Crossmint’s Agent Vault: Save Your Card Once and Give Your AI Agent a Budget

You're watching stable dash live, the modern money show. We're bringing on Alex here from CrossMint. Alex, good to see you. How's it going? Good to see you guys. It's going great. Welcome in. Welcome in. Head of Agenic Payments at CrossMint. What an exciting role. I feel like CrossMint has, in some ways, been a huge part of the narrative even around this, in my view, just like a company that's really been pushing the needle, pioneering a lot of things. So it's it's awesome to be talking to you.

Excited to dive into what you're seeing. When I think about CrossMint, I think about probably the first stop on the journey a lot of developers and builders will actually go to first to to be like, hey. Here's what I'm thinking. What do you think? Help me think through this. So I can't even imagine, like, what a typical day in your world looks like, but excited to learn a little bit from you today.

Yeah. I mean, it's it's really cool. Like, we've been shipping agentic products literally since 2024. We put out this thing we called the the great on chain, agentic toolkit. And but even even despite that, like, when I joined Crossman about five months ago, I was like, my job title really about to be the head of agentic payments? Sounds kinda crazy. And, you know, for me, it's like, yeah. Awesome. Like, I always wanna go where it's like, maybe this thing might not actually work.

Like, that's actually more fun than, like, oh, yeah. This thing's working. Like, let's scale it up. Yeah.

What is when when you think of, like, head of Agencik Payments and I love talking to people who was kinda, like, pioneering a new role. There was the go to market engineer, head of Agencik Payments. People were head of the first head of digital assets. What does it mean to start, like, taking on this role? And when you think about the problems, like, what is your mandate within the company?

Yeah. You know, it it was pretty broad when I came in. And, because even though it's still so nascent and there's very little, like, initial volume in the agentic space, it's growing, but it it's still pretty small. That volume to us really takes three different shapes. One, you have the, like, what what people are typically calling the machine to machine. This is x four zero two. This is MPP. An agent is paying for some resource it needs for itself. Two, you have the consumer assistant piece, which is blowing up.

And three, we have the, like, theoretical still, mostly business, like, agents that can handle payroll and vendors and procurement and stuff like that. And when I came in, it's still very much blue ocean. Like, should we try to tackle all three? Should we steer towards one? Do we wanna make a bet on some order in which these things happen? A lot of what I'm doing is working with the founders and and really strategizing, like, what do we have today while we do wallets? Okay.

We can do some stuff in the next four or two world versus what do we have to build and, like, what order do we wanna do things? So we very much treat the agentic team like a startup within Crossman. We're doing a lot of other great stuff on on ramps and stable coins, but, like, this is a startup that gets to chart its own path. And Mhmm. So I've really approached it that way. Like, we're staying nimble. Like, we're ready to move. Yep. We we also have to have some opinions about where we think the market's going.

Yeah. How do you approach that, like, personally? You know, you have a history as as a founder working with Demo. Yeah. I imagine it's it's it's fulfilling to an degree to be able to join a company and still have that founder mindset. So unpack unpack that a little bit more, especially maybe in the context of today's news. Because I think part of what you're working through is like a chicken and the egg problem.

It's like, do people not use pay for stuff with agents because it's hard right now, or do they you know, which comes first? Do they need the path? Do they need the tools like what you announced today? And maybe you can break that down a little bit. Or is it something else that we need to lay the foundation for when it comes to using agents and then the one pay?

We are believers that and there you know, there's a lot of stuff that will have to exist for agent payments at scale. There are things like know your agent, you know, protocols. How do you identify the agent itself? How do you identify the human behind the agent? A big unsolved problem in this space is how are we gonna handle risk, disputes, chargebacks? There are many unsolved problems. We cannot wait for these problems to be, like, solved on paper and in the diagram before we start shipping this stuff.

And what I look for as a founder is, like, are there people who are trying to build this stuff, like, trying to get from point a to point b regardless of what, you know, all the tools that are out there? And so we saw that. Like, even six months ago, even a year ago, we saw people who are out there saying, I'm trying to get my agent to pay something. Maybe it's a bill. Maybe I'm trying to reorder my coffee or buy myself lunch. But I'm you know, it's kind of the classic, like, Chris Dixon thing.

Like, what people are doing on their nights and weekends is gonna become a big category down the line. And so when we saw what those builders were doing, we knew that eventually this is gonna become a product, and we just had conviction that there's gonna be a thousand AI assistant agents of all different flavors. They're all gonna need a wallet. And until a month ago, maybe that looked questionable. How many assistants had a wallet? How many agents had a wallet? They didn't. And then all of a sudden Yeah.

Instinct reviews came out right away with wallets and it was like, okay. Yeah. Of course, agents are gonna have wallets. Like, every agent's gonna have a wallet. So for us, like, the the product that we just announced today, this sort of universal vault, it's actually an expansion of the agent cards product that we've had. And really, it just makes it more flexible and more useful to agent builders.

So previously, we only supported cards that were compatible with the newest agentic protocols, these intelligent commerce, Mastercard, and agent pay. Today, we now can support saving any card for later use by the agent. It's it's just like, what can we do now that gets people from agent to transacting? And then we'll fill in all the other details around it over time.

So what are we what are we seeing happen here in this visual? This was the kind of demo visual that you guys shared on x just a few hours ago. Like, walk us through this as as it's showing Of

course. So this is a a hypothetical agent called Impulse. And what most agents are doing today is if you are a user and you want to buy something, they will prompt you and will say, hey. Go ahead and use Stripe link. Stripe link is a pretty great product. Like, it works really well if you hook it up to an agent. I've used it extensively. We think that the thing that's missing one of the things that's missing is a more white label experience. It's sort of like product one zero one.

Like, if you can avoid kicking the user out of your product surface and into some third party app, you should do that. And so what you have here is basically a demo of how we handle that today. So you can save a card directly to the agent's vault itself. It's all white label if you're an agent builder, and then that card is accessible for later purchases.

The other thing that's really nice about this model is the cards themselves are gonna access these new protocols, which instead of minting, like, a new virtual card every time, they handle the card creation and the card numbers a different way. That's a little bit better for the merchants as well. So in our mind, this is ultimately gonna be the future of what all agent payments look like, a white labeled vault, a white labeled wallet, and a cleaner way to transact with merchants.

And what was the hit? Like, what was typically happening before this? Like, how do you compare and contrast where cards enter the scene, where they where they get issued single like, we've seen, like, okay. Here's a dedicated card. Put a balance in it. Or here's a single use card for a single transaction. Like, where do you see what what have these approaches been with the intersection of, like, cards and agents, and how does this compare or contrast?

The very first approach was you had a bunch of, you know, brave souls just give their card number to their own personal agent. These were probably the open clock people and whatnot. And I don't know. Forget the numbers, but they were, like, studies and whatnot, and something like 10% of these card numbers ended up leaked. And so, obviously, that couldn't continue. There there have been a few models. So one is, like, basically, the, like, issue a new virtual card and, you know, Stripe does this, Ramp does this.

Basically, each time you want your agent to transact, it creates a brand new card number, and then you you basically give that agent the card, and it goes and spends money. That's nice because it actually you can limit the amount of spend on it. You, in theory, could do this with Stablecoins. Although, in our view, like, moving stablecoins to a new wallet, to a new card every time you wanna do something like that, it it's fine. It just adds some steps. Although, we're working with some new banks to do that already.

The way that the what we have built works is these protocols, these intelligent commerce, Mastercard, agent pay, they create basically they call it a a card token. Now we're not talking like a blockchain token. This concept of tokenization has been in the payments industry for a long time. It's basically like, you know, a packet that has information about the card. It's tied back to the original card number. You can pass this through the payment network without having to expose the original card number.

Importantly, these new agentic protocols, they stipulate a way that eventually, not all of this is built yet, eventually, we will pass information about the agent itself even if you want about the human behind the agent who's making the request, which is really important if you're gonna start to solve things like disputes, and it's something the merchants really care about. Merchants don't want a future where they get millions of unidentified agents buying from them.

They wanna know it's Alex or it's Zach or, you know, whoever it is doing the transaction. So for us, we're just using the cutting edge of, like, how you create these these one time use card numbers in this, like, agent token way.

So it's am I understanding correct? So it's it's you could put in your card information Yep. But then it's a it could store it in the vault, but then the end merchant is not storing any of that payment data. It's No. One time, done. They don't see it. But the agent can still reuse it, or do you still have to revisit each time and and upload those? Is it almost like a Stripe link experience for whatever agent you're in kind of?

You save it once. Yeah. So it's a Stripe link like experience, but white label. You save it once. And then after that, as a user, you will create what are called order intents. This is a common term in the industry, and the idea is you're linking, like, the creation of this one time used card number with your intent as an individual to, like, buy x. That's really important because it will eventually tie into, hey. How do we handle disputes? How do we know what the person wanted to buy?

How do we know if the agent got it right? Things like that. The cool part of this is you can these order intents that we leverage with Visa and Mastercard, they they are not they can be time bound, but you can have longer lived ones. So if I tell my agent, hey, you know, monitor the price on this jacket and then buy it when it's, you know, 50% off, my agent actually could execute that. I could give it the money ahead of time.

And so we're incrementally sliding from, like, one purchase, one approval, one card to I'm a start to give the agent a budget, and I'm gonna give it a few things to look for, and maybe it'll have a cap on what it can handle.

But that's not just a tech problem. It's a behavior question. Like, when are people gonna be ready to start doing that? I'm I'm I mean, I'm right there. Like, I've been really enjoying Rockbot. It has my agent cash. You know, I'll load up some stable points, but I feel okay. I'm like, at the end of day, whatever. You spend you spend $5 and zero it out. Not a big deal. Do I wanna give you my link? Even though I can give you my credit card through link, do I want to do that? I get a little still apprehensive.

I wanna take more of, a builder approach to this conversation. You know, when somebody looks about building a business around Crossman and being able to embed these payments into Agencik, is this is this the broad use case personal AI assistant that we're kinda seeing more and more recently, the Grock bot them uses the instincts, or is your opinion that we're actually gonna go to more of a specialized agent use case, which we're seeing pop up in more and more apps, you know, the Amazons of the world or Yeah.

Whatever it may be that's like, hey. I'm I'm your agent to do this thing really well. How do you think about people building businesses now around this technology?

My take is Big inhale. I've just been going to you know, it's just constant, man. No. No. My take is I I think that the, like, the consumer experience around assistance is going to look less like I have one assistant that does everything for me and more like what my what the average consumer's phone looks like today. So the average person has about 80 apps on their phone, and the top four apps constitute something like 80% of their usage. I don't I think that'll be high.

Like, I don't think most people are gonna use 80 assistants. But I think, you know, you'll probably have a stable of different assistants, and you'll have a variety of specialists that you turn to in certain instances. But you'll have one, two, maybe three assistants that are, like, your go tos. I think somebody like a muse is obviously, like, super well positioned to be a generalist. Instinct's gonna take one of the biggest swings in recent memory and try to try to go that route as well.

The vast majority of the agent builders that I speak with are looking for a little bit more of a wedge, a little bit more of an angle that's, hey, we're just general. Maybe it's we help you with travel or we are agentic shopping for collectibles is a really interesting area. Obviously, that market's going crazy. Like, you know, if you can be the intelligence layer for collectibles and then also, hey, I'm buying you this sick Pokemon card, like, nice. Now you've carved out a niche.

Wellness is one that we hear a lot from. Like, I'm building the wellness agent and and that comes in many different flavors. And okay, now you're my wellness agent. Where am I buying my lunch to get my macros? What supplement am I buying? So I'm definitely a believer.

People are gonna use more than one agent and that that specialization is gonna create lots of opportunity for these agent builders to carve out a niche, and it's gonna just increase the number of wallets and, you know, transacting agents that we see out there in the world.

What what would be the quick tldr on Muse? This this seems like one of the newer progressions in recent weeks with Meta. Yeah. What what's getting so much traction about this? Like, where where are you seeing this? Obviously, because it's Meta, but, like, what what's been your experience so far?

Yeah. Setting aside distribution advantage, I personally just like that there's more of a UX. You know, iMessage, it's great. I like to text my instinct. I like to text my Hermes agent. Like, it's nice, but there's more that you can do if you get the user into an app. And Muse has done a really nice job. They have this little, you know, like, the news feed. I think the way that they communicate, the way that the UX shows, hey. I'm thinking now. I'm writing you a response.

The little little Muse guy, like, types in his little window up there. Like, the UX is crisp. Now it did, like it has thrown up on me a few times when I asked it for pretty simple stuff. Like, it did, like, pasted walls of, like, unintelligible text to me, which this thing hasn't done yet. But, you know, I I I this is, I think, a classic example between, like, there's no right and wrong, you know. Like, some people want a little bit more UX around their products. They want a little bit more UI.

Some people are like, no. Give me the minimalist. Give me the straight text. And so there's no right and wrong here. Facebook's gonna nail the UI, like, that's what they're already doing. Have you guys tried it? Like, have you guys, like, tested it all?

I haven't yet. No. Zach, what what's your No. Yeah. My I was actually thinking about this earlier. We had Karim on from Entendre or Entendre. Gosh. I can't get it right. You have. And we were talking about benchmark. He was benchmarking all these models. And, you know, there's granted this is for accounting, and there's, you know, 2% variance. And the question I ask myself is how much does that really matter?

Would these models be better the more context you give them, and is it better to stay consistent with a model, build in an agent, give it context? And, granted, yes, be agnostic to a degree so you can take it into different harnesses. Yeah. I haven't gotten on user instinct mainly because this this cost of switching mentality is Yes. I don't I don't wanna spend thirty minutes, an hour trying to figure it out.

I just did that, you know, two, three weeks ago with GrockBot, and that took me two or three weeks of of seeing GrokBot pop up on my timeline. So I'll probably do it. And I think that's that's the paradigm, and that's the struggle I have personally, which is why not just stick with what I know is working is or is it gonna be that impactful that it's worth spending an hour or two hours getting the hang of when, I feel like I'm just now hitting my groove with the other tooling that I also recently just started to use?

So I think that's a really fascinating point. And I you know, testing agents two months ago, I the thing that was becoming like, the standard, I wanna say, like, two to three months ago was, like, hey, I'm your new agent. Give me a call. Like, let's do a thirty minute interview. And, like, that was how people were trying to solve the time to value problem was, like, let me get as much information from you as I can upfront so that now my assistant the assistant can be as helpful as possible.

And I think what's really interesting is that, like, Instinct and Muse ask you next to nothing upfront anymore. Now you could argue, alright. Muse has everything about me from Facebook and Instagram. So, like, they probably have a bunch of context. But, like, the model is now how do you, you know like, anything. Again, like, we tend to think of all this AI stuff as, like, oh, it's so new. It's all brand new. But, like, the product values and principles are the same. Lower the time to value.

And, like, Instinct and Muse can deliver value to you really, really quickly. And that's you know, you wanna build up the context. You wanna know the customer better over time, but you don't want them to have to spend hours, days, etcetera, like, speeding the AI brain about yourself. And so I think that that problem is getting better, which means the switching costs are getting lower. However, you know, this little stuff will probably be like a one password.

I'll never leave one password because I have 5,000,000 things in there. Like, I'm sure there's an easy way to export and whatever, but like like,

I'm stuck there now. So who knows? Maybe you and I will be stuck in the people we're the AIs we're using in five years. I I will say, this was a pleasant one. You know, I had I've been on the Claude train for as long as I can remember now. But then when Astra came out, I was like, I really need to give this a shot. I'd actually deleted ChatGPT, started a new subscription, and then when I was going on an onboarding, it actually says, hey. Do you wanna just import? I pretty much I don't know.

I would I would love to revisit this because I kinda breezed through it because of actually how easy it was, but it pretty much was like, I see you're doing this in Claude. Do you want me to just import all of these skills, all of these MCPs, and everything here? And I was like, actually, that's exactly what I would like. And it's actually synced now. So I will say that is one area where I was like, oh, okay. That wasn't as scary as I thought it was.

I I do think it's interesting. I I look. I don't remember where I saw this on Twitter or whatnot. Who knows if it's true? But I read something that, like, instincts one of their unique approaches was not to try to remember everything, but to actually forget 90% of the stuff that happens, which I think is fascinating. Because I've I've always thought it was really funny, like, what we were trying to do with AI, which is, like, telling you to remember everything.

Because, like, you don't have to go very long before you've told your AI contradictory things. That to you, as a human, you can square how those things are contradictory and in what situations each statement apply, but an AI can't. And I have been thinking about that for a long time, like, how are people gonna solve that problem in memory? I also think it's really interesting that, like, as a human, like, you just you'll be walking down the street and remember something random.

And, like, that leads you on a train of thought that takes you somewhere else. And, like, when it like, what's the corollary to how agents handle something like that today? So I think that, like, the idea that I'm gonna have some vault with all my context, every message, every prompt, everything I've ever asked in AI is, like, probably incorrect in the long run. Like, I've written skills I've don't use. I've made NCP connections I don't use.

Like, filtering out what matters to knowing Alex or or, you know, anyone is gonna be, like, what separates the best agents from the all sororants.

So bringing this back as we close Yeah. To CrossMit a little bit here. When you think about the specific, like, focus right now, you said you guys are you're thinking about this from a like, ex founder, operating like a founder within CrossMint. CrossMint is building first party a lot of different things. Where are you really seeing the suite, the full AgenTic payment suite here? I'll pop it up again because I think it's really Yeah. Yeah. Helpful in terms of seeing all these different things side by side.

You have wallets, AgenTic cards, ramps, APIs. Like, where where are you seeing this? Yep. Are you seeing these right now live standalone successfully and you're trying to piece them together, or are you seeing more customers or, like, developers not really know what they need and they're kinda just, like, picking and using a lot of this? We got we got pinned earlier in the show or I got pinned earlier in the show for being a little bit of, like, the go to market bent.

So that's kind of what I'm asking around is, like, where where does this come to fruition on the on the customer side?

We do occasionally have agent builders show up and say, I need it all. I'm gonna give it a wallet, and it's gonna make x four or two transactions. And therefore, I'll need the on ramp, but I'm also gonna do cards and, you know, let my agents buy stuff on ecommerce. That said, the 75% of the agent builders who come to us to pick a round number are here for cards and ecommerce.

And they recognize that, look, if the vast majority of things that the that people are buying on a daily basis, the vast majority of an individual's spend the average median individual's, whatever you wanna call it, spend is going on a card through some website. And if you wanna capture some of that, like, you need a way for the human, the person, to show up with a card they already have and give it access in a secure way to the agent.

So between what we provide in our vault and the checkout API that we provide, we let your agent start buying for the human operator behind behind it. And that's where the vast majority of what we're doing the demand is. And so over the next few weeks here, you're gonna see a bunch more agents launch with that stack, you know, white label vault and using our commerce APIs to to purchase. Yeah.

Alex would love to feature and, again, common theme. Keeps on bringing bring bring brought the so what so what are people doing with this tech? Any any more consumer driven apps here that come to mind that you think would be great a good fit for for this segment? We'd love to chat with them. No. No need to pay many now. Feel free to think about it.

But if you do have any, feel free to share. I have some names. Maybe we'll talk after, and we can I can suggest some folks to to come on? No. I I it's a great question, so what? And I actually would take it a step back from just payments alone, like, the AI so what? Because, I don't know. You guys probably also see this on Twitter every now and then. There are those people who say, we have AI. I'm more productive. I'm doing more than ever. Why do I feel like my to do list is longer?

Why do I feel like there's more to do and that I'm not on top of shit? And I think that's you know, for all of us in this world, like, that's just the way we're gonna feel for a long time while, you know, we're bathed in the chaos of something new every single day. And maybe, like, our internal barometers for that will adjust. Yeah. Ultimately, like, the belief that motivates me here is that, like, if you can help people in businesses put stuff like this on autopilot, you can give them back time for something else.

And that's what I think this boils down to. Like, I wanna be able to put a bunch of stuff on autopilot personally. And then, yeah, as a former business owner, there were a million things, invoices and contractor payments and all this stuff that I had to approve. It's like an agent can do that. 90% of it, 95% of it, and maybe eventually a 100% of it to including to, like, sending the payment. So that's the thing that motivates me. I just wanna unlock more time for for people.

It's awesome. Thanks for the energy today. Just do it again.

We'll do it again. I was gonna say it was like, that felt like a that felt like a subtweet or, I was like, I feel I feel that pain. But Yeah. Yeah. Great to have you on today, I appreciate appreciate your time. We're sending you off with some, CrossMit CrossMit fireworks. I love it. CrossMit fireworks.

Didn't I even expect fireworks, but they look awesome. There we go. Later, guys. Cheers.

Thanks, Alex. Stabilash live continues Thursday, the September seventeenth.