Six months in agentic payments

Last summer, I wrote a memo that changed my life.

It was simply titled “Agentic payments memo.” It was a document about the problems, opportunities, and unanswered questions in a category that, at the time, barely existed.

That memo led to a (publicly announced) $10m seed round, a team of 12 that will soon be 25, and the foundation for what may very well be my life’s work.

The memo was informed by countless conversations with frontier founders. What was striking about those conversations was how early the market still was. When you suggested that agents might eventually execute payments, the most common response was a blank stare.

Of all the venture-backed founders I spoke with who were building frontier agents, only one had ever seriously considered that payments might become agentic. It simply was not part of the prevailing mental model.

Today, just 6 months later, the landscape looks radically different. My timeline is full of people debating the future of agentic payments. Every payments company is articulating some version of a strategy, and there is no shortage of takes about whether the volume is real, premature, or overhyped.

I’ve learned a lot over the last six months. Being early and in the arena, here are a few of my personal observations to help cut through the noise.

1 / Volume

It is true that agentic payments volume today is still de minimis.

It will not remain small.

There is now significant investment flowing into this category, and, yes, much of that investment is ahead of the market. Agentic payments do not yet represent a meaningful portion of GDP, but we have already crossed one of the most important thresholds: from non-existence to top-of-mind. That matters.

Entire ecosystems do not form around nothing. Numerous smart people do not commit years of their lives to a category by accident.

It is also worth remembering that building in regulated, complex markets (especially financial services) takes time. Mercury, Chime, Revolut, Stripe, and others all took years to build the infrastructure, partnerships, and operational scaffolding required of them by banks, processors, and regulators.

What many people are not seeing yet is the amount of infrastructure being built beneath the surface.

At Natural, we’ll be live with two direct bank integrations by the end of Q2 and three banks by next year, alongside our own ledgering, program management, risk and compliance, and multi-bank settlement infrastructure.

That work is not sexy. But it’s what turns a thesis into a market.

2 / Stablecoins

I’ll say it again: the idea that stablecoins are the only viable rail for agentic payments is wrong.

For many domestic use cases, fiat rails are cheaper, faster, and more controllable. Stablecoins become more compelling in certain cross-border contexts, but they are not the universal answer.

There are also many use cases where cards will continue to matter.

We have a merchant-acquiring product going live shortly that still relies heavily on card payments, while integrating that money movement into Natural’s broader ecosystem.

The future is not one rail. It is a routing and orchestration scheme across multiple rails and a settlement layer that abstracts that complexity without requiring an opinionated choice by the consumer.

3 / Point solutions

I’ve heard many versions of the same idea: build just identity, just a policy engine, just wallets, just one slice of the stack. I am extremely skeptical that this can work.

Founders do not want to stitch together a dozen disparate point solutions, especially in a category that is already operationally complex. More importantly, there are limited compounding advantages in owning only one thin layer of the stack.

The strategy with the most value accrual is much broader: identity, authorization, payment execution, settlement, risk, disputes, and everything around them.

That is obviously the harder path. Internally, I sometimes describe it as trying to thread ten needles at the same time. It doesn’t matter if we get nine needles right because they are all interdependent. We must thread all correctly in an exceptionally fast moving market.

But hard is not the same as wrong.

4 / Regulatory and compliance

The market is still underweighting this.

To developers, that may feel like a feature today. In 18 to 24 months, it will look like a bug.

It does not matter if you can generate payment volume today if a regulator is going to force your operations to shut down a year from now. This is still money movement. It needs to be treated with the same seriousness as traditional financial services, and, in many cases, with even more scrutiny.

This category’s winners will not be the teams that corner-cut regulations. They will be the teams that absorb the regulatory density of this space and build products that obscure that complexity.

5 / Use cases

Most of the public use cases you see today are toys.

That is not an insult. It’s how most important technology waves begin: smart people hacking on things they find compelling before the market fully forms.

But the companies that will move meaningful payment volume in incumbent industries are, for the most part, still quiet online.

Over the next three to six months, I expect a real shift: away from demos, experiments, and indie projects, and toward products that serve real users at meaningful scale.

That is when the signal gets easier to separate from the noise.

6 / MCP vs. CLI vs. API

The market changes its mind on the “right” interface roughly every eight weeks.

Anyone speaking with too much certainty here is mostly projecting their own preferences.

The agent infrastructure layer is still evolving extremely quickly. The tooling for getting agents to reliably execute long-running workflows over long context windows, without drift, and without human intervention is still immature.

Agentic payments companies will need to support every relevant activation surface until it becomes clear where durable developer mindshare is settling.

Betting too early on one interface is less conviction than it is fragility.

7 / Know your agent

Most opinions in this category are still preliminary.

Here is the first useful simplifying frame: in regulated environments that require real, verifiable identity, the easiest way to reason about an agent is to tie it back to an entity you can model through KYC or KYB.

That level of friction will not be required for the entire market, but it absolutely will be required for some meaningful subset of it. Who ends up with the most dense and valuable database of unified identity tokens? That remains an open question.

But the harder problem is this:

If you model a persistent identity around an agent workflow, at what point does the workflow change enough that the original identity no longer applies?

Human identity is anchored to a physical being. Agent identity is not. I can establish an identity for an agent, then rewrite the code, logic, and decisioning underneath it. At what point is it no longer the same agent?

How much underlying change is required before identity itself has to be recharacterized?

That is a real problem. And it is still mostly unsolved.

8 / Solution complexity

One of the most common failure modes I see is unnecessary complexity.

If I, a second-time founder in payments, struggle to follow your docs or feel like I need to become even more of a payments expert just to implement your product, you have already lost the plot.

The job of agentic payments is not to expose complexity. It is to absorb and abstract it.

An agent, or a developer who knows nothing about payments, should be able to implement a reliable solution without needing to learn the difference between custodial and non-custodial wallets, what step-ups are required for KYC or KYB, or how many days they have to issue provisional credit on a refund.

The best products in this category will make difficult things feel obvious.

9 / Commerce vs. non-commerce

Much of the market initially approached agentic payments through a commerce lens.

The canonical example was: if I say, “I want to buy a sweater” in ChatGPT, how does that transaction actually happen?

Our view has long been that commerce may be one of the last major markets to fully evolve here, not the first. With the rollback of Instant Checkout in ChatGPT by OpenAI, I think that’s even more true.

There are other categories where the need is more immediate, the workflows are more structured, and the value is easier to capture before fully autonomous consumer commerce becomes mainstream.

The market started with the most visible use case, not necessarily the most durable one.

10 / Protocols vs. products

The last major divergence in this market is the split between protocols and products.

Protocols aim to create shared, interoperable standards. In theory, they feel collaborative and neutral. But what people often miss is that products are where margin exists to support risk, fraud, compliance, operations, and customer trust.

Products are much harder to build. They are also where the real enterprise value accrues.

Consumers do not care about protocols. They care about the brand moving their money, whether it feels safe, whether it is fast, whether it is cheap, and whether the interface works.

Most protocols I have seen in agentic payments so far are not solving anything particularly novel. They are often overextended marketing positions solving narrow problems. Unless they expand meaningfully, I expect many of them to struggle to maintain long-term relevance.

There is simply not enough differentiation there today to sustain durable adoption.

In closing

Natural is still early in our journey, and I’m admittedly biased, but I believe we’re building the most complete agentic payments platform in the market, at the deepest layer of the stack. We have an exceptional cap table, we’re assembling an exceptional team, and we’re executing with real speed. If you want to build with us, we’re hiring across design, engineering, operations, and go-to-market. And if you’re building agents, send me an email. I’d love to get you set up.