Agentic commerce is the ecosystem in which AI agents participate in shopping tasks on behalf of people or businesses. Those tasks can include discovery, comparison, merchant interaction, transaction, payment authorization, and post-purchase coordination.
It is a macro movement, not a single product category. An external AI sales channel, a chat-contained shopping assistant, a merchant’s on-site AI shopping agent, and a payment-authorization protocol can all participate in agentic commerce while solving different jobs.
What makes commerce agentic
An agentic system can pursue a bounded goal through declared information and supported actions. In commerce, that might mean gathering product options, comparing offers, asking a merchant system for an available capability, or progressing a transaction under user-defined limits.
Three properties matter:
- State: What product, merchant, customer, or interface information can the agent use?
- Authority: Which actions is it allowed to request or perform?
- Verification: How does the system confirm the requested result occurred?
Natural-language conversation can be part of the experience, but conversation alone does not define agentic commerce.
The main interaction models
The market uses “agentic commerce” for several distinct surfaces.
External AI channels
The shopper asks an external assistant to find or compare products. Merchant data reaches that channel through crawling, feeds, catalogs, or platform integrations. The shopper may be referred to the merchant or complete part of the flow in the external channel.
Chat-contained shopping journeys
The conversation becomes the primary interface. Product results, recommendations, controls, and sometimes transaction steps render inside chat.
Alternative AI storefronts
A merchant or provider builds a new shopping experience around an AI-led flow rather than retaining the existing storefront as the main workspace.
Adaptive search and discovery
The search or discovery surface changes around intent while remaining part of the merchant experience. This can include conversational search, adaptive result layouts, and guided product exploration.
The ecommerce product-discovery map separates retrieval, recommendations, merchandising, and guided selling in more detail.
Connected existing-interface agents
The merchant’s existing interface remains the workspace. A connected host supplies current state and visible targets, exposes supported actions, and provides a resulting state that can be verified.
These models can coexist. The category label does not tell an operator which data, permissions, interface, or actions a product actually uses.
Agentic commerce and agentic storefronts are not synonyms
Agentic commerce names the broad ecosystem. An agentic storefront is one merchant-facing implementation within it.
The storefront term is also ambiguous:
- Shopify uses “agentic storefronts” for selling through external AI channels.
- Other vendors use it for AI-led alternative storefronts.
- Some use it for on-site assistants or adaptive discovery.
Use the interaction locus and implementation evidence rather than relying on the label.
The protocol layer
Several protocols address different functions:
- MCP and the experimental WebMCP proposal concern callable capabilities in server-side and browser-side environments.
- ACP and UCP describe commerce interactions with different scopes.
- AP2 addresses agent-led payment authorization.
- A2A addresses communication between agents.
They are not one ladder from basic to advanced. See the agentic commerce protocol stack for the full map.
What agentic commerce is not
It is not automatically:
- a chatbot;
- an on-site shopping assistant;
- a recommendation engine;
- a product feed;
- a checkout protocol;
- a promise of autonomous purchasing;
- evidence of higher conversion.
Those may be components, implementations, or claimed outcomes. Each needs its own proof.
What ecommerce operators should ask
When evaluating an agentic-commerce product, ask:
- What is the primary job: discovery, support, guided selling, transaction, or post-sale?
- Where does the interaction occur?
- Which state is connected, and how current is it?
- Can the shopper see the elements the agent refers to?
- Which actions does the merchant or host expose?
- What permissions and confirmation rules apply?
- How is the resulting state or transaction verified?
- What evidence supports any commercial outcome?
These questions separate category language from product behavior.
A bounded kn8 example
kn8 is an AI shopping agent designed to work through a connected storefront interface. It uses current state and visible targets the host exposes, guides the shopper in the interface already on screen, invokes only supported actions, and verifies what changed. That is a product mechanism within agentic commerce, not a definition of the whole category. No catalog, inventory, cart, checkout, Shopify, or outcome claim is implied without a connected demonstration. See the interaction model →
Key takeaways
- Agentic commerce is a macro ecosystem, not a single on-site product category.
- Different interaction models place the agent in different channels and interfaces.
- State, authority, and verification matter more than the label “agentic.”
- Protocols cover different functions and do not by themselves prove a merchant experience.
- Product behavior and commercial outcomes require separate evidence.