The Agentic Commerce Stack: Product Data, Discovery, Checkout, Payments and Fulfillment

The agentic commerce stack is the set of layers that let an AI agent shop for a person: the shopping agent, discovery, product data, the commerce API, cart, checkout, payments, order, fulfillment and post-purchase service. Protocols connect the agent side to the merchant side: UCP and ACP for commerce capabilities such as cart, checkout and orders, MCP as a tool and transport layer, and AP2 and network token programs for payment authorization.

Ownership is clear. Agent platforms own the agent and its interface. The merchant owns product data, prices, checkout state, orders and fulfillment, and stays merchant of record. Payment providers and networks own credentials and authorization.

The agent interprets the shopper's intent, applies their constraints (budget, size, delivery date, preferred brands), compares options and presents choices and confirmations. It belongs to the agent platform (an assistant such as ChatGPT or Gemini, a retailer's own agent or a third-party app). Merchants do not control it, but they influence it through data quality, policies and how easily their checkout can be completed. How this changes the classic funnel is covered in how AI agents change the ecommerce funnel, and the business overview in agentic commerce and AI shopping agents.

Agents discover products through catalogs and feeds the platforms index (Google Merchant Center, Shopify Catalog, OpenAI's product feeds), through UCP catalog search on merchants that support it, and through web search and page reading. Visibility depends on eligibility, data quality and the platform's own ranking. Content and search visibility for AI answers are covered in making your site discoverable in AI search.

Agents act on structured data: stable IDs that match checkout, per-variant options, price, availability, shipping, return and seller policies. This is the layer merchants control most directly and the one that most often decides success. See AI product feeds, product data for AI search and, for Shopify stores, Shopify agentic commerce.

Behind every agent channel sits the merchant's commerce logic: catalog, inventory, pricing, tax, cart, checkout, orders and returns. Build it once with agent-grade guarantees (typed schemas, idempotency, delegated authorization, signed webhooks) and expose it through protocol adapters and MCP tools. See agent-ready ecommerce API.

Carts collect items; checkout sessions turn them into an order with authoritative totals, tax, fulfillment options and status. ACP and UCP both define checkout operations (create, update, retrieve, complete, cancel) with explicit statuses, escalation when the shopper must act and strict idempotency on completion. See agentic checkout and ACP vs UCP vs MCP.

Payment authorization stays with payment providers, networks and issuers. Agent channels use scoped credentials (delegated or shared payment tokens, wallet credentials, network agentic tokens) and, increasingly, signed mandates that record what the user authorized. Keep agent authorization, purchase consent and payment authorization separate. See AI commerce payments.

Agent orders enter the same order management and fulfillment systems as any other order, tagged by channel. Signed webhooks keep the agent platform informed of creation, shipping and delivery; UCP defines an order capability with webhook-based updates, and ACP defines order and fulfillment event structures. Operations, fraud signals and confirmation accuracy for these orders are covered in orders placed by AI agents, and order flow design in ecommerce order management.

After purchase, the shopper may ask the same agent where the order is, to cancel a line or to start a return. That requires order lookup, cancellation and return capabilities with the shopper's delegated authorization, and clear handoff to human support when needed. See orders placed by AI agents and the ecommerce post-purchase experience.

Measurement spans layers: AI referral traffic to your site, orders placed inside assistants, assisted journeys that end on your site, and post-purchase outcomes such as returns and support contacts by channel. See tracking AI-referred ecommerce sales for a reporting framework.

Agentic commerce is not one technology but a stack. Agent platforms own the conversation; merchants own data, checkout, orders and fulfillment; payment systems own authorization; and protocols connect them. Strengthen the layers you control, starting with product data and order operations, build commerce capabilities once behind your own API and adopt protocols as your channels require.