The Agent Commerce Gap

AI agents are learning to make purchases. The harder challenge is connecting those decisions to the payment systems the physical world already uses.

By Sana Mirza

An AI agent can already do something rather interesting with money: pay for a digital service without requiring a person to complete every step of the transaction.

Imagine an agent working on a research assignment. It needs access to a dataset, so it requests information from an API, receives a price and, if authorized, pays for access. The service delivers the data, and the agent continues its work.

The product is software. The payment is digital. Both sides can communicate through machine-readable instructions. There may be no shopping cart, no cashier and no reason for a person to interrupt the workflow simply to approve a small, previously authorized purchase.

This is where agent commerce makes immediate sense.

Machine-payment protocols such as x402 are being developed for these kinds of interactions, while broader agent-payment frameworks are addressing questions of identity, authorization and delegated spending.

The infrastructure is still evolving, but the basic commercial interaction is relatively easy to imagine. Software requests something from software, pays through a supported mechanism and receives a digital result.

Now consider a different request.

A traveler asks an AI assistant to arrange transportation from the airport, reserve accommodation and handle an evening meal within a specified budget.

The agent might find suitable options and understand the traveler’s preferences. It may even have permission to spend. But completing those purchases means interacting with businesses operating across different payment environments.

The hotel may accept cards. The restaurant may display a local QR code. The transportation provider may depend on a domestic payment service.

The agent’s ability to decide what to buy is only part of the transaction.

The rest belongs to payment infrastructure that was largely built for people, merchants and financial institutions — not autonomous software.

Where Digital Commerce Meets the Physical World

The difference between buying an API response and paying a neighborhood restaurant is easy to overlook.

In a software-to-software transaction, the service provider can expose a payment interface designed specifically for automated interaction. Payment conditions, authorization and delivery can potentially be coordinated within the same digital workflow.

A physical merchant usually has no reason to build such an interface.

A restaurant in Bangkok may already have a functioning QR payment arrangement. A retailer in Jakarta may use a local acquiring provider. A business in Manila may depend on domestic banking and payment infrastructure.

These merchants have systems that work for their customers. They are unlikely to replace those systems simply because an AI agent wants to make a purchase.

The APC whitepaper identifies five layers of infrastructure that matter when agent payments move beyond software and into physical commerce. Each presents a different problem.

1. QR Networks

QR payments are familiar across many markets, particularly in Asia, but a QR code is only the visible entry point to a payment arrangement. The information encoded in it, the supported payment scheme and the underlying network can differ between countries. An agent may be capable of reading a merchant’s QR code, yet still need compatible infrastructure to interpret the payment request and complete the transaction. Understanding the code is not the same as being able to pay through the network behind it.

2. Local Acquiring Systems

Merchants typically receive payments through established acquiring relationships, payment service providers or other authorized financial participants. Those arrangements determine how transactions are accepted, processed and credited. An agent-payment system therefore needs a way to interact with the merchant’s existing acceptance environment. Giving an AI agent a wallet does not automatically establish a payment relationship with the business it wants to pay.

3. Banking Rails

Even when a transaction begins with digital assets, the receiving business may operate through conventional banking infrastructure. Domestic payment networks, bank accounts and local financial rules remain relevant to the transaction’s completion. An agent could initiate a payment at any hour, but that does not mean every subsequent movement of funds follows the same timetable or operates under the same conditions. The payment must eventually reach a financial environment the recipient can use.

4. Liquidity and Conversion

An agent might hold or be authorized to spend a stablecoin while the merchant expects payment in local currency. Someone — or some connected infrastructure — must handle the conversion, subject to available liquidity, pricing and applicable requirements. Exchange rates and transaction costs also need to be established clearly enough for the agent’s spending authorization to remain meaningful. A purchase approved within a budget should not become an unpredictable obligation because the funding asset and receiving currency differ.

5. Local Fiat Settlement

The final challenge is making sure the merchant receives usable funds through its supported settlement arrangement. A blockchain transfer may be confirmed before the merchant’s local-currency settlement is complete. Those are different events, and a functioning agent-payment system needs to account for both. For the business accepting the payment, successful settlement matters more than the technical novelty of the instruction that started it.

Together, these layers explain why the next stage of agent commerce cannot be solved entirely within an AI model or a digital wallet.

An agent may have the intelligence to select a purchase and the authority to initiate payment. Neither guarantees that the merchant can accept the resulting transaction.

The Merchant Shouldn’t Have to Change

There is a sentence in APC’s whitepaper that deserves more attention than most discussions of autonomous payments give it:

Merchants should not need to understand AI protocols or manage digital assets.

It is a simple requirement with significant implications.

Much of the conversation around agent commerce focuses on what the agent needs: an identity, a wallet, a spending limit, a way to authenticate and permission to execute transactions.

Those requirements are important. An agent acting on someone’s behalf must operate within verifiable boundaries, with clear rules governing what it can purchase and how much it can spend.

But the merchant has a different problem.

A small business owner doesn’t necessarily care whether a customer discovered the store through a search engine, a recommendation app or an AI assistant. The business needs to know that the payment is valid and that the expected funds will arrive.

Requiring merchants to learn AI payment protocols, maintain digital-asset wallets or introduce an entirely new checkout process would make adoption considerably harder.

Consider a restaurant that already accepts local QR payments. Its employees understand the process, its accounting procedures are built around it and its customers know how to pay.

If an AI agent wants to purchase a meal on behalf of a customer, the more practical approach is to connect the agent’s authorized payment instruction to that existing acceptance environment.

The restaurant should not need to become an AI infrastructure provider to participate in agent commerce.

This shifts the engineering problem away from merchant adoption and toward interoperability.

The infrastructure connecting agents to merchants must handle the translation between delegated spending instructions and established payment systems. It must also preserve authorization controls, transaction verification and the requirements of the receiving network.

There is a useful lesson here from earlier payment technologies. Widespread acceptance often depends less on how innovative a payment method appears than on how little disruption it creates for the business receiving the money.

Agent commerce may follow a similar pattern.

Three Environments That Developed Separately

The infrastructure problem becomes clearer when we consider the three environments now beginning to converge.

AI agents are developing the ability to interpret objectives, select services and initiate actions on behalf of users. Their payment requirements include delegated authority, identity, spending restrictions and records that make automated decisions accountable.

Programmable digital assets provide ways to represent and transfer value through blockchain-based systems. Stablecoins and machine-payment protocols create possibilities for automated financial interactions, particularly where software can communicate directly with software.

Real-time local payment networks have evolved around the needs of consumers, merchants, banks and domestic commerce. Their strength lies in established acceptance and settlement arrangements, often operating within particular markets and currencies.

Each environment has developed its own infrastructure, standards and assumptions.

An agent can be authorized to spend without being connected to a merchant’s payment network. A stablecoin can move across a blockchain without solving local-currency settlement. A merchant can accept instant payments without having any knowledge of AI agents.

Connecting the three requires more than placing a payment button inside an AI application.

APC is developing this connection through its broader payment infrastructure, linking programmable transaction initiation and digital-asset settlement with supported real-world payment environments. APC Agent Pay remains in development.

The significance of the approach is broader than any individual product. It recognizes that autonomous commerce will need to work with financial systems that already exist, including systems that were never designed with AI agents in mind.

Two Transitions, One Infrastructure Problem

For years, digital-asset payments have faced a practical question: how does value held in a blockchain-based environment become useful in everyday commerce?

Agent commerce introduces another question: how does software acting under delegated authority participate in that same commercial world?

The APC whitepaper describes these as two transitions:

CRYPTO → REAL-WORLD PAYMENTS

AI → REAL-WORLD COMMERCE

They are related, but they are not identical.

The first concerns the movement of digital value into payment environments that businesses and consumers already use. The second concerns the ability of authorized software to initiate and complete transactions within those environments.

Their intersection is where the difficult infrastructure work begins.

A capable AI agent does not automatically have access to the financial systems it needs. A programmable asset does not automatically become an acceptable form of payment at a physical merchant. And a functioning local payment network does not automatically recognize instructions from autonomous software.

For agent commerce to move beyond digital services, those separate capabilities must become interoperable while preserving the controls that make payments trustworthy.

The future may involve more transactions initiated by software, but the businesses receiving those payments will still have invoices, suppliers, employees, local currencies and financial obligations.

An agent’s intelligence may determine what should be purchased. Its authorization rules determine whether it is allowed to proceed. The payment infrastructure must then make the transaction possible in the world where the merchant actually operates.

That is the agent commerce gap — and closing it is a considerably larger task than teaching software how to spend money.

About AIPaywithCrypto

APC enables autonomous AI payments, programmable settlement, stable-coin infrastructure, and real-world merchant connectivity across global payment networks.