Agentic Commerce: How AI Shopping Agents Buy in 2026

Agentic commerce, defined

Agentic commerce is online buying where an AI agent discovers, decides, and checks out on the shopper’s behalf. The human supplies the intent; the agent does the browsing and the transaction. The line that separates it from a smart recommendation is the checkout: an agentic system does not just suggest a product, it completes the purchase — through an open protocol to the merchant and a payments standard that proves the shopper authorized the spend. For the full plain-language version, see what is agentic commerce.

The protocol stack

Agentic commerce runs on two layers that do different jobs. A commerce layer handles the checkout; a payments layer proves consent. They compose rather than compete.

• ACP — Agentic Commerce Protocol (OpenAI + Stripe) — Checkout: connects a buyer’s agent to a merchant to complete a purchase. Open standard, in beta.
• UCP — Universal Commerce Protocol (Industry working group) — The full commerce lifecycle around the transaction, not just checkout. Emerging companion to ACP.
• AP2 — Agent Payments Protocol (Google → FIDO Alliance) — Payment consent: verifiable credentials + cryptographic Mandates proving the user authorized the spend. Announced Sept 2025; v0.2 donated to the FIDO Alliance Apr 28, 2026.
• Card-network frameworks (Visa + Mastercard) — Trusted-agent acceptance on existing card rails. Visa Intelligent Commerce Connect (protocol-agnostic on-ramp) shipped Apr 8, 2026; Mastercard Agent Pay merchant framework.

The split matters: a store integrates the commerce layer to be discoverable and checkout-able, while the payments layer is what makes an agent’s spend auditable and authorized. For how the two compare and where Visa and Mastercard fit, see ACP vs AP2.

What changes for a store

Agentic commerce adds a second customer: the agent. Your storefront still serves the human, but an agent reads structured data and calls APIs, not product photography and persuasion copy. Being "agent-ready" is therefore a data and checkout problem — clean product feeds, machine-readable inventory and price, and a checkout an agent can call — not a redesign. The persuasion surface shrinks; the data surface grows.

Who the agents are

The buyers are AI assistants with checkout built in (ChatGPT), answer engines that complete purchases (Perplexity), and retailer-owned copilots. Each reaches the merchant through the same protocol layer, so integrating once exposes the store to many agents. The landscape of which agents matter, and what each needs from a catalog, is covered in AI shopping agents.

How to prepare

Start where agents read: a clean, complete product feed with stable identifiers, real-time price and stock, and policies (shipping, returns) in machine-readable form. Then make checkout callable — adopt the commerce protocol your agents use and wire the payments layer so authorization is provable. Sequence it by revenue: expose your highest-velocity, most-reordered SKUs first, because routine reorders are where agents land earliest.

The verdict

Treat the agent as a customer you have not designed for yet. The work is structured data and an agent-callable checkout, wired to a provable-consent payments layer — not a new storefront. Stores that expose clean data and adopt the protocol stack early get discovered and transacted by agents before their competitors are parseable at all. Protocol landscape last verified 2026-06-10; this space moves monthly.

What is agentic commerce?

Agentic commerce is online buying in which an AI agent — ChatGPT’s checkout, Perplexity, or a retailer’s own copilot — discovers products, decides between them, and completes the purchase on a shopper’s behalf. The human sets the intent ("find me running shoes under $120 that ship by Friday"); the agent does the browsing and the checkout. It is distinct from a chatbot that only recommends: the defining step is that the agent can actually transact, using open protocols to talk to the merchant and a payments standard to prove the shopper consented.

What is an example of agentic commerce?

A shopper tells ChatGPT to reorder a printer cartridge that fits their model and have it arrive before the weekend. The agent identifies the correct SKU, checks price and delivery across merchants, selects one, and completes checkout through the Agentic Commerce Protocol — confirming payment with a cryptographic mandate that proves the shopper authorized that exact amount. The person never opened a store. The earliest live examples are instant-checkout flows inside AI assistants and retailer copilots that complete the purchase in the chat rather than handing off to a website.

What does "agentic" mean here?

Agentic describes software that can take actions toward a goal with limited human supervision, not just answer questions. An agentic system plans, calls tools, and executes steps — in commerce, that means searching, comparing, and buying. The word matters because it marks the line between a recommendation engine (suggests, you act) and an agent (acts on your behalf within the bounds you set). The bounds are enforced by the payments layer, which only authorizes spend the user explicitly mandated.

What is the difference between e-commerce and agentic commerce?

E-commerce optimizes for a human navigating a storefront: product pages, images, reviews, a visual cart. Agentic commerce optimizes for a machine reading structured data and calling a checkout API: the persuasion surface shrinks and the data surface grows. In practice you now serve both at once — the human still browses, but an agent increasingly does the routine reordering and shortlisting. The store that wins exposes clean machine-readable product and inventory data and an agent-callable checkout, without abandoning the human experience.