# Dakota ships agent primitives for bill pay, contractor payouts and treasury policy

Stabledash Live, a Stabledash show. Aug 29, 2026. Ryan Bozarth.
https://live.stabledash.com/shows/the-live-show/episodes/dakotas-new-agent-primitives-bill-pay-payouts-and-spending-ZnL7NNlL12Q

CEO Ryan Bozarth says agents should be treated like new hires: given a role, a card and an expense policy, with guardrails that make written cash policies actually run.

Dakota is adding a set of agentic finance primitives on top of its money storage tools, giving agents policies and guardrails whether they are hosted or unhosted. One of the first uses, according to the company, is Slack agents that take an invoice, pay a bill or pay a contractor in USDC, with human escalation built in.

## Built from its own bills

Co-founder and CEO Ryan Bozarth says two things drove the build: tools Dakota made for its own team, and AI-native companies describing how they would build the next fintech. Internally, invoices arrive by email and go to an agent that has guardrails, checks for duplicates and makes the payment.

From neobanks, he says, treasury operations and payment operations were the two most requested pieces, with customers wanting agents waiting for them when they open an account.

## Money out and money at rest

Bozarth describes payment operations agents as handling expenses, invoices and contractor payouts, controlled through policy engines, audit trails and allow-listed recipients. Treasury operations agents instead look at stored money and ask whether it is set up well, for example holding some assets in a stablecoin chosen for liquidity and others in stablecoins that earn rewards.

## Policies that actually run

Bozarth argues that cash management policies are often written down but rarely followed. He says Dakota helps fintechs and neobanks codify those policies into their platforms so they can sell them to their own customers. He points to the SVB era, when many companies were told they needed a cash management policy, but adds that it was not always done and memories are short.

## An agent is a role

Bozarth says an agent should match a role or job in an organization, which usually contains several skills, rather than a single task. He compares it to hiring a new payment operations person who gets a credit card along with expense policies and a monthly spending limit, and says agents should face similar constraints.

Taken together, Dakota is presenting itself as the regulated layer under other companies' agents, with its own controls on spending and recipients. The interviewers also noted that the company has an active application for a National Trust Bank Charter.

## Key takeaways

1. Dakota now offers policies and guardrails for hosted or self-hosted agents, with an early use case of Slack agents paying invoices and contractors in USDC.
2. Ryan Bozarth splits the work into payment operations agents for money out and treasury operations agents for money at rest.
3. Bozarth says written treasury policies are rarely executed, and Dakota helps fintechs and neobanks turn them into enforceable policy engines.

## Chapters

- 0:00 Dakota moves from developer tools to agentic finance primitives
- 2:39 Internal needs and AI-native fintechs drove the build
- 4:56 Payment ops handles money out, treasury ops handles money at rest
- 7:40 Written treasury policies get converted into enforceable policy engines
- 9:02 Neobanks offer codified cash management to differentiate
- 11:23 Agents should map to roles bounded by spending constraints

## Transcript

**Drew Rogers** (0:00)

Ryan Bozars. Ryan, good morning. You again. Welcome. Good to see you guys. Appreciate having me on. But... Yeah. I love you, bud. Lots of lots of fun fun movement here with Dakota. I feel like each each week I go back to your website, something new is shipped, something new is developed. And I guess, right, you're you're a very... I think from what I've gathered, a very AI native, very AI forward team, which I I have to imagine this is helping with the the speed at which you can ship.

But bring us into this moment now on the side of AgenTic payments of what you've also been deploying for now your customers? You ran the AgenTic challenge. That was awesome to see to come to a close. But, like, what is this next iteration of what Dakota can... You know, the the tooling you're you're building for agents with Dakota? Yeah. 100%. So we started earlier this year with really building out a set of tools that made it easy for people to build on Dakota. Everything from agentic SDKs,

**Ryan Bozarth** (0:52)

to be able to run challenges with folks, to be able to kind of really push on what we are building and make sure that they could really get up and running as they're building in new ways today on top of our APIs. And then what we're doing now is kind of the next iteration of that is providing people a set of primitives so that they can build agents into their own products. And so what that looks like today is that we have core primitives for thing, money storage.

The next piece is a bunch of primitives around agentic finance. And so what that means for us is being able to have policies and guardrails for agents that are hosted or unhosted. You can bring your own that leverage Dakota's architecture and infrastructure to build things like pay bills.

And so, one of the first use cases that we saw was people building Slack agents that allow them to call Dakota within Slack and be able to give it an invoice to pay its bill, To be able to make a payment to a contractor with USDC, and really kinda managing a lot of the kinda payout process, where that's gonna be bill pay, and increasingly things like contractors and payroll, all within a kind of secure ecosystem.

Not just a one time payment, but also being able to have kind of human escalations, be able to schedule payments. And as we play this forward, we're kind of really building towards a world where your finance team and your product can both play the role of the kind of financial operations agent supplemented by your human team. And so, as we see the next generation of neo banks and fintechs products be built, one of their key areas of thinking about how to arrive at this offering for their customers.

Our goal is to be able to manage the kind of money movement, the regulated part of that infrastructure, so that they can be able to apply the logic layer, the agent layer on top to really manage products like bill pay or finance So

**Drew Rogers** (2:39)

walk us through, Was this in the road map, or was this a, hey. This is how people are using a product? Maybe we were thinking about it six, twelve, eighteen months down the line, but, actually, we need to accelerate this because, I mean, I think we... We've even done this in in house where we just start to build these things, and it's not necessarily how anybody intended their app to be used.

And then next thing you know, we see features rolling out the next month and next month and next month so they can serve this in house. What were those learnings like? And and walk us through, like, what did you start seeing first that said, okay. We act... Like, unpack that a little bit more. Yeah. So there's really two things that kinda, like, drove this particular, like, product build out. One was our own team, and so we kind of, like, have built a lot of things internally that, like, we end up using.

**Ryan Bozarth** (3:21)

And the second was really, like, AI native companies coming to us and saying, when I build the next kind of AI native, like, fintech, this is how I'm how I'm envisioning it. Here's how I want it to work. Here's what I want to kind of happen, and really kind of pulling on those threads to be able to, okay. Here's the right feature set for you. And that might look a little different than someone who is kinda not leaning into AI in the same way.

And so from our own experience, a lot of the bill pay pieces were things where we were looking at that and saying, hey, wouldn't it be nice if, you know, as these invoices come in via email, as they kinda come into Dakota, we can just be able to send it to an agent to take care of It could have the right guardrails in place. It can make the payment for us. It can check if it's a duplicate.

It can do all of the work that you might expect someone on a financial operations do, but just allow us to kinda, like, leverage the technology that exists today and the architecture we've already built. The kind of regulated aspect of the license aspect of the money movement all taken care of. And so, like, we looked at our own use cases, and then we also have people coming to us and saying, my version of Neobank actually looks like this.

And that was one in which agents were built into the Neobank itself so that when a customer arrived at that neobank, they were greeted not just with an interface, both agents that could perform work for them. Treasury operations and payment operations being the kind of two most common requested pieces that say, once one arrives at this neobank, I want there to be people with it in the form of agents. Can you maybe break those two areas down? We can kind of get into the difference between treasury ops, payment ops.

Like, how do you typically like to compare and contrast what that means and then the people at the companies who love love what's possible differently there? Yep. 100%. On payment operations, to maybe to to ground us, I'll think about this as like we have, like, money in, kinda money storage, and then money out. A lot of payment operations is gonna help facilitate the money out part of the equation.

So the expense management, That whether I have a bill like an invoice coming in, whether I need to pay a contractor, like, what are the ways in which there is an agent that could help me make those payments, and how do I set that up so that they have the right guardrails in place to do that? Whether that is going to be the policy engines, whether that's going to be the audit trail, whether that's gonna be kind of kind of recipients that are allow listed.

But all of the infrastructure you need to be able to say, hey, as this invoice comes in from a vendor via email, I can send it to a, oftentimes, a group email alias, kind of your, you know, finance app, your billing app, things like that, that then allow it to get automatically paid and also have the kind of ability to then manage that payment going forward, whether it's a schedule or something else. The treasury ops is a bit different.

The treasury ops is more about looking at the money that's at rest, so the money kind of at storage, and saying, what is the optimal setup? And is that optimally set up today, or do I make changes for the future? And so one optimal setup might be, I wanna keep some assets in a stablecoin that is optimized for liquidity. Other stablecoins that are being optimized for my ability to earn rewards on those stablecoins. As I think about cash, being able to move between cash and stablecoins.

And so you're looking at that, and the business is deciding that says, I would like this kind of storage or treasury setup that kind of indicates where I'm at with liquidity and operations, and where I'm at pre optimizing for things like yield or return, and be able to monitor that and be able to make those movements given the direction that says, just for simplicity, May is 8020, where I want 80% in long term and 20% optimized for liquidity or six months of operating expenses optimized for liquidity.

I can then build that in into a framework, and that really helps for a lot of cash management policies, as an example. A lot of companies at a certain scale will run off of a cash management policy that says, I need to hold my cash in two to three or more institutions to manage my risk. I need to be able to earning as well as having liquidity for certain events that might take place, for certain amount of payroll, for example.

And being able to codify that and hand over to an agent to monitor that is one of those things that very rarely happens. The policy might be written down, but is very rarely executed on. And so that treasury management piece is one of those areas where it's not only a kind of opportunity for yield, but it's opportunity to manage your risk as well. How do you go about setting up those policies in place? I'm thinking especially for some of these orgs who, you know, have been around five, ten years.

They're transitioning into this neobank,

**Drew Rogers** (7:40)

stablecoin, global finance world. Do they bring their PDF? Like I said, they wrote down these policies, and does AI scan it and they build a policy around that? Is it working with your team and working with the current CFO to build those out, or is this they're just going to the platform and and they have a a kind of a flow to set those up? How do they go to zero to one to start start building these Yeah.

**Ryan Bozarth** (8:01)

So the first thing is to write them down. The beauty of of kind of AI is that we can take kind of plain English language, and we help convert that into code or into policies. And so generally, we've seen that people start with what is often a written document that is the cash management policy or the treasury management policy that they might have, and say, this is how we think about it. This is how the executive team or the board has signed off on it. And then the next job is to codify that.

And so depending on the size and shape of the organization, we've had people do it on their own. We've definitely also leaned in to help them. They'll say, this is what this looks like when you convert that document into effectively what is a policy engine. And the policy engine can then be governed through a lot of the primitives that we have. So we offer the policy engine, but we also offer things like mandates or caps that say, you could only operate within this kind of refined constraint.

So for example, only send to these institutions. Only send up to this amount across this period. But once you have that policy engine, then a lot of the individual primitives come from Dakota or they can come from Dakota, in which case, we can say, here's how that is materialized.

**Zach Fowler** (9:02)

Here how that is that kind plays out as you go forward and as you manage your treasury, manage your operations. Brian, how do you see the... Like, I'm hearing this and I'm thinking about our business, maybe some other, like, you know, SMB mid market style use cases here where, like, maybe younger companies, my hunch is, like, we're gonna be more risk on, more willing to try new things, more willing to, like, use stuff like this compared to perhaps the larger enterprise or Fortune 100 or something?

And who probably do have all of those things living in a PDF somewhere? Like, how how are you dealing with that maybe? Like, it seems like the the use case is is ideal for those who have these types of systems and rules in place, but those who may... Might be ready to, like, test and get started with this may not actually have a lot of that. How is Dakota thinking about even going back? I think about, like, the Stripe Atlas use case of, like, hey. We're just...

This may not be, like, directly core to the business, but if we can help you incorporate the business, then we're right there when you're ready to take payments. Like, are you also thinking about ways to even help set up some of these policies or things like that? Yeah. So one thing that we've seen is that when we're kind of working with a a fintech or a neobank,

**Ryan Bozarth** (10:13)

part of their value proposition comes in the form of bringing the agents alongside the software layer, for example. Increasingly, they're finding other ways to differentiate, which can be things like this, which is like, let me also... As I bring that bank account for you, that banking platform, I can help you manage your risk. And for a lot of companies that kinda came out of the kind of SVB era, for example, we all actually probably saw a lot of those documents that were written that said, hey.

You really need a cash management policy. But to your point, like, wasn't always done, and oftentimes, we have a short of memory we'd like to admit. But now these platforms are starting to do the heavy lifting of saying, you know, I can help you. And so we oftentimes are helping the fintech, helping the neobank kind of codify that and build that into their platform, but then they can offer that as a value proposition to their end customers.

So even though that next, you know, let's say, startup c stage company might not be thinking about cash management, they might say, hey, if I can have a smaller finance team, if I can have a better kind of cash management, treasury management policy, I might use that as a way to differentiate this neobank from the next one, or this PSP or fintech from the next one. Really, really cool. I know this was Right. Was the last kind of Yeah. Last question. I'll let you close it.

We got a last one, and then we have Eugene coming on from OSL.

**Drew Rogers** (11:23)

The word agents right now, it feels like a bit of a catch all. And I think about the agents that we use internally, externally, externally, that we use as kind of, like, individual... You know, they're kind of like these siloed workers versus the ones that are supplementary.

**Ryan Bozarth** (11:38)

You know, I think about the the the CPA that has five agents, you know, and they're... It's really this augmented CPA. How... When you think of agents, how do you define them? And if you were to talk to somebody that was just entering the workforce and you were to say, hey. Here's an area where you could go zero to one with agents, what would your advice be to them?

I think one of the biggest things with agents that we've thought about and learned about is it matters a lot having a shared understanding of like the altitude that you're discussing, or the abstraction. I think one thing that we've come back to over and over again is that an agent oftentimes is aligned with a role that can be done, or a job to be done in an organization. And that job to be done oftentimes has multiple skill sets inside of it.

And so, for something like a payment operations agent, that is typically a role or a function at an organization that has several tasks or several skills inside of it. So that agent, for example, will have access to those skills, but is out of abstraction layer that says it's a collection of these skills that provides a job to be done for an organization. So very similar to a role. And so we think about the altitude. We think the roles are incredibly important kind of abstraction.

And then from there, you can start to say, hey. What are the roles or the jobs to be done? These oftentimes might be one or more workflows as an example. And so our ability to kind of assign skills, give it a kind of scope of work or a definition, so this is your lane. This is your role, but then also give it the ability to kind of make decisions within a set of constraints.

So for example, if you were to bring in a new payment operation person, you would have just given them an unfettered credit card, but you would say, hey. Here's Nick. Here's a here's a credit card. Here's some expense management policies that we have. You have up to x dollars on monthly basis to spend it down on. Very similarly, we think of those types of constraints, but also not being prescriptive about saying, hey. If you can, you know, find research, you can find work,

**Zach Fowler** (13:19)

then you have the ability to kinda spend down up to that amount. It's awesome. Excited to continue following this, Ryan, to keep us updated. I know you guys have some active application out for the National Trust Bank Charter. A lot of stuff to be covering here for you guys. Great to have you on. And and we're... We'll have two actually neo segments, neo powered by you guys later today in the show. So it's been fun working with you guys on that. 100%. Thank you, guys. Appreciate the time. Thanks for the time, guys.

See you. Bye.