The Payment Network Nobody’s Talking About

The Payment Network Nobody’s Talking About

In 2023 I watched a $4,200 payment take six business days to clear between two domestic businesses. Both used major banks. Both used standard ACH. Nobody could explain why it took six days. It just did. That’s the payment infrastructure underlying most of American B2B commerce: slow, opaque, tolerant of multi-day float because everyone built their businesses around it.

Now I’m watching AI agents execute purchasing decisions in seconds. And then waiting six days for the payment.

That mismatch is going to break things. Or it’s going to force a payment infrastructure upgrade that the industry has been deferring for 30 years.

The Speed Gap in Agentic Commerce

When AI agents are doing procurement, the decision layer collapses from days to seconds. But the payment layer still moves at human speed — or slower, because legacy payment rails were designed for batch processing, not real-time execution.

This creates a bizarre situation where the most sophisticated part of the commerce stack (the AI decision layer) is constantly waiting on the least sophisticated part (the payment rail). It’s like designing a Formula 1 car and then putting in a governor that limits top speed to 40 mph because the roads haven’t been updated yet.

For consumer commerce, this matters less. Individual purchase amounts are smaller, humans tolerate a day or two of pending status, and cards do provide something close to real-time authorization even if settlement takes time. But for B2B? We’re talking about $50,000 orders that an agent can evaluate and commit to in 90 seconds, then wait a week for payment to actually move. The agent’s speed advantage evaporates entirely at the payment layer.

What the New Payment Infrastructure Needs to Be

The Universal Commerce Protocol is addressing this partly through payment method standardization — ensuring that whatever payment method is available to an agent is exposed as structured data in the transaction schema so the agent can select and execute without human intervention. That’s a meaningful step.

But the deeper problem is that most real-time payment infrastructure (RTP, FedNow) hasn’t been built with agent-to-agent commerce as a design target. The authorization models assume a human is authorizing each transaction. The fraud detection models assume human behavioral patterns. The dispute resolution processes assume a human can respond to an inquiry.

Agent payments need different primitives: programmable payment conditions, pre-authorized spend envelopes that can be drawn against in real-time, transaction metadata that travels with the payment and enables automated reconciliation, and dispute protocols that agents can navigate without escalating to a human for every exception.

The Business I’m Actually Building Around This

In my restoration operation, we’ve moved about 60% of our supplier payments to a system that at least approximates this: pre-authorized spend pools by vendor, auto-reconciliation against POs, same-day settlement for approved vendors. It’s not perfect — it’s still built on ACH with some add-ons — but it’s reduced payment-related friction in our procurement by something like 70%.

The vendors we pay through this system get paid faster. They also get more business from us, because our agent prefers vendors where the full cycle (order to payment to confirmation) is clean and fast. The vendor with the best product at the second-best payment terms loses to the vendor with a good product and frictionless payment. Consistently.

I think about what the payment rails of the fully-built agentic commerce economy look like. It’s essentially a real-time settlement network with programmable conditions, agent-readable authorization chains, and automated reconciliation that doesn’t require a human to touch anything until something breaks. That network isn’t built yet. But the economic pressure to build it is mounting every month.

The company that builds the payment layer for agent-to-agent commerce is going to be very large. I’d rather be their customer early than scramble to integrate when everyone else is already there.