Agentic Commerce: How AI Agents Are Changing Ecommerce

Agentic commerce is shopping carried out by AI agents on a person's behalf: an assistant searches catalogs, compares options and, with the shopper's approval, builds a cart and checks out with the merchant. It runs on open protocols that standardize how agents and merchants talk: the Universal Commerce Protocol (UCP, Google and Shopify), the Agentic Commerce Protocol (ACP, OpenAI and Stripe), Google's Agent Payments Protocol (AP2) and the Model Context Protocol (MCP). In documented implementations the merchant stays merchant of record. It's live on some surfaces and early access on others.

Ecommerce has moved from browsing catalogs, to searching, to asking assistants for recommendations. Agentic commerce adds delegation: the assistant doesn't only suggest a product, it can act, such as checking stock, building a cart and completing checkout within limits the shopper sets. The diagram above shows the layers: assistant surfaces on top, protocols in the middle, merchant systems underneath.

Several standards emerged between 2025 and 2026. They overlap and are evolving, so treat this as a map, not a final picture.

Discovery moves partly into conversations the merchant doesn't see, so product data does more of the persuading. Checkout may happen outside the store's own pages, so policies and service need to be clear without the store's design around them. Attribution becomes channel-specific. Brand still matters, because shoppers ask for brands and assistants cite reviews and content, but the agent's shortlist rewards facts over slogans. See AI product discovery and product data for AI search.

Claims that agents will soon handle most shopping autonomously, that stores need special “AI files” to be seen, or that one protocol has already won aren't supported by current evidence. Google states no special markup or files are needed for AI Overviews or AI Mode (Google Search Central). Build for what's documented and measurable.

Agentic commerce, as used here, means AI agents acting for shoppers across stores. It's different from AI agents that work for your own team (catalog QA, support drafting, reporting), covered in AI agents for ecommerce, and from on-site assistants that help shoppers in your store, covered in AI shopping assistants. The three share foundations (structured product data, accurate policies, reliable APIs) but raise different questions about control and measurement.

Orders that arrive through AI channels still need fulfilment, service and returns. Check that these orders are identifiable in your admin, that confirmation and service emails work, that returns and disputes follow your normal policies, and that fraud checks apply. Decide how customer service will handle questions about what an agent told a shopper, since you may not see the conversation. These processes are emerging along with the protocols; review them as platforms publish more detail.

Measurement is still immature. Track what's available: orders attributed to AI sales channels in your platform, referrals from AI assistants in analytics, and post-purchase survey answers mentioning AI tools. Compare order value, returns and repeat rates for these customers with other channels over time. Avoid drawing firm conclusions from small early volumes. See ecommerce attribution.

Agentic commerce is the shift from assistants that suggest to agents that act, underpinned by emerging protocols in which merchants stay responsible for the sale. It's real but early and changing fast. The durable response is the same whatever happens: accurate data, complete feeds, clear policies and reliable checkout. For merchant preparation, see AI shopping agents; for Shopify, see Shopify agentic commerce.