Agentic Commerce Explained: How AI Agents Will Buy Things for Your Customers

Agentic commerce is the model where AI agents act on behalf of customers. They understand intent, search catalogs, compare options, build carts, and complete purchases. The customer never touches a traditional storefront. ChatGPT, Perplexity, Gemini, and Claude are already shipping versions of it in 2026.

This is not a feature added on top of e-commerce. It is a structural shift in how products get discovered and bought.

The traditional purchase flow is linear. The customer thinks of a need, opens Google or a retailer site, browses, compares, clicks, and pays. Every step is manual. Every step happens on a surface the merchant controls.

In agentic commerce, the customer describes intent in natural language. The AI agent interprets it, queries one or several catalogs in real time, evaluates options across price, availability, and structured attributes, builds a recommendation, and executes the purchase. The merchant surface becomes optional.

The agent does not guess. It reads structured product data: titles, prices, stock, shipping conditions, return policies. If the data is clean, it gets selected. If the data is broken, it gets skipped.

A real example. A user opens ChatGPT and types "I need a winter jacket for hiking, waterproof, under 200 euros, delivered to Paris by next week."

The agent breaks the request into structured filters: product type, use case, attributes, price ceiling, delivery zone, delivery deadline. It queries connected merchant catalogs through protocols like OpenAI's Apps SDK (the layer that exposes a merchant catalog inside ChatGPT) or Google's Universal Commerce Protocol. It pulls real products with real prices and real stock from real stores.

The agent presents 3 to 5 options. The user can swap, ask for cheaper alternatives, or refine constraints in plain language. Once the cart is finalized, the user clicks to validate. Payment happens either natively in the agent surface or on the merchant site through a redirect, depending on the integration.

The whole flow takes minutes. No browsing. No filters. No tabs.

The infrastructure is no longer theoretical. Several major retailers have moved.

Carrefour launched the first European grocery app inside ChatGPT in March 2026, letting 26 million French users build carts through conversation. Shopify integrated with OpenAI's Apps SDK, exposing every Shopify merchant to ChatGPT users by default if they opt in. Walmart, Target, and over 20 other retailers joined Google's Universal Commerce Protocol and are building Gemini integrations. Booking, Expedia, and Adobe ship dedicated apps inside ChatGPT.

According to Google's NRF 2026 announcements, agentic commerce features will roll out to all merchants connected to Merchant Center over the next 18 months.

The pace of adoption is unlike any previous commerce shift. Mobile commerce took nearly a decade to become table stakes. Agentic commerce is compressing that timeline into quarters.

Carrefour announced its ChatGPT integration in March 2026 and exposed 26 million French customers to a conversational grocery experience in one move, according to Carrefour's own announcement. Google launched UCP in late 2025. Within four months, Walmart, Target, and over 20 other retailers had joined. Shopify flipped the switch on OpenAI Apps SDK integration and every opted-in merchant became immediately reachable inside ChatGPT.

What makes this different is the infrastructure layer. UCP, Apps SDK, and MCP (the Model Context Protocol used by Claude to connect to external systems) are open standards. Once a merchant plugs in, the exposure is not gradual. It is immediate across every agent that supports the protocol.

This is why the 18-month window in Google's announcement is misleading. For merchants who act in the next two quarters, it is a land grab. For everyone else, it is a catch-up.

Agents do not pick brands the way humans do. There is no logo, no design, no marketing copy. Just data and context.

• Catalog readiness. Real-time stock sync, complete product attributes, clean variants, accurate pricing, structured shipping and return terms. Missing data means the agent skips the product.
• Protocol compatibility. UCP for Google, Apps SDK for OpenAI, MCP for Claude. Merchants need to expose their catalog through at least one of these open standards or risk being invisible to entire agent ecosystems.
• Trust signals at the data layer. Reviews, ratings, return rates, and structured customer feedback all feed into how the agent ranks options. These are no longer marketing assets. They are decision inputs.

Brand still matters when the user asks for a specific brand. When they do not, the agent decides, and the agent decides on data.

The center of gravity moves. The website is still useful, but it is no longer the only entry point. Most of the strategic work happens upstream of the customer ever seeing a page.

Three concrete shifts:

• Product feed becomes the new storefront. Whatever your feed says is what the agent sees. Investments shift from page design to data quality.
• Conversion optimization extends into the agent layer. A clean structured response to "what is your return policy in plain text" matters more than a hero banner.
• Reach is no longer earned through ads alone. It is earned by being readable. A small merchant with a clean UCP-compatible feed can be selected next to Walmart in the same agent response.

This is also why getting cited by AI search engines matters in parallel. The same agents that recommend products also recommend the editorial content that frames the buying decision. Visibility and selection are linked.

You do not need to wait for the next platform launch. The infrastructure is live and the early movers are gaining ground.

• Audit your product feed. Completeness, accuracy, freshness. Every missing attribute is a missed selection.
• Choose at least one protocol. UCP if you are on Google Merchant Center, Apps SDK if you are on Shopify or want OpenAI reach.
• Structure your trust data. Make returns, shipping, and policies machine-readable, not buried in PDFs.
• Test the agent surfaces. Search your own products in ChatGPT, Perplexity, and Gemini. See what they return. Fix what they get wrong.

This is exactly what we build at L'Atelier Growth. Product feed infrastructure, protocol integrations, and structured commerce data that make catalogs readable by AI agents across every surface they operate on.

This is not consulting. These are systems we design, build, and operate. If your catalog is not ready for the agent layer your customers are already using, get in touch.