Agentic commerce depends on machine readable price and permission

A software agent found an API that could complete its task, then stopped at the pricing page. The service required account creation and a monthly plan. A human could make sense of the options. The process could not determine whether one request was worth the subscription.

Stripe’s Machine Payments Protocol announcement identifies this mismatch directly. Existing payment systems often require account creation, plan selection, and checkout steps designed for people. Machine payments move price and proof into a protocol that software can interpret. We take the trend seriously because the cost appears after the first successful demonstration.

Stripe’s 2025 annual update reported that businesses on Stripe processed $1.9 trillion, equivalent to about 1.6 percent of global GDP. Its agentic commerce work reflects an expectation that software will participate in more of that economic activity. The direction of the evidence strengthens our view that quality and ownership must be measured together.

A machine readable offer needs more than a number. The client needs the unit being purchased, the conditions of delivery, and a limit on what it may spend. The seller needs a way to verify payment without receiving reusable credentials.

We take the position that machine commerce requires explicit prices, permissions and settlement records. Autonomous purchasing without those records creates accounting and security debt.

A price becomes actionable only when the buyer can interpret its unit. One cent per request, one dollar per document, and one dollar per successful result describe different obligations. A machine readable offer must identify the unit and the conditions that complete the purchase. Otherwise the client can compare numbers without comparing products.

We designed WunderCorp MPP around that position. WunderCorp MPP publishes payment aware API products for outcomes such as app planning, previews, completions, and image generation. Discovery metadata and structured schemas give a client enough information to evaluate and call an endpoint programmatically.

The model fits bounded outcomes better than open ended services. A request with uncertain duration or liability may still require a contract and a human approval process. Machine payment protocols do not eliminate commercial judgment.

We expect this claim to be measured. Permission adds another layer. A process may be allowed to spend five dollars on research and nothing on account changes. The payment request needs enough context for the policy engine to recognize that boundary. Commercial automation therefore depends on both the seller’s offer and the buyer’s delegated authority.

Quotes also need time boundaries. A seller may change price after discovery, while a client may hold an offer during a long task. An expiration field lets both sides know when the original terms no longer apply. Without it, a correct payment can still produce a dispute because the parties acted on different versions of the price.

Our position is practical. Agentic commerce will grow where the purchase can be expressed clearly. Price, permission, and result must all be legible to software.