Agentic Commerce

Agentic commerce is online buying in which an AI agent acts on a person's behalf: it searches for products, compares options, builds a cart and, with the user's permission, completes the purchase and handles follow-up such as order tracking. It is the transactional end of AI shopping. The assistant does not just recommend a product; it buys it within limits the user has set.

The idea is ahead of everyday behavior, and 2026 showed both progress and limits. OpenAI launched Instant Checkout in ChatGPT in September 2025, then wound it down as a native feature in March 2026, saying it lacked flexibility; reporting at the time said few merchants were active and users researched in ChatGPT but rarely bought there. Purchases from ChatGPT now complete on merchant sites or in merchant apps. Other players kept building. Microsoft launched Copilot Checkout on January 8, 2026. Google offers agentic checkout on price-tracked items, where it buys the product once the price falls within the shopper's budget after they confirm, and introduced Universal Cart at I/O 2026. Amazon's Alexa for Shopping can auto-buy at a set price and use Buy for Me to purchase from other retailers. Some analysts argue that many 'agentic' checkouts are really buy buttons inside a chat, with little autonomy involved, which is a fair caution when reading vendor claims.

A set of open protocols is forming the plumbing, each handling a different layer:

• The Agentic Commerce Protocol (ACP), maintained by OpenAI and Stripe with Meta on its steering committee, covers product feeds, checkout sessions and delegated payment tokens.
• Google's Universal Commerce Protocol (UCP), launched January 11, 2026 and co-developed with Shopify, Etsy, Wayfair, Target and Walmart, covers checkout, cart transfer and identity linking on Google's AI surfaces and beyond.
• The Agent Payments Protocol (AP2) proves what an agent is authorized to buy using signed mandates. Google donated it to the FIDO Alliance in April 2026.
• x402 embeds payments in HTTP so agents can pay per request, and is now governed by the x402 Foundation at the Linux Foundation.
• The Agent2Agent protocol and the Model Context Protocol let agents talk to each other and reach merchant tools and data.

For merchants, the requirements are practical. Agents cannot interpret a hero banner, so they rely on structured catalog data: price, stock, variants, shipping times, return policies and reviews, delivered through product feeds and consistent product pages. Where checkout runs through ACP, UCP or a payment partner, the merchant usually stays merchant of record and keeps the customer relationship, but has to support the integration and decide which agents to accept. Clear, machine-readable policies matter because an agent comparing two similar offers can only weigh the terms it can actually read.

Measuring readiness has two parts. First, visibility: do assistants select your products for the prompts you care about, with correct prices and stock? Second, transactability: can a supported agent complete a purchase with you, on which platforms, and how often does it fail? The second is newer and depends on which protocols and partners you integrate. Our guide on how ecommerce brands can prepare for agentic commerce walks through the steps, and agent experience optimization covers making the rest of a site usable by agents.