.gif){kind=spacer}
Agentic Commerce Glossary for Retailers: The Terms Retail Leaders Need to Know in 2026
An agentic commerce glossary helps retail leaders separate AI-assisted shopping from agent-executed transactions and distinguish product-data work from checkout, payment, and interoperability work. These terms matter because artificial intelligence can mediate discovery through purchase, while the technical standards that connect agents, retail systems, and payments serve different roles.
McKinsey’s agentic commerce consumer research, published on October 28, 2026, surveyed approximately 21,000 consumers across 11 markets. It measures AI-mediated shopping behavior at scale, but it should not be read as evidence that completed agent-led purchases are already universal. McKinsey
What is agentic commerce?
Agentic commerce is shopping in which artificial intelligence (AI) agents mediate or execute parts of the journey, from discovery to purchase, on a consumer’s behalf. The term is not limited to an agent completing checkout. It can also describe product discovery, comparison, basket creation, or another step in the journey. McKinsey
Conversational does not automatically mean agentic. A chatbot with canned responses is conversational but not agentic, according to Forrester. Retailers also need to distinguish between owned experiences on their sites or apps and non-owned experiences in external answer engines. The retailer has different control over data, presentation, and transaction flow in each setting. Forrester
The category is still developing, so treat these definitions as working terms rather than one industry-wide standard. McKinsey focuses on agents mediating or executing the journey, while Forrester also examines where the experience occurs and how much action it takes.
The agentic commerce glossary
Use this glossary to identify whether a term concerns the shopping experience, retail data, commerce transactions, or payment authority.
Sort the terms with a four-question framework
This is a decision tool, not a maturity model. Start with the layer that matches the business problem.
- Can an agent understand the offer?
Review product descriptions, structured data, feeds, variants, shipping, returns, and availability. This is primarily an ecommerce and product-data responsibility. Google’s merchant listing documentation identifies product details such as price, availability, shipping, and returns as structured data fields that systems can interpret. Google Search Central - Can an agent access approved retail information or actions?
Review application programming interfaces (APIs), permissions, and Model Context Protocol connections. Technology teams should decide which product, inventory, customer, or order information an AI application may access and which actions require approval. MCP specification - Can an agent create and complete a transaction?
Review cart management, checkout sessions, fulfillment choices, payment processing, and order handling. Commerce protocols such as ACP and UCP address these interactions, but their supported surfaces and integration routes differ. ACP UCP - Can the retailer verify payment authority?
Review payment controls, mandates, constraints, user confirmation, and transaction records. AP2 focuses on payment authority and verifiable intent, including checkout and payment mandates. AP2
The distinction is direct: MCP connects AI applications to tools and context; A2A connects agents to other agents; ACP and UCP specify commerce interactions; AP2 focuses on payment authorization and verifiable intent. These protocols can work together, but they do not perform the same job. MCP A2A ACP AP2 Google UCP
Before implementation, confirm the protocol version, integration route, available surface, and partner support. The UCP page lists some capabilities as upcoming and includes a waitlist. AP2 states that more integrations are in progress and that standardization work continues.
Worked example: a hypothetical retail flow
Hypothetical
A shopper asks an AI assistant:
“Find a waterproof hiking jacket under $180 in navy, size medium, available for delivery by Friday.”
Structured product data describes the item and offer. A tool connection can retrieve approved inventory or order information. An agentic checkout flow can manage the cart, fulfillment option, and payment step. Before payment, mandate-based controls can record the shopper’s constraints or authorization. Webhooks can communicate shipment and refund updates. Google Search Central Stripe AP2 MCP specification
Use this flow to assign ownership:
- Ecommerce and merchandising: product data, variants, price, availability, shipping, and returns.
- Technology: approved tool access, APIs, identity, and integration controls.
- Payments and risk: payment authority, mandates, constraints, and transaction records.
- Operations and customer experience: order status, delivery events, refunds, and support.
The point is not to select one protocol for every layer. It is to identify the missing connection and give it to the team responsible for that connection.
Common misconceptions about agentic commerce
- “Agentic commerce means fully autonomous shopping.” Not necessarily. In OpenAI’s initial Instant Checkout flow, shoppers confirmed order, shipping, and payment details before completing a purchase. OpenAI
- “Schema markup enables agentic checkout.” No. Schema markup describes product-page information. ACP separately specifies checkout, delegated payment, and webhooks. Google Search Central Stripe
- “MCP, A2A, ACP, UCP, and AP2 are interchangeable.” They address different connections or stages. MCP handles tools and context, A2A handles agent communication, ACP and UCP handle commerce interactions, and AP2 addresses payment authority.
- “Protocol compatibility proves a live rollout.” It does not. Check the supported surface, implementation status, partner route, and operational responsibilities. Google’s UCP page identifies some capabilities as upcoming and invites merchants to join a waitlist. Google UCP
