Why AI Agents Still Can't Make Payments and How MoonPay Plans to Fix It

Man, it's a good day. Let's go. He's on the media tour right now. How's the media tour been?

You know, it's not been so crazy. It's it's not been that crazy. So so you guys are are, I think, like, second or third on deck. Let's go. That's great. So I saw you were on MTS. How was that? MTS was great. Yeah. Lot of fun. And by the way, that intro video was kinda fire. It was really cool. Thank you. Appreciate you. Thank you. So earlier in the show, we were showing this. I'll just pop this up because it's been a crazy

last six months for MoonPay. Talk a little bit about where you guys fit into this stack, and we can dive into Dawn, the, you know, new role that you have at MoonPay and what you're building. But maybe just to we open this can of worms at the beginning of the show of, like, the MoonPay acquisition strategy and how all of these things are coming together into a suite. What was it about that that made you, on your end as the founder, want to join this team and and this ecosystem? I think it's incredibly

competitive right now to win. And the reason why it's so competitive to win is because the rate of advancement of AI is causing a lot of services to coalesce into one interface. You see that with almost every industry. And in particular, look at OpenAI and Anthropic. OpenAI just launched this forward deployed engineer company that is attempting to put AI into every company that exists. Anthropic is doing something similar when it comes to consulting businesses and working with different consulting firms.

You see this, like, really, really big verticalization and horizontal shift in the industry. And the reason why is because AI enables these big companies to move at the pace of small startups while still leveraging their massive distribution advantage. And so to me to me, m and a is, like, the angle by which these companies are aggregating the resourcing and the talent and the infrastructure in order to be able to deploy that vertical and horizontal experience.

And as you can see, MoonPay is really, really killing it with that

that angle as well. Yeah. Yeah. It seems like a lot of these different pieces of what an enterprise or modern financial company would need all coming together. We were talking about the history of MoonPay a little bit and where they've started versus where they are now. It seems like in the last really few months, they have really stepped up in terms of bringing a lot of this together to serve the modern financial organization. Maybe bring us into now what's next. So so Dawn is in the family.

You're, you know, building and leading within this organization now. Talk a little bit about the agent strategy and and what is going on with MoonPay agents.

I was looking here. This seems to be a really core part of the stack in terms of how MoonPay is thinking. Yeah. Well well, Dawn Labs is an applied research lab that that attempted to apply AI to finance, and there's a couple thesis we had at at Dunlabs. The first is that finance is gonna be democratized, no longer accessible to the few quants and traders at at bigger firms.

And I think you're gonna see this plethora of anyone being able to access and test out the same the same ideas that these quants of bigger firms can can do. The other two TCs are that agents are gonna be the primary operators of finance. You already see kind of that shift taking place. A lot of companies are now pushing for agents to use finance, and x four zero two and MPP are examples of protocols that enable agents to use finance on chain.

They're strad steadily growing in adoption, and I anticipate by the end of the year or next year, you're gonna see exponential growth. And the third thesis that I have is that the future of finance is on chain. And so given those three theses, Don and MoonPay, to a larger extent, is focusing on figuring out the right ways for agents to interact and operate on money. The difference between money and other industries where AI is being applied is that money inherently carries a lot of risk in terms of movement.

And because of that risk, there needs to be different sets of permissions and particular interfaces in order to actually move that money. As a human, you wouldn't be comfortable with an agent moving around a million, you know, a million dollars and putting it in some sort of, you know, LP because what if the agent messes up? Even if the agent is better than the human and makes less mistakes than human, a lot of humans would still not trust the agent to do that. There's actually one to one with how radiology works.

In radiology, essentially you have these scans, and you have radiologists essentially are able to see from those scans whether someone has cancer. And computer vision models are actually really good at taking those scans and determining if someone has cancer. They're much better than human radiologists. But still, you need the human radiologists because people want to trust the human radiologists, not the not the machines. Same with cars. You know, cars nobody trusts Waymos or Teslas.

They would rather prefer a human driver, even though Waymo and Tesla's better than a human driver. And so we're gonna see that with money too, and so we have to design the right interfaces so that people trust agents to move your money. Intro us a little bit to OWS

or Open Wallet Standard. Does are you guys calling it OWS? Is that the short internal name for it or no? That is the internal name for the Okay. Intro us to this and what that now means for you and where you're really trying to spend your time.

So, like, finance and AI or I think I was listening to something recently about Menthropic reporting on just their core segments on the enterprise side and where a ton of the volume is coming from and money and finance and those use cases are really driving a lot what of what they're seeing. Where do you you see spending your time from a priority standpoint in terms of OWS

and what those core strategic priorities are for you as you enter the firm, as you build this out now? OWS is actually the backbone behind all our CLI products. The Dawn CLI, the MoonPay CLI, they all rely on AWS as the shared wallet layer locally in order to for the agent to interact with your wallets. So AWS solved a real problem that people had, which is all these CLI tools that agents and humans used to have access to spun up their own wallet stores. That caused fragmented liquidity.

That also caused this problem where people might forget where they place their wallets and lose funds. But primarily, the big problem is you only want to on ramp your funds once to a wallet, and you don't wanna deal with many different wallets in different places. OWS collapses that fragmentation. There's one wallet store. That store is protected pretty, pretty well.

And multiple services, whether it's Dawn, whether it's MoonPay, whether it's D Flow, whether it's Polymarket, and so on and so forth, access that same wallet. Even Tempo can. And so AWS solved this really critical problem where we had incredible wallet fragmentation.

All our services are built on top of AWS. Can we just while we're looking at this fund flow, maybe walk us through this as if you were explaining this to a seventh grader or maybe a fifth grader. Golden what this

breakdown looks like. It seems like a really simplified way to look at it. Yeah. Let's say that you are attempting to tell your agent to do something that involves money on chain. Those that category of things is pretty wide. Maybe it's interacting with the lending protocol. Maybe it's trading a hyper liquid, trading on Polymarket, on Solana, Tempo, so on and so forth. How do you do it? You need a wallet. And there are some tools that you can use to spin up a wallet for you.

That wallet is stored literally in a file on your laptop. And each of those tools will spin up a wallet that is stored on a file on your laptop, and then your agent will open that file, read the private key, and then execute the transaction in order to interact with that lending protocol or that swapping protocol. That's really insecure. You don't ever want your agent reading the private key.

You might have noticed, like, there was news today that broke about how there's an incredible number of hacks that just happened with a lot of NPM packages. And I think that behavior is just gonna become increasingly common. In general, you want to limit access to critical elements that hold money as much as possible. And so that previous paradigm is is not good enough. What OWS does is it manages the wallet part completely. All these tools that people develop, they don't need to think about the wallet stuff.

OWS has already figured that out. OWS already has a great security model. It already has a great storage model. It already has implemented a policy layer. A policy layer is essentially a layer saying, may only create a transaction that has maybe a daily limit of a thousand dollars, and that limits the surface area of attack. And so OWS just has the full set of primitives that any tool could ever want in order to manage wallets.

And that enables any person, nontechnical or technical, to use a service like OpenClaw or and in my opinion, I'm a big fan of the Codex desktop app in order to interact with these tools and do, like, really complicated things by just telling their agent to do it, and and also

know in their heart that, you know, it's pretty safe. Yeah. Because that's that's where people's mind will immediately jump when you sort of say Open Claw and my money. Right? And the larger sums you get that, the sort of hair on the neck starts standing up a little bit or codex or whatever agentic use you have here. But it's interesting talking about all you're saying, like, the protocols, the policies built in here to sort of prevent any worst case scenarios.

I can see that being a a big trust barrier to overcome, but I love to hear how y'all are addressing it. Look, I think that we're gonna see, like, news on a lot of hacks in the in the next couple months when people's money is being arbitrarily deployed on on on random vaults, and I think AWS is is here to solve that problem. And so you see in the demo video that you just watched, AWS is the infrastructure layer on top of which MoonPay agents works. Yeah.

That's that's where I was gonna go next is sort of the so, like, right now, you know, on the individual side of the moonpay.com

website, this Moon Agent's card is kind of front and center. It's it's a way to get involved. It's a way to bring some of this to life. Talk a little bit about this product and maybe maybe one, like, where this OWS ecosystem under it, how it's fitting in to this specifically, but then also the what I'd imagine to be the opportunity for other like products or similarly architected products this represents

for other builders or other teams. And I'm I'm just curious, like, more on this card. This is the first time I'm seeing this card. This is sweet. It is really, really well designed. It's a stablecoin debit card that operates on on your OWS wallet. And the way it works is you spin up an OWS wallet, and you generate a card that spends directly from the USDT on that card. You can use that card anywhere.

You can set your approvals on the card via the blockchain, via these approval transactions, and then you can take that card and put it in Apple Pay or Google Pay, or you can even just give it to a browser use agent, and the browser use agent will then take those credit card credentials, go to a merchant's checkout page, input the credit card credentials without any input from the user, and do that entire process headlessly. So the big problems that we've seen in finance is finance is made for humans.

Checkout pages are made for humans. Currently, the payment methods are cards, and cards are also made for humans. And so we built MoonPay agents card in order to connect interfaces made for humans with interfaces made for agents. This effectively this product effectively allows people with OpenClaw or people which use other clients like Perplexity Computers, Zoho Computer, pope.com, and so on and so forth to use their to spin up a credit card that's linked to their OWS wallet and

spend it headlessly. Right. So this card and how it's positioned seems really focused on and I'm I'm in the kind of individuals individuals part of the website. What is your view on because I know Sodot Sodot was a huge acquisition lately building into the institutional narrative and the business side of MoonPay, which we can get into. But where do you see the differences here for, like, more institutional enterprise b to b use cases versus the individual ones?

For us, for our business, StableDash, like, I'm even thinking right now is, like, how can we start getting involved with spinning up things like this? Are things like this available today for enterprises, or is this more so focused on individuals

at least right now? And how you just think about the the differences there. Frankly, you just saw our b to c, and I think when you think about what b to c and what b to b means, I also think those definitions are collapsing with with AI agents to some extent. There is a big b to b project that we're working on when it comes to AI agents that we think is gonna transform how agents interact with money.

Even when I think about OWS, Open Wallet Standard, and when I think about MoonPay agents and when I think about existing CLI products, I still don't think that retail and businesses can effectively let agents access money from any client. And the reason why is that agents are extremely capable when provided a file system as a tool. The reason why is because agents, foundation models, OpenAI, and Anthropic, train coding specific models.

These coding specific these domain specific models applied to coding are really good at using a file system and file system commands in order to operate, which means that we've built all these tools that operate on the file system. But it turns out the clients people use don't have access to a file system. The most popular clients that people use are actually mobile apps, like ChatGPT and Claude. Those mobile apps, their back end is a database. And there are a couple solutions to this.

One, I think the coolest solution is virtualized file systems. Virtualized file systems are this concept where you take an interface. The most popular thing, think, is called Just Bash. It's authored by Roussell. And essentially, that interface has implemented all the Bash commands and transforms them into something that interacts directly with the database. And so you get a database is speed and low cost, as well as a file system's, like, command structure.

This relates, like, entirely to how we're training foundation models. Everybody uses Codex. Everybody uses Opus. These foundation models are trained almost entirely on code. They're not trained on other domains. That's why, as software engineers, we think these models are so intelligent and so cracked, and actually people in other domains are like, wow. Wait a minute. Like, these models are kind of stupid. And the reason why is because they're domain specific models.

So we Where does your mind where does your mind go with this? My mind's going a few different directions.

When you had that site up with the MoonPay agents card, I was just looking at the few kind of listed use cases, you know, shopping and errands, travel booking, subscriptions, recurring spend, business purchasing. I thought it was interesting to think about the collapse of these b to c and b to b use cases. Even even what you just said, Niraj, about the sort of, like, software engineers, I think these models are so smart.

Other people in other domains are kinda like throwing out something like agentic payments, giving an agent your card can seem kind of far fetched. But then you think, wouldn't it be nice to have an agent recurringly buy groceries for me? Think about you managing a household. You got three kids. You're all this. You know? And then it's like, of course, all the b two b payments that are recurring that need to be handled that you get behind on and all this.

It's having the ultimate assistant, and it feels like the gap in sort of handing this off is so right. I don't know about you guys, but if you rewind a couple years ago, I'm remembering offshoring, hiring virtual assistants, and authorizing somebody from another country to say, like, here's a credit card that I had some rules on, and I would watch and all that. But I trusted this person, but it's like, I was doing this with another human.

And, Jay, you're pointing out, like, maybe that human she was very trustworthy, but maybe it's like, can we really trust humans as much as we could trust these agents anyway? It's like, just the natural progression. An assistant, quote, unquote, that is agentic can only be so effective without the ability to transact. And it's taken it in my mind from, oh, this is far fetched out there.

People just wanna talk about agentic payments because it sounds sexy to, like, of course, this is coming, and it's coming really soon. And I can actually see the adoption use cases becoming more real and

more accelerated quickly. Personally, I'm thinking like, yeah. How can I start using this? Yeah. I mean, that's exactly right. Maybe question question for you just on the opportunity here.

So we have, like, when I was entering my career, it was sort of the reign of, like, the API software sales world where, like, APIs were just dominating and, like, that was the paradigm tech was in, where it's like, you sell an API, you process that you you price that on a per call basis, and then you can build extremely scalable businesses that are super profitable based on top of that.

And we have seen, like, the API paradigm for the last, like, ten, fifteen years really reign, but there was a very human involved paradigm. Where do you see this next it almost feels like we're kind of at the dawn of something oh my gosh. I said it. The the beginning damn. The beginning of something new. Great branding. Good god, Raj. I guess that clicked for me in that moment. But how do you how do you see what what now happens? It seems like there's so so much that is now possible.

But how should people be thinking about what's possible in this, like, agent native economy?

Human native methods to pay for APIs involve these things called subscriptions, where you pay a lump sum for some lump sum of credits, and then those credits run out, and then you pay another lump sum for another lump sum of credits. And now the paradigms introduced by MPP and x four zero two mean, instead of paying lump sum, which is inefficient, you may not use some credits, you're paying on a per credit basis.

And you can do that because the existing blockchain technology is fast enough and cheap enough to support that. Yeah. I think that will be the case. I actually don't think it's, like, an incredible amount of friction. I don't see it as the, like, most transformational outcome. I think, again, like, we can fit these building blocks of human native methods of payment and and agent native methods of payment together forcefully, if need be, with browser use and computer use and agent based credit cards.

And agents can still operate in this in this world. To me, I actually think that the harder problem is not getting agents with payment credentials to pay for something. It's actually granting agents the payment credentials in the first place. That's still there's still a lot of friction there, and there's still a lot of, like, trust assumptions that haven't been solved. And so I think that the players that are able to effectively give agents payment credentials easily are going to, you know, win massively.

That's Mhmm. That to me is like What's so difficult about that currently? Like, why is that such a barrier to entry? I'll give a simple example. ChatGPT or Claude mobile apps. I don't know how to give my payment credentials to either of those two apps. I think it's literally impossible right now. And that to me, that shocks me. It means that people with access to the two most popular artificial intelligence apps of all time cannot cannot pay whatsoever. And there's really no avenue to do it right now.

And what we're working on at MoonPay, with all these acquisitions and resourcing that we have, the regulatory licensing that we have, is to give the ability of any client anywhere, on mobile, on desktop, on different hardware services, the ability to retrieve scoped payment credentials.

Do you think that's something from from those two companies specifically? They're obviously thinking about this. They're not, like, aloof to this opportunity. What what is the putting yourself in their shoes the reason they may not be doing this right now? Is it a risk based decision? Is it a build complexity that they're facing? Is it something they want to be really thoughtful over time to introduce? Obviously, it's not there today. Why do you think that is? These companies are really taking

territory of vibe coded swap. So if there's any company that can be vibe coded swap, and you're okay with that, then OpenAI and Anthropic are almost certainly gonna win. You see that with primarily, like, with data related companies, and you see that with industries where if the data is wrong, there's not an incredible cost to that. However, when it comes to payment credentials, there's actually an incredible cost to being wrong or being hacked, And so I don't think that you can vibe code this.

On top of that, there is actually like a lot of government regulatory licensing that is required, like things like money transmission licenses and and other stuff in order to effectively enable payments. So, yes, they may be working on it, but it's not something that they can easily work on without a partner that specializes on payments. Can you speak a little we we talked we reported on this at top of the show around

the history that that MoonPay has had, the regulatory capture, and just robustness on the licensing side that they have because they've been around for so long. Can you speak a little bit to that and the kind of advantages that that provides in this conversation? I know we were we were talking specifically about Max von Wallenberg as he was thinking about the idea of selling the business and and do going through this.

One of the key things for him that he said was just the depth of regulatory licensing and and things like that that MoonPay brings and how fast that can accelerate certain businesses who are younger or are in the startup world. How are you thinking about that in terms of your decision to partner with MoonPay in this way? And and what does that look like for those who aren't familiar? I mean, Max has such an incredible background in finance. He really knows what he's talking about when it comes to

regulatory licensing and and all the blockers that's required in order to get a finance app out. In my opinion, I think there are four moats that exist in launching any product today. The first moat is technical talent. The second moat is capital, money. The third moat is regulatory licensing, and the fourth moat is distribution among users.

And I think that OpenAI and Anthropic and other foundation model companies are able to consume or subsume any businesses that only rely on the first moat, the technical talent moat, because it turns out foundation models are now better technically than most people. However, the latter three moats actually meaningfully still exist. The latter three moats being

capital, regulatory, arbitrage. Was like they're I was forgetting that one.

Yeah. And, you know, maybe capital's not as relevant. Foundation models, companies are very, very well capitalized, but regulatory moats and distribution still matter. Regulatory moats, they just take several years, and a lot of time, and a lot of trust in order to develop and gain. Also, back to the capital mode. It's not, you know, as simple as just having capital. It's also thinking about effectively moving capital. Think about on ramping and off ramping.

It's actually like an incredibly complicated thing to do despite seeming simple. It involves a number of regulatory licenses. It involves like complicated movements of money. There's a lot that goes that is involved that is way beyond just the technology that that perhaps can be vibe coded. And and when you think about MoonPay specifically

in those three kind of last areas, capital, regulation, and distribution, especially in those kind of complex areas around on ramping, off ramping, things like that, how did how did those strengths play into you personally being excited to partner with them in this way? One, I like that

at this company, we're not afraid to take on the biggest players and the biggest problems in the space. You see it from our acquisition strategy. We genuinely are playing to win. I think that's really important. Back to your question, though. I think at MoonPay, we have all four moats. We have the technical talent, especially through these acquisitions. We have the capital. We have the regulatory licensing. We've had the regulatory license.

We've we've always had a lot of regulatory licensing, and we've also continuously acquired companies in order to get that regulatory licensing. And the fourth mode is distribution. We already have a consumer app that that is very, very popular, and with all the new projects that we're launching, we're we're also have, you know, incredible amount of mindshare. So I think MoonPay, as far as these four modes are concerned,

is very, very well positioned. Maybe my last question here would be just who are those big competitors or big organizations that you're going after? Like, when you even think about the opportunity that's ahead and who is running that race, maybe specifically with the agent side of what MoonPay is doing, how do you think about the lane that you're in, and who else is perhaps in that that you're trying to go up against in this way? I'll preface this by saying I hesitate to use the word competitor.

I think, like, the financial payment space is extremely large, and the pie is massive, and there's a lot of complementary things that we're doing with everyone. I would say, like, there are a couple players that maybe matter in the Asian payments space, and those players would be Stripe, Coinbase,

MoonPay. There you have it. Yeah. It is an interesting dynamic. You point that out. Even even on a sort of vibes level, one is a huge pie. There's a ton of opportunities for collaboration and integration as so it's sort of like competitors versus collaborators. There's also kind of this these are people who have moved early. I feel like there's kind of a game respects game. It's like, yeah, we might we might be up against each other in some ways, but there's also I see what you're building.

I like how you're taking this thesis That your answer even kind of reflects, I feel like, the vibe of the industry. Yeah. Hey. Look. There there are no, like, eternal enemies or no eternal friends, I guess. Niraj, anything we anything we've missed on this or not covered or not highlighted

in terms of the acquisition,

what you're building next, or, like, final thoughts from your side as we as we You know, my other disclaimer is I I think it incorrectly says in the in the tagline that I'm a business officer. I'm not. I'm an engineer. So take my business with a grain of salt. There it goes.

And then I think the second thing I'll highlight just as this show is coming to a close is we are launching a trading competition with the Dawn CLI that gives nontechnical users the ability to trade, and I'm kind of excited to see what people come up with. The trading competition's likely gonna start this week. What does that look like? Like, what do you what do you mean? So, like, we're both Zach and myself are nontechnical. We probably fall in that bucket. What what do we do? Where do we go? What is at stake here?

Well, here's what I want you to do. One, install the Codex desktop app. If you haven't used the Codex desktop app, I'll do some free advertising for OpenAI. It's incredible. It's really good. And I think it supercharges nontechnical people to work on technical things without seeing any of the technical complexity. So one, install the Codex desktop app. Two, go to cli.don.ai, and copy and paste that one liner on the website into your Codex app. Click enter.

And that's really all you need to do in order to start launching any strategy that will automatically trade for you. The nice thing about Dawn is that it enables you to trade with paper money. Paper money essentially means you trade virtually, and you're you're able to see how well these strategies do without any without risking any real funds. And then, of course, you can trade with real funds. It reminds me of a finance 1,000

at college. Remember this class where they, like, gave you, and at my university, they gave us, like, a, yeah, fake account with, like, who could who could earn the most. It's a great way to learn kind of getting that. What is the Claude equivalent of Codex?

What's that called? There's the Claude desktop app. And how do you compare those two, and why does why is Codex in your mind superior for this use case? Because I think for us, our organization is pretty Claude heavy. Like, we've we've done most things on on Claude so far. Yeah. Claude desktop app is also really good, and it will work in this case. You just have to go to the Claude code tab on the Claude desktop app. There's three tabs. There's the Claude chat app, the Claude covert tab, and the Claude code tab.

Go to that third tab, and then you can do the same flow. The reason why I think the Claude desktop app is slightly worse than the Codex app is it imposes too much structure on the agent. The structure is the fact that there are three tabs and three different actions. Because agents are intelligent, you don't need to impose structure.

We saw this early at Dawn where we tried to impose specific coding frameworks and specific tools and specific abilities into the agent, and the agent actually prefers to be flexible in this case. And so Codex automatically detects if you're asking a question, you're trying to interact with documents, or you're trying to do a task that might involve coding. And that's completely extracted away from the user.

And I'm sure Cloud Desktop is gonna copy them in a few days, but you can you can completely do this from Cloud Desktop. We're in the we're in the world of days now, which is crazy. What is the reward for this competition, or what is the incentive for those who win? There is, I think, $7,500 total in prizes, maybe more. Don't quote me on that.

Sit down dinner with Ivan.

Definitely sit down dinner with me.

I don't know if it's that valuable, but yeah. Man. Well, thanks for taking your time on such a huge week. We we're extremely excited to, one, continue covering the intersection of, like, agents and money and where this is all going. I think for us, there's a huge opportunity to actually carve out more, like, intentional programming specifically around this and and really help people get up to speed and get involved and on base with some of this.

So, yeah, when we think about companies to work with and partner within that, MoonPay is top of that list. I think you guys are doing some really cool stuff. So thank you for the time, and congrats on such a it's funny thinking back to the first conversation we had where I don't I don't know what the timeline was and when this was finalized or what, but it's nice to have all of us out in the open now and

excited for the future that's ahead at MoonPay. Thank you so much, and we love you guys.