Fragment raises from FirstMark to build programmable money infrastructure for AI agents

Speaker 3: back to papers. Back to letters. Anyway, we

Speaker 2: have I think we're way I think we're way over. Yeah. Way over. It's time to bring in our first From Fragment. Of the show. The co founder of the Dear friend.

Speaker 3: Welcome to the show.

Speaker 2: There he is.

Speaker 3: How are you doing?

Speaker 6: Good luck, guys.

Speaker 2: What's up, dude? It's only been a few years. It's great to see your face.

Speaker 6: You too. Looking good, Jordy.

Speaker 2: Looking good. So

Speaker 3: you two know each other, but I don't know if the audience does. So why don't you kick it off with an introduction?

Speaker 6: Me or Jordy?

Speaker 3: Yes. You, the guest. Tell the audience.

Speaker 6: Honestly, they're guest. We... Ran a company called Party Round back in the day, and I had a company called Cove. Cove was a... Kind of like a... An automated fund admin for private investments. And we took on the task of administrating Party Round's deals.

Speaker 2: Mhmm.

Speaker 11: Yes.

Speaker 3: Cool.

Speaker 6: Well, along the way, we are actually... So Party Round was one of them, but we we administrated a bunch of real estate deals and private equity deals. Mhmm. And that's... Doing that with two people was was a pain... A super big pain, and that is the catalyst for building Fragment.

Speaker 2: Alright. So catch us up to speed. You go from from Cove, it evolves into Fragment. You raise, if I remember correctly, a seed round from from Stripe. Was this in 2022 or or maybe 2021?

Speaker 6: We raised a a pre seed from from, from Box Group

Speaker 11: and then a

Speaker 6: seed from Stripe. Yep.

Speaker 2: There you go. And then, yeah, catch us up to speed on the last on the last couple years.

Speaker 6: We we built a ledger API, which is basically a system of record for fintech companies. And fintech companies built all of the products that run those companies on this system of record. So companies like WAP and Ramp and build.com run on that infrastructure. And over the course of the last six months or so, we saw that those those those folks were really pushing the system with different types of transactions and these adjective transactions. So they were using the system... Using agents to actually run a lot of the back office work for these fintech companies. And so we started following those that track very closely and have built now a financial system that's purpose built for agents and just announced a funding round with with First Mark.

Speaker 2: John, why don't you hit the gong as hard as you possibly can?

Speaker 3: Would love to.

Speaker 2: And I have some follow-up questions.

Speaker 1: I would love to. Bang.

Speaker 2: Oh. And it shook shook the room. Alright. So go... I I I would love if you could get more specific around how this actually works today and how you think it'll work in the future because we've been getting agentic... Various like agentic payments pitches for a while now, and none none of them have been that compelling, at least to me. What I'm not hearing you say is like we're gonna we're gonna replace Stripe because Stripe couldn't possibly be good at at at at moving money. It's more... It's like a different layer of the stack. Is that is that right?

Speaker 3: I I mean, I... The... All the sustaining innovation pitches that I've heard have been like a 100% I'm on board. Like Stripe saying, like, yeah, agents are gonna be moving money over Stripe rails. I'm, like, I'm in. Plaid, I... You know, I've I've I've used Plaid in Chatuchiki to authenticate a bunch of bank accounts. It's great.

Speaker 2: Or even with personal agents, the the the most compelling pitch on how they're gonna make money Generally, short a term is like, yeah, interchange fees. Yep. But but you're talking about working with these enterprise companies that are using agents to, like, basically run processes on the back end and then update a ledger, trigger payments, things like that.

Speaker 6: Yeah. So we wanna automate commerce.

Speaker 2: Mhmm.

Speaker 6: And we're gonna do that by outsourcing the the work of finance to agents. And agents can't really do that work by operating all of these sort of human operated financial systems. Why? Well, so to automate commerce, you kinda need to know what commerce is. Commerce is a set of of promises. It's it's like these instructions for who owes what, to whom, when Mhmm. And money money is the system that settles them. And so where do these where do these instructions exist? Well, in b two b, they're, like, contract terms. They're they're buried in PDFs and docs and whatnot. And what what do those terms include? Well, they're they're like invoicing terms, and SLAs, and payment terms, and refunding, and return rules, and billing, and accounting rules. And in existing... In in traditional commerce, those instructions for settling them are are distributed across all these different systems. And that makes it really hard if you're an agent to go across all these systems and and and settle. Each system sort of has a portion of the story, and each system plays its own part. And humans sort of carry the context across them. And it's that human bit where the humans are are are part of that story that, you know, once transactions exponentiate, creates economic strain. That's... That that creates a problem.

Speaker 3: Is that true? I'm I'm so confused about, like, agents can't go across different systems because it feels like the narrative over the last couple months has been, like, the agents are in too many systems. Like, it's, like, the opposite. It's, like, like they're reading PDFs they shouldn't read, and they're in the system that they weren't even supposed to be in. It feels like agents are, like, really good at, like, reading rules and then, like, interacting with different systems, like writing API code on the fly, using MCP, using computer use. Like, I I I understand that, like, there's a lot of problems with that, but it feels like it's, like, firmly in the path. But I still think there's, like, a lot of value here. But I'm wondering, like, like, how does the how does the company interface with, like, the model capabilities getting better? So

Speaker 6: the model capabilities are gonna get better at intelligence and reasoning. Mhmm. The modeling capabilities are a little bit less likely to get better at the deterministic stuff that are the rules of money and variants and the instructions that exist in these contracts.

Speaker 3: Okay.

Speaker 6: So it... Let's just like take take an example. If if you're an agent and you are trying to, like, disperse a claim, for example, an insurance claim, well, one system is gonna have the the balance for the claim. Another system will have, like, what's already been paid. Another system will have what you've accounted for. All these rules exist in different systems.

Speaker 3: Sure.

Speaker 6: And the... In order for the agent to know what to disperse or what to account for, it doesn't have a... It doesn't have all the context. It doesn't have a a clean view of the instructions across all of these different systems. Yeah. So you're basically telling an agent to go into each of these system and operate the flow that the that the the application prescribes. What we want to do is let the agent go nuts. Like, let them actually... Let them let them let them be intelligent. We're going to constrain them from doing anything stupid, but we want them to be intelligent so that they can refund what they need to. If... When you're solving a problem, you you need to refund what you need. You want account for the the way the the deal is supposed to be accounted for. You want to pay maybe a little bit now and a little later on. You might want to use credits. There's all these different ways that you're going to Sure. Sure. Easily solve the problem.

Speaker 3: Very cool. Yeah. That's...

Speaker 2: Put it more simply, like, there's a lot of stuff with money that you want to be deterministic. Is that is that another way to think about it? And like have...

Speaker 6: Yeah. You want the agent.

Speaker 2: You don't want the model like reasoning through... You want it to be reasoning through like maybe like certain parts of the process but not...

Speaker 3: And probably not every Yeah. Time. Like you don't necessarily want your... Like every time you you you get a transaction on your e commerce site, you don't want like, you know, GPT six pro to like figure out if it's the right amount of money. Yeah. Yeah. It's like, just use a calculator. Right?

Speaker 6: Well well, you want the agent to take the rules out of all these contracts.

Speaker 3: Yep. The

Speaker 6: the the rules in these contracts are what programs the money.

Speaker 3: Sure. Sure. Sure.

Speaker 6: So you take the rules from all these contracts Yeah. You put it next to the balances that those rules need to update

Speaker 3: Yep.

Speaker 6: And now you have now you have programmable money. And the agent will execute the... Those rules and update those balances accordingly. Yeah. That's that's that's the system we're building.

Speaker 3: Yeah. Yeah. That's very cool. Awesome. Well,

Speaker 2: thank you. You hiring anybody?

Speaker 4: Oh, yeah.

Speaker 2: Got a bunch

Speaker 3: of new...

Speaker 2: No. Got a bunch of money? Where are you guys building the team these days?

Speaker 6: We're in New York. We are hiring like crazy go to market folks, Yeah....

Speaker 2: Anyone.

Speaker 3: Very cool. Crushing it too. Yeah. New York is New York is crushing it. It's been it's been awesome to watch. Well, thank you so much for coming on the Great

Speaker 2: to get the update, dude.

Speaker 3: Have a great rest your day. Congrats. And we'll talk to you soon. Goodbye. Yeah. Alright. Let me tell you about Railway. Railway is the all in one intelligent cloud provider. Use your favorite agent to deploy web app servers, databases, and more while Railway automatically takes care of scaling, monitoring, and security. Our next guest, Danielle Jane, here with us in person in the studio.