What is agentic commerce? Definition and examples

Agentic commerce is buying and selling where an AI agent finds, compares and purchases products or services for a person, within limits that person sets.

How it works

The shopper states a goal ("running shoes under $120, size 10, delivered by Friday") and an AI agent does the legwork: searching, comparing, filling the cart and, with permission, paying. For that to work safely, merchants, agents and payment providers need shared rules, and several were published in 2025 and 2026:

• Agentic Commerce Protocol (ACP). Co-developed by Stripe and OpenAI and announced on September 29, 2025, alongside Instant Checkout in ChatGPT. Payment uses a Shared Payment Token that is "scoped to a specific merchant and basket total", and merchants keep control over what is sold and how orders are fulfilled (Stripe).
• Agent Payments Protocol (AP2). Announced by Google on September 16, 2025. It uses "Mandates", cryptographically signed records of what the user authorized: an intent mandate for the request and a cart mandate for the exact items and price. Google describes it as usable as an extension of the A2A protocol and the Model Context Protocol (Google Cloud).
• Universal Commerce Protocol (UCP). Published by Google in January 2026 with retailers including Shopify, Etsy, Wayfair, Target and Walmart. Merchants describe their capabilities in a JSON manifest at /.well-known/ucp, so agents can discover checkout and other features without custom integrations (Google Developers Blog).

These efforts overlap and are still evolving; as of October 2026 there is no single standard.

Why it matters for your website

Most B2B purchases don't end in an instant checkout, but the research phase is changing first. An agent asked to "shortlist three tools that fit a 30-person team and a $500 monthly budget" reads pricing pages, plan limits and integration lists. What it can't find or parse, it can't compare.

Practical steps that help both people and agents:

• Publish prices, or clear starting prices and what drives them.
• State plan limits in text and tables, not only in images.
• Describe your company and products with structured data.
• Make the next step, such as a trial, quote or demo, something an agent can complete or hand back to the user cleanly.

Example

A buyer asks their assistant to compare three scheduling tools for a 30-person team. The assistant opens each vendor's pricing page, extracts per-seat prices and seat limits, and returns a table. One vendor lists only "Contact us" for teams over 25, so the assistant marks its price as unknown and suggests the buyer ask for a quote. The other two are compared on published facts.

Related terms

• AI agent: the software doing the shopping.
• Agentic browser: where many of these tasks run.
• Structured data: machine-readable product and price facts.
• WebMCP: a proposed way for sites to offer actions to agents.
• Model Context Protocol (MCP): a standard AP2 can extend.