The Headless
Merchant Era
Has Started

Noah Levine's guest post for a16z crypto introduces a precise term for what's emerging: the headless merchant. A business with no storefront, no accounts, no marketing funnel — just a set of API endpoints, machine-readable documentation, and a price per call. When an agent needs a capability, it checks the docs, checks the price, and executes the transaction in a single HTTP request.

The numbers from the article make clear this isn't theoretical. In one marketplace's first week live, approximately 900 agents executed over 31,000 transactions. That's not a pilot — that's a production signal.

The storefront was always for humans

Every assumption baked into modern e-commerce infrastructure — the homepage, the account creation flow, the checkout funnel, the subscription tier — was designed around one constraint: a human being has to navigate it.

That constraint produced a particular shape of business. You invested in UX because the human needed guidance. You built a subscription model because the friction of per-transaction billing was too high for a human to tolerate at high frequency. You ran a marketing team because discovery happened through human attention — search rankings, social feeds, word of mouth.

Remove the human from the buying side, and none of that structure is necessary. An AI agent evaluating a service doesn't browse — it reads documentation. It doesn't need a checkout flow — it executes a payment. It doesn't need an account — the payment itself proves it's authorised to use the service.

Levine's term for what remains when you strip out everything designed for human buyers is apt: headless. The merchant still exists, but without the face it used to need.

Payment as authentication changes everything downstream

The most structurally significant idea in the article is that payment and authentication collapse into a single act. In the traditional model, these are separate steps: you create an account (authentication), then you pay (transaction). The account is the relationship; the payment is a recurring event within it.

In the headless model, there is no account. The payment is the relationship — or rather, each payment is a complete and self-contained relationship. The agent pays, it receives the capability, the interaction ends. No ongoing identity, no subscription management, no churn.

This creates a radically different unit of commerce. Instead of acquiring a customer and managing their lifecycle, a headless merchant is just maintaining an endpoint that agents find reliable enough to call again. Retention is a function of uptime and price consistency, not relationship management.

Old model vs headless model

The infrastructure problem that article doesn't fully address

Levine describes the Machine Payments Protocol — a collaboration between Stripe and Tempo — as the technology making this possible, alongside tools from Visa and others. The framing is that the payment rails are essentially solved, and the opportunity is in building headless services that run on them.

That's partly right, and it's a useful corrective to infrastructure-first thinking. The services themselves are the product. But the payment layer for headless commerce has a specific shape that existing rails weren't built for.

When 900 agents are executing 31,000 transactions in a single week — at fractions of a cent — through a single marketplace, a few things have to be true about the underlying payment infrastructure:

• The agent needs financial identity. Not a human's card on file, not a service account borrowing credentials — actual financial identity tied to the agent itself, so that policy enforcement and audit trails are meaningful.
• Spending policies need to be programmable. At $0.003 per call, volume is the variable. An agent executing 10,000 calls in an hour needs automated guardrails — counterparty allow-lists, spend caps, category restrictions — that don't require a human in the approval loop.
• Settlement needs to be agent-native. When the paying entity is an agent and the receiving entity is a headless API, routing that through human-designed payment flows adds latency and compliance friction that doesn't belong there.

The headless merchant model works at scale only if the financial infrastructure underneath it is as headless as the commerce it supports.

The intention economy is a payment design problem

Levine uses the phrase "intention economy" to describe what replaces the attention economy: commerce moves from places — stores, apps, websites — to moments, the exact second an agent identifies a need and executes against it.

In the attention economy, payment infrastructure was a back-office concern. By the time a transaction happened, the merchant had already done the hard work of capturing attention and converting it. The payment was the final step of a long funnel.

In the intention economy, the payment is the entire funnel. There is no browse phase, no consideration phase, no cart abandonment. The agent evaluates options, selects based on price and reliability, and transacts — in a single exchange. The payment infrastructure is the product experience.

That's a different design constraint than payment infrastructure has ever been asked to meet. It's not about fraud prevention or checkout conversion rates — it's about latency, programmability, identity, and the ability to operate at a scale no human-driven commerce has ever approached.

Further reading: Autopilots Don't Just Process. They Pay. · Building on x402 — The Open Payment Protocol for AI Agents · "Trapped in a Box": Why AI Agents Need Their Own Money