I'm going and purchasing, you know, high-value or high-ticket item purchases and at that point I think that's where, you know, for example, the card networks are going to continue to have an advantage. But I think if you believe that a lot of these, you know, it's not going to be these enterprise agreements and major SaaS agreements and again, it's just, you know, I have a a specific project that I'm building and maybe I want to use this tool once, you know, I I don't think fraud is as big of an issue because at the end of the day it's not a big enough amount of money that I'm spending where

it materially impacts me. All right, so in a moment we're going to talk a little bit more about the different types of agents that will exist and how this will play out, but first a quick word from the sponsors who make this show possible. Step into a new era of wealth. Discover Nexo, the premier digital wealth platform. Manage your crypto portfolio with confidence and control. Receive interest on your digital assets. Borrow against them without selling. Trade a wide range of cryptocurrencies, all in one platform.

Now available in the US with 30 days of exclusive privileges for new clients. Experience Wealth Club Premier. Access enhanced interest rates, reduced borrowing costs, and crypto cashback on swaps. Get started today at nexo.com/unchained. Back to my conversation with Noah and Robbie. So Noah, I actually wanted to ask you what you thought about Robbie's like taxonomy of the different types of agents and how you think um you know, that part might play out in terms of like, you know, ones that transact more than others.

Yeah, for sure. I mean, look, I think at the end of the day like today it kind of feels like there's a lot of conversation about having all these different agents doing these different things, but you know, frankly, at least from my perspective, it feels like they're all sort of part of the exact same ecosystem. Like, you know, for example, you can spin up different sub-agents within that can focus on different tasks, but at the end of the day it kind of feels like it rolls up to the to the same product or to the same agent. And so I think what we're going to see is that and and we're starting to see this develop already with sort of this new skills ecosystem that, you know, there are certain sort of almost like character builds that you can have where a certain agent has infinite amount of

intelligence, but when you give it very specific guardrails or very specific, you know, kind of constraints, it will be more useful for that task. I think how that impacts commerce though is is kind of immaterial because to kind of to Robbie's point, if the human or the developer's ultimately the one that's making the payment, it really only matters, you know, if you have one balance. And so I think one of the areas where a lot of the commentary in my opinion has gotten a little bit sideways is there's this notion that like I'm going to have 10,000 different sub-agents and each one's going to need their own wallet and each one's going to have to have their own controls and all these things. And I think frankly a lot of that is a little bit overcomplicated and I I probably tend to gear more towards Robbie's opinion there where at

the end of the day like, you know, if a given task has a certain mandate of how much you're willing to spend, you know, which one how much each agent or participant spends is kind of irrelevant. So I think from that respect, you know, we're likely to see more subdivision between different sub-agents of what they do, but I think from a commerce perspective, it won't be as complicated. All right, so to go back to the fraud issue, I wondered what other roadblocks you see that, you know, need to be solved in order for the agentic commerce space to be built out and and grow.

Yeah, I think a lot of people have talked about, you know, the rails being kind of the ultimate bottleneck and, you know, the thought experiment is like, "Hey, if we can, you know, solve the rails and have instant payments and micropayments, then, you know, the agentic commerce will, you know, it'll it'll completely take off." I think my perspective is it is not the rails that's the bottleneck and it's actually human sort of social structures. And I think like like I don't know that people are willingly going to just outsource everything to an agent. I think like that implicitly takes a lot of trust and that's not consistent with how, you know, decisions have historically been made for consumers. So that would be one example. I think more consequential examples are, you know, bureaucratic inertia, right? Like I think, you know, if you're if you believe that a lot of these agents are

going to be these commercial agents that are ultimately going to be deployed sort of top-down within organizations, you know, there are multiple layers you have to ultimately get through uh to actually be able to sell that, um you know, within an organization. And I do think it'll be a competitive necessity down the road and it'll be almost like a liability if you don't integrate these agents, but I think that actually takes a lot a lot longer than people intuitively think. The other thing is government spending is, you know, a massive part of the economy. Uh and I think you'd be crazy to think that, you know, governments are going to be adopting agents within the the next few years. Like I think that's also going to take a very long time and there's plenty of precedents throughout history that would would also be consistent with

that. Um the other thing is legal and regulatory uh bottlenecks as well. And I think that's another thing, you know, there isn't enough people kind of discussing that. Um where, you know, it's not just, you know, the rails that weren't engineered for machines, it's also, you know, our entire regulatory system, our our entire legal structure. Um and I I I don't think it's it's going to be something, you know, in the same way that like, you know, a protocol upgrade uh is is sufficient to fix kind of the rails. There's no protocol upgrade that can be applied to legal structures, right? There's a lot more, again, just inertia there. So I think that's what everyone's sort of underestimating is is there are so many bottlenecks uh to adoption. I think the technology will be there and you could almost argue it's already there. Um but

that doesn't necessarily mean, you know, everything's going to be adopted overnight. Noah, do you have anything to add? Yeah, I mean, I I personally am more bullish on on, you know, how much how much comfortability humans will get with agents taking off. I mean, I think, you know, if you go back to the early internet days or, you know, there was similar fears I think of, you know, is my data going to get taken? Like, you know, how secure, how safe is this? And over time, you know, I think people got more comfortable with it and the technology and the guardrails improved. And I think people are going to to optimize for what is, you know, the the lowest barrier to entry and the how do they create the most efficiency in their lives even if it means taking a little bit of a risk. I mean, even if you look

at like the Open Claw ecosystem, like there's a lot of setups of Open Claw that, you know, I think are are less than ideal or or not very safe, but people still do it because they they like experimenting. And I think over time as the the guardrails get there, that's how you start to cross the chasm and you get more mainstream adoption. I think for me, like if I was to say like what is the the biggest barrier, at least especially in this this sort of new headless merchant and agentic economy that I kind of talked about, I think it's honestly less about the the rails. I agree with Robbie there and I think it's more on the discoverability layer. You know, if you have thousands and thousands of new potential endpoints that you could be hitting and they all kind of look like perfect complements to

one another, how do you differentiate between them? And so, you know, I think it's an open question of like, is this something that gets developed from a a private company or versus is this a, you know, sort of an open ecosystem? I think we see that that debate happen a lot in crypto itself as as well, you know, with this topic in and outside of it. Um but I think, you know, in order for this to be useful, you know, there needs to be more signal in terms of like actually how, you know, how safe is this? Like, you know, how reliable is this? And I think that will grow the network significantly. Yeah, so on that score, I actually did want to ask about some of the different standards that are out there because um so Noah mentioned this marketplace in his piece and that is powered by MPP.

Um the I'm just like it's merchant payments protocol by Stipe and Tempo. Um and um you know, another big name that people are talking about is Coinbase's X402 or or they developed it. It's like an open standard. Um you know, I think the short version of the difference between these two is MPP have something called sessions, which is sort of like opening a tab at a bar with a budget. It's it frankly also to me it would sounded just so super similar to lightning channels or state channels.

And um you know, then uh but the one other detail about MPP is it's basically at this moment it's more centralized. Um it's it's only settling on Tempo. Um it requires Stripe, but it does connect to traditional payment rails like Visa. And they say that, you know, they will eventually be chain agnostic. Um and uh at the moment they also have Lightspark on it, which is um a lightning, you know, uh payments company with uh Bitcoin. X402 is just more permissionless. It supports any ERC-20 token at the moment. It's gasless for users. Um but it doesn't have the sessions aspect, so you're, you know,

transacting with every request. Um and I just wondered if you had thoughts on like different type of architectures and which ones you think will be adopted more quickly um and and why you thought that. And you can name any others because there's a whole bunch of others like, you know, Visa's CLI tool, the the Google agent payments protocol AP2. There's a bunch. So you can, you know, talk about any anyones. Yeah, no, I think it's a good question. I mean, look, I I I think I'm pretty optimistic on on both X402 and on MPP.

You know, I think obviously the benefit that MPP has is that they obviously have the the Stripe ecosystem behind them. So I think there's definitely a head start there where if you have existing Stripe merchants who, uh you know, want to participate in this new agentic economy, you know, it'll potentially be a lot easier for them to just click a button and have that enabled versus, you know, maybe with X402 you got to go and set up a facilitator and, you know, potentially will require more bespoke integrations. And you know, it's worth noting that Stripe also integrated X402 prior to announce the announcement of MPP. So I think it's still a, you know, it's unclear um how this will develop. Um you know, I think obviously that the the crypto ecosystem tends to prefer more of the open permissionless standards, but I also think that there are some some

benefits you have when you can control the product more. And you know, I think to your example, things like sessions and also the ability to use traditional payment methods is a a huge step in the right direction. I do think it's worth, you know, Visa CLI is uh you know, a new product that Visa's developing where it's it's not a protocol, but rather it's sort of a new uh CLI-based wallet. And so, you know, you similar to, you know, I think some of the other wallets that we're seeing, you can, you know, download it into Open Claw or Clawd Code. You can go and transact at a bunch of different uh endpoints. It's comes with a whole directory of of endpoints I think that they've developed as well as connection to existing ones. And unlike, you know, I think some of the other MPC-based wallets, like it comes, you

know, default with card. Um and so again, I I think at this point, as I mentioned in my paper, I feel like the rails are are largely there. And you know, who wins is is probably less relevant because it's ultimately going to come up come down to what merchants want to uh want to integrate with. I think the bigger question is is, you know, who are the new merchants that are going to come out and um you know, what services are they going to create that is useful to developers to the point that they want to adopt this. I think that's me is is the bigger question and in the bigger opportunity for innovation. And but do you want to elaborate on that? Like what type of merchants you think will come out for, you know, Yeah. Well, I mean, I think there's a, you know, one area that is less talked

about is there's a lot of interesting stuff on the creative side. So, you know, the ability to buy code music, the ability to uh you know, have image generation. I think we've seen a lot of creative things there. And so, um you know, the ability to generate a song with, you know, uh explaining the the vibe and the the mood is is interesting. But I think it's going to expand much more into more general developer services. I think if you you've used, you know, any of these these um coding platforms, um you notice that there's a lot of commerce moments that happen when you're when you're using it, you know, whether you need to get a domain name, whether you need to set up a server, whether you need to get access to proprietary data.

I think there's if you, you know, there's all these commerce moments where, you know, you have to leave the terminal, go and get an API key, maybe get a subscription, then come back into Clawd Code or Open Clawd and and and paste it in. And it's it's a big friction point. And so I think the more that people develop on the kind of the buyer side, it will lead to more inspiration and influence on the seller side. And Robbie, what about you? What um you know, thoughts do you have about which types of platforms will um will do well in this world? Yeah, I think part of the irony at this point at least is a lot of them are backwards compatible with cards. Uh and I think, you know, Noah's made this point. I think it's probably an area we actually do do agree on is um you know,

there is this inertia again where, you know, people are used to making kind of payments with cards. And um again, I think one of the really important things is these card networks do have, you know, all these tokenized credentials, right? So they have solved for this kind of fraud and risk scoring issue um that blockchains just haven't solved for. And I think it'll take really a long time. I think one nuance that that I've been thinking about more is, you know, do these do these tokenized credentials and these kind of, you know, fraud graphs that the card networks have made um and you know, have refined over years and years, do they actually map well to agentic commerce, right? So if an agent is making transaction, you know, obviously that's fundamentally different to to a human. And they're the behavioral sort of patterns will also

look, you know, quite different. So I'm also, you know, curious that, you know, if if we end up seeing, you know, uh almost like new fraud graphs and new new risk scoring, you know, have to evolve for agents. And um I actually think Noah sort of alluded to this in one of his pieces where, you know, maybe that is the opportunity for for blockchains, right? Um you know, it kind of creates this gap where because no one has solved for it, uh you know, then maybe that's something, you know, blockchains could actually solve for. And that's kind of the wedge that they need to ultimately compete with card networks on the agentic side. Oh my gosh, that is so interesting. Um that makes a lot of sense to me because um >> [laughter] >> I mean, I yeah, I just as you know, there's so much crime in crypto. So I

feel like through crypto I've learned a lot about kind of the behavior of criminals. And I remember, you know, um when Katie Haun was a prosecutor and she figured out that there were two separate federal agents that were stealing the Bitcoin that were um you know, what's the word? Seized from Silk Road. The The way she figured it out was that somebody tipped her off they thought an agent was doing this. And then she she thought no, they wouldn't be and she wanted to prove this person wrong. But then when she looked at the behaviors, she was like, "Oh, it's not just that somebody's doing it but there's two people because their behaviors are so different. It was so obvious that they were two different people."

So, um yeah, like with an agent you wouldn't have that kind of I mean you you might but but it's it would just be harder to figure out like what are the idiosyncratic kind of behaviors that, you know, mark a person. Um I did also want to ask um about a thought that I had which was I think something that is a little bit more permissioned you you know, like the MPP protocol will maybe succeed first especially because they already do have so many merchants.

But that like maybe over time XRO2 or or just anything that's more permissionless might do better. Um and I know that MPP is heading in that direction so so it may it may not be that they fall behind at that point. Um but it just sort of feels like there could be an analog to the early internet where you know, AOL was how everybody got online first and then of course, you know, like nobody uses AOL anymore. So, um you know, it it feels like there there is something like that. Like you have this sort of hand-holdy kind of set up to begin with and then then it moves beyond.

Um so okay, so I did also want to ask um like I imagine that the rise of AI agents and you know, as as you're looking at the future of agentic commerce that it's changing a lot about how you think about just VC generally. Like I know for me even just as an entrepreneur, um it's changing how I think about like you know, what types of businesses would would um be viable in this new world. So, I was just wondering, you know, how is all of this affecting either how venture capital um you know, like what types of projects you're looking to to fund um or what types of projects you're even being pitched.

Yeah, I'm happy to go first and then Noah uh comment as well. But I do think a lot of it distills down to sort of value capture and where you ultimately think value accrues sort of within the stack. And I think some something that a lot of people have been saying is, you know, what these agents do fundamentally is they abstract away the entire front-end interface. Um and instead of, you know, uh you know, trading whether it's, you know, crypto and you're trading or, you know, you're, you know, depositing some funds into, you know, these yield protocols. Well, what you're actually going to do is you're going to say tell this agent, you know, go do all this for me. Uh and then you don't need those you don't you don't need the chart or you don't need any of the the kind of interface um on the front end. Um I

actually fundamentally disagree uh with that take and I actually think, you know, especially on the trading side and I wrote a piece a while ago maybe 2 years ago called the fat wallet thesis. Um and the the sort of core uh sort of, you know, thesis there is that whoever owns the end user ultimately captures the most value. Uh and then I think that's been consistent, you know, across literally, you know, every every emerging vertical probably throughout history. Uh and I think it's especially true with sort of in the in the internet era. Um and I think, you know, a lot of people are saying that, you know, agentic you know, all these agent interfaces are going to undermine kind of this fat wallet thesis and and so on.

And um I think the reason it won't is because people actually like interfaces. Uh and I think again this is another nuance that the entire sort of agentic thesis misses which is, you know, if I'm if I'm trading for example in crypto, um the chart itself governs my decision of whether I want to go long or short or, you know, or buy or sell or whatever. Um and I think like that is not something that users will want abstracted away ever. Uh and I think that's actually something they're going to they're going to always want to keep. And maybe it's AI I will have the agent, you know, find me the best yield source, right? Like I think that's one example where maybe it is sort of interesting, right? You know, I have some USDC sitting in my wallet and I say, "Hey, just go find me the

best, you know, risk-adjusted yields." Um and it goes and on the back end, you know, kind of deposits that. Um but I think especially on the trading side and obviously that's sort of been the principal, you know, consumer use case for for crypto is you know, people just want to trade, you know, crypto. So, I think that that agents don't really, you know, disrupt that. Um and then the other thing I would add is, you know, in terms of, you know, okay, how does value accrue then? So, maybe some of this accrues to the front end um and whoever owns the end user. I think you also want to own kind of the settlement layer as well. And I think it's kind of this barbell of where value accrues. And on the settlement layer intuitively there's more there's more of a network effect there where like the more volume that

settles the more volume that settles, right? And sort of volume and liquidity begets volume and liquidity and so on. Um and I think that's where, you know, and I don't know who's going to win and, you know, I think at this point I think Tempo is probably, you know, maybe the front runner uh for for at least all the agentic stuff given um they kind of have the right partners, you know, who are in the mix and they have kind of the the merchant distribution of Stripe and the and so on. So, um but I think ultimately, you know, with respect to VC and how I think about value value capture, I think it's, you know, you want to be on one one one side of kind of that barbell. And actually just one question when you said um like I I forget how you phrased it. Did you

say, "Oh, you want to like own the platform like where people or agents are congregating?" But so right now we are seeing like websites pop up but I I don't imagine that agents will like congregate at websites or So, am I right? >> Yeah. I was more referring in in the the crypto context of, you know, uh sort of wallets and trading and sort of the use cases that people are, you know, um using within crypto. And I think that's where you want to own whoever, you know, has the end own owns kind of the end user. Um I think like, you know, if you're if you're then to extrapolate that out and say, you know, you look at all agentic use cases, um I think that probably looks more like the chatbot interface, right? So, whether that's, you know, ChatGPT or or Claude. I think

that's the other layer that you want to own because again, they own kind of the end user and I think downstream of that there's there's a lot of different ways to monetize. Mhm. Noah. Yeah, no. Well, first of all, the the fat wallet thesis I that was a great piece. Highly recommend anyone uh to to read it. It was it was it was great. But I the one area that I I will kind of push back is is I I gen- I tend to fully agree with you that like whoever is owning the end customer uh will ultimately accrue a lot of the value. But I think what's so interesting about AI and there's actually this really I think interesting crossover where, you know, what we're seeing is that the front end and the back end are becoming increasingly separated and or increasingly separate and you know, the

the parallel is I think I had this, you know, one of the big eye-opening moments for me in crypto I think was when I first used a self-custody wallet and I actually exported my keys and then imported the keys into another wallet and you see the the interface changes but the back end is the same. And so I think there's an a very interesting thing happening in AI where the front the back end is is static but the front end is becoming increasingly more modular. And so for example, like while I agree with you that people do have preferences over what the front end is and how they engage with it, what I think we're seeing with these tools is that it's actually much easier to create your own front end. And so if for example, if you really like a certain trading app, right? There but maybe you

don't like one or two features, what you could do is as long as you have the back end in and the back end's in place then you can very easily customize what that front end looks like. And so I think from that regard, you know, whatever the hardest thing to do is or whatever the biggest sticking point is in the back end in the infrastructure where there's some sort of mo- in in some sort of capability that they uniquely have, I think that ends up being very valuable because ultimately what AI is largely going to do is it's going to make it so that you can build any front end that you want and you know, you can customize it to a much greater degree. All right. Well, this has been super fun chatting with you both. Thank you so much for sharing your thoughts and

coming on Unchained. Awesome. Thanks for having us, Laura. Thank you. We will catch you later.