The Plumbing Under the Agent Economy

The Original

This week Stripe announced the Machine Payments Protocol (MPP) — an open standard, co-authored with Tempo, that lets AI agents pay for things programmatically. No login flows. No pricing pages. No human hovering over a checkout screen. An agent requests a resource, gets a payment request back, authorizes it, and moves on.

The launch partners tell the story Stripe wants you to hear: agents spinning up headless browsers, printing physical mail, even ordering sandwiches from a deli in New York. Payments settle through the same Stripe infrastructure businesses already run on — same dashboard, same payouts, same fraud protection.

In a few lines of code, your agent can spend money.

Our Take

Let's be honest about what just happened. Stripe didn't invent agent payments. They claimed the namespace.

And that matters more than the technology. MPP is a PaymentIntents wrapper with a protocol spec stapled on. The actual financial plumbing — stablecoins via Tempo, card rails via SPTs, settlement into existing Stripe balances — that's all infrastructure Stripe already had. What's new is the framing: agents aren't users of human payment systems. They're a new category that deserves its own standard.

This is the Stripe playbook executed perfectly. They did the same thing with Connect (marketplaces), Billing (SaaS), and Issuing (cards). Find a pattern that developers are hacking together, name it, wrap it in a clean API, and let the ecosystem crystallize around you.

The question isn't whether MPP is useful. It obviously is. If you're building an agent that needs to buy things, you'd be foolish not to look at it. The question is whether the model of agency it assumes is the one that actually emerges.

Where This Goes

MPP assumes a world where agents are individual economic actors — each one authorized to spend, each transaction atomic and self-contained. Agent sees resource, agent pays for resource, agent gets resource. It's clean. It's HTTP-shaped. It maps beautifully onto how Stripe already thinks about the world.

But watch what's happening in practice. The agents people are actually building don't operate as lone economic actors. They operate as extensions of humans and organizations. They have budgets, policies, approval chains. The interesting problem isn't "how does an agent swipe a credit card." It's "how does an organization grant spending authority to a fleet of agents with different trust levels, different scopes, and different risk profiles."

That's not a payment protocol problem. That's an authorization and governance problem. And MPP, as specified, mostly punts on it.

There's a deeper tension, too. MPP and Stripe's broader Agentic Commerce Suite are designed for a world where agents buy things from businesses. But the most interesting economic patterns emerging in agent systems are agent-to-agent. One agent hiring another. Swarms negotiating resource allocation. Micro-auctions for compute time. The Prospect Butcher sandwich demo is cute, but the real action is in the plumbing between agents — and that's a coordination problem that payment rails alone don't solve.

Here's what I think Stripe gets exactly right: agents need financial identity. Not human identity bolted onto autonomous systems, but something native. The idea that an agent can be a first-class participant in a payment flow — not through screen-scraping and browser automation, but through a protocol designed for it — is genuinely important.

But financial identity without financial governance is a loaded gun. And the first company to solve agent-level authorization — budgets, scopes, revocation, audit trails — will own a layer that sits above any payment protocol, including MPP.

Stripe is betting that payment rails are the foundation. I think they're the second floor. The foundation is trust infrastructure — the system that decides which agents can spend, how much, on what, and who's accountable when they get it wrong.

MPP is a good start. But the hard problems are above it, not below it.