The Coming Market for Machine Customers

Most discussion of AI shopping still assumes a familiar consumer with an unfamiliar assistant. The person states a need, the agent researches the market, and the person approves the transaction. That is probably the right model for early consumer adoption. It is not the only market being built.

In June 2026 Mastercard launched Agent Pay for Machines, designed for software purchasing services from other software at high speed and, in some cases, for fractions of a cent. Three months earlier Stripe and Tempo introduced the Machine Payments Protocol, an open standard for agents to pay businesses and one another. AWS is already working with Stripe on HTTP 402-based infrastructure that lets software agents pay to access protected content and data.

These projects point toward a different commercial actor: the machine customer.

The buyer and the customer can separate

The economic principal remains human. A company allocates a budget; a consumer sets preferences; an organization defines policy. But the operational buyer can be software that searches, evaluates, consumes and pays without a person participating in every exchange.

That distinction is already familiar in advertising and financial markets, where algorithms transact on behalf of principals. What is new is the expansion into general-purpose services.

A coding agent may purchase a short-lived API call, a data record, a document conversion or a few seconds of compute as part of a larger task. A procurement system may compare approved suppliers and place replenishment orders within policy. These transactions can be too small, too frequent or too context-specific to justify human approval individually.

Machine demand changes product design

Interfaces exist because people need help understanding choices. Software buyers want something else: structured capabilities, explicit prices, predictable contracts and machine-verifiable outcomes.

This can change the shape of the product. A service designed for machine customers may be sold in very small units rather than monthly subscriptions. Authentication may be delegated rather than account-based. Pricing and license terms may need to be returned programmatically. Quality may be evaluated continuously rather than through brand familiarity.

Mastercard describes machine transactions as “programmatic, always-on and executed between systems.” Stripe makes a similar point: the current financial system often forces agents to create accounts, navigate pricing pages and enter credentials—steps designed for people, not software.

A second audience for every digital business

The implications extend beyond payments. Businesses may increasingly have to communicate with two audiences.

The human customer responds to brand, trust, narrative, design and experience. The machine customer responds to structured evidence, interoperability, price, permissions and reliability. The software ultimately serves a person or organization, so the two are not independent. But the path from demand to transaction differs.

B2B procurement and digital services are likely to move first because policies, budgets and repeat purchases are easier to formalize. Consumer commerce will follow selectively where preferences are stable and the cost of error is low.

The strategic challenge is differentiation. If agents compare every offer on explicit attributes, merchants that cannot express value in machine-readable terms risk becoming invisible; merchants that express only attributes risk becoming commodities.

The coming market for machine customers will therefore reward a combination that ecommerce has not previously had to optimize: computational legibility without surrendering the reasons a human principal prefers one supplier to another.

The economic buyer may remain human. The operational customer can increasingly be software.

“These transactions are programmatic, always-on and executed between systems in the background of digital commerce.” — Mastercard