The shift from content to executable commerce
Traditional web architecture assumes a human interprets product information and uses an interface. An autonomous or semi-autonomous agent needs machine-readable product data, explicit policies and authorised actions. Simply scraping the same page is not enough when the agent is expected to reserve stock, initiate payments or accept commercial terms.
The architecture therefore separates descriptive content from authoritative product, pricing and availability data, and from the actions the agent is permitted to execute.
Trustworthy action requires several layers
- Authoritative data: product, price, availability and eligibility information must come from systems that are responsible for those values.
- Policies: returns, restrictions and commercial rules must be represented as constraints rather than prose alone.
- Execution-time rechecks: a decision based on earlier information must be validated again before placing an order, reserving stock or initiating payment.
- Authorisation: the agent must operate inside explicit limits for identity, payment and permitted actions.
- Audit trail: the system should record enough detail to explain why an action was allowed and which data was used.
Protocol-agnostic boundaries
Protocols will evolve. The important architecture question is where authoritative commerce data and authorisation live inside the merchant system. A protocol adapter should expose that data and those capabilities, not become the only place where business rules exist.
This keeps the merchant system able to support multiple agent interfaces without duplicating eligibility or checkout logic across each protocol.
Human and agent channels must agree
If a website says a product is available but the agent interface exposes different stock or policy information, the business is presenting inconsistent versions of the same data. Agent readiness therefore depends on the same disciplines required for APIs and ERP/e-commerce integration: authoritative data, clear interfaces and reconciliation.
Commerce data, authorisation and interfaces for AI agents
The architecture brings these parts together with an emphasis on verifiable data, authorised actions and auditability rather than relying on prompt-layer automation.