Agentic Payments are a Side Show
The main event is elsewhere
Welcome to Stablecoin Blueprint, the weekly analysis about the opportunities and growth strategies in Stablecoin Payments and Onchain Finance
Got feedback or suggestions? Reply to this email or find my contact details online.
Everyone is debating the wrong question about agents and money
The discourse has settled into a familiar pattern. One side argues that agents will need wallets, that card networks weren’t built for machines, that stablecoins are the natural substrate for autonomous transactions.
The other side pushes back: most agents won’t transact autonomously, consumer agents will orchestrate but humans will authorize, most of the agentic economy will be billed monthly.
But both sides are arguing about plumbing. Which pipe carries the money is a second-order question. The first-order question is: Where do agents create new economic activity, rather than just replace it?
The AI-physical interface is where the growth is
Andrej Karpathy recently laid out a framework that I think captures what’s actually happening better than the payments debate does.
See Andrej’s appearance on No Priors podcast
He describes three zones: the digital world, the physical world, and the interface between them.
The digital world is already agent-native. Agents can write code, run ad campaigns, generate content, and manage an increasing number of SaaS operations. This is the water they swim in today.
There’s still a huge amount of innovation to do here (just look at the rapid proliferation of agent skills and MCPs). We still haven’t had enough thinking cycles for all the information that’s already digital, but the infrastructure largely exists. Agents can operate in this space with minimal friction.
The physical world (manufacturing, robotics, chemicals) is still years away from meaningful agent autonomy. Atoms are, as Karpathy puts it, a million times harder than bits. The capital requirements are enormous, the feedback loops are slow, and the tolerance for error is low. You can’t autoresearch an in-person physical inspection.
The interface between digital and physical is where the biggest growth opportunity sits. And it’s largely unbuilt. This is where agents need to reach into the real economy. Not to pay other agents, but to acquire data that doesn’t exist online yet, source physical goods, verify real-world states, hire contractors, and negotiate with suppliers.
My favorite concept from the Karpathy talk was when he describes information markets (seriously, go watch it). Imagine agents participating in prediction markets creating priced bounties for verified real-time photos from a developing situation on the ground. The internet enables global coordination, and agents are the coordinators, operating 24/7 to decide, act, and allocate resources accordingly. The digital intelligence is ready. The connective tissue to the physical world is not.
People are already building at the interface and hitting the wall
Ben Cera (@Bencera) is a solo founder building @polsia, a platform where AI agents run businesses on behalf of users. His experience is the clearest illustration of where the interface breaks down.
If your business is purely digital (build software, run Facebook ads, do email outreach, handle customer support) you can be almost 100% autonomous today. Polsia’s agents do all of this. They research markets, build MVPs, manage ad spend, even handle refunds with empathy. The digital layer works.
The moment you need something physical, you hit friction. Want to sell high-end pajamas made in Italy? The agent needs to email factories, negotiate quotes, inspect materials, verify quality control, and coordinate with a fulfilment center for shipping. Several steps at the interface between digital and physical are bottlenecks that the current infrastructure wasn’t designed to support. Phone calls are increasingly being automated, in-person evaluation (inspecting materials, assessing a factory floor, verifying quality) is far more challenging.
This friction creates a selection pressure that I think is underappreciated. If you’re a solo founder (or a small team running mostly on agents) which vendor do you favor? The supplier who exposes an API, accepts programmatic payments, and can be integrated into your automated workflow? Or the one who requires a phone call, a manual invoice, and a human in the loop?
Agentic-driven teams will route their spend toward suppliers who fit their workflows. Businesses that make themselves automatable will capture this demand. Those that don’t will lose it. This is how the interface shapes commerce; not through new payment protocols, but through the selection pressure of who can be bought from programmatically.
And this isn’t just existing demand being rerouted. As the barrier to starting a business falls, there are founders who wouldn’t have existed a year ago, building agentic-driven businesses that generate entirely new demand for goods, services, and labor. The interface creates commerce that didn’t exist before.
The recipient determines the rail
Payors favor the path of least resistance.
Merchants just want to get paid.
For maximal acceptance at existing merchants, cards will be the default. The interchange cost is real, but PSPs handle fraud, reconciliation, and multi-method acceptance. That’s complexity merchants would rather pay to avoid. This covers the vast majority of established commerce today.
When the recipient isn’t inside that network (the photographer in Tehran, the data labeler in Lagos, the freelance researcher who isn’t a Stripe merchant) stablecoins become the natural option. Not because of a theoretical argument about settlement times or interchange fees, but because permissionlessness is the only practical path to paying someone who can’t or won’t onboard to traditional payment infrastructure.
And yes, there will be agent-to-agent commerce. Agents are already discovering and using APIs without authentication, and there will be machine-to-machine payments where stablecoins or other programmatic rails make sense. That could be very interesting. But it’s small today (albeit growing). The vastly larger opportunity is between agentic-driven workflows and the real economy, agents buying on behalf of humans from businesses and individuals who already exist.
The opportunity: agentic-driven business
The agent economy is small but growing fast. The real economy is already massive. The interface between them (where agents reach into the real world on behalf of humans) is where the biggest opportunities exist today.
The discourse has been fixated on whether agents need wallets or cards, whether stablecoins will displace Visa, whether x402 or MPP will become the standard. These are interesting questions that I myself love to debate. But they are far less relevant than harnessing the net new economic activity that using agents creates.
The important question is: what happens when AI capabilities meet physical-world needs at scale? How much new commerce gets created when the cost of finding, evaluating, and purchasing from the real economy drops to near zero?
The builders who win will be the ones who make themselves (their goods, their services, their data, their expertise) accessible to the agentic-driven economy. The opportunity is at the interface. The rails will follow.
Thanks for reading!
Found this helpful? Let me know by clicking the like (🤍) icon and share it with someone who might find it helpful too.