Agentic commerce, explained

Agentic commerce is shopping that happens through an AI assistant instead of a website. A shopper asks ChatGPT, Gemini, Claude, Perplexity or Copilot to find and buy something, and the assistant — the agent — discovers products, compares options across stores, and completes the purchase on the shopper’s behalf using open protocols.

The store’s human-facing UI stops being the only way in. The agent reads your catalogue through a machine-readable feed and transacts through your checkout. A merchant who is “agent-ready” appears inside these assistants; one who isn’t is invisible to them.

Why this is happening now

Three things converged in 2025–2026:

1. Discovery is shifting into AI assistants. A growing share of “what should I buy” questions now starts in a chat assistant rather than a search engine.
2. The protocols stabilised. OpenAI/Stripe shipped the Agentic Commerce Protocol (ACP), Google and Shopify shipped the Universal Commerce Protocol (UCP), Google and the card networks shipped Agent Payments (AP2), and Anthropic’s Model Context Protocol (MCP) became the default way agents connect to live systems. See the protocol stack.
3. The platforms moved. Shopify launched Agentic Storefronts and opened its catalogue to non-Shopify merchants — proof that the catalogue feed, not the storefront UI, is the asset that matters.

What it means for merchants

The honest, evidence-based read:

• AI assistants do not become the store. Even in the bull case, the assistant is the discovery and conversation layer; it never becomes the merchant of record. Orders, payments, fulfilment and the customer relationship stay with the merchant. Value concentrates in whoever owns the agent-ready catalogue feed and the transaction backend.
• In-chat checkout had a false start. Native in-chat purchase launched in 2025, saw weak adoption, and the surfaces refocused on product discovery, routing the actual purchase back to the merchant’s checkout. The durable shift is discovery, not the in-chat payment UX — so be protocol-ready without over-investing in in-chat checkout while it settles.
• “Generate a store from a prompt” is now table stakes. The moat is being buyable by every agent and charging nothing for it.

How to be ready

1. Complete your product data. Agents rank and answer on structured attributes: image, price, GTIN, identifiers, a real description, brand, and category. Gaps make products invisible on some surfaces.
2. Serve the discovery manifests. Publish your /.well-known/ protocol profiles and a current product feed so agents can find and trust your store.
3. Expose an agent-completable checkout and an MCP server so agents can transact in plain language.
4. Measure it. Track which products are agent-eligible per surface and fix what’s missing.

On Mercentia all four are default-on, on every plan, at 0% transaction fees. The dashboard ships a catalog AI-readiness score that grades each product and auto-fixes the safe gaps, and every store serves the full protocol stack out of the box. See the agentic commerce hub for the live conformance matrix.