Agentic commerce: how AI agents discover, compare and buy products
Agentic commerce is buying done through AI agents that research, compare and purchase on behalf of a person or a company. Instead of a shopper clicking through a store, an agent reads product pages, structured data and APIs, and completes the purchase within rules the buyer sets.
How an AI agent buys a product, step by step
- Understand the request. A person or a procurement system asks for something: "reorder corrugated boxes under $0.50 a unit" or "find a standing desk with next-week delivery".
- Discover candidates. The agent searches the web, queries product feeds or APIs, and browses supplier sites. Sites it cannot reach or read drop out here.
- Read and compare. It reads product pages and structured data for price, currency, unit, availability and specifications, then compares options.
- Choose within the rules. It picks an option inside limits the buyer set: approved vendors, contract prices, spend limits. Exceptions go to a person.
- Check out. It completes checkout, as a guest or with the buyer’s account, or hands the last step to a person to approve.
- Confirm and record. The order is confirmed, and for businesses it flows into procurement and finance systems.
Every step depends on the site being readable and usable by software. A site that hides its price, challenges automated visitors or needs a person at each step is likely to be passed over before a human ever visits. See what agents look for in what agent readiness means.
From people browsing to agents buying
- Yesterday: buyers visit your site, read your pages, compare prices and check out themselves.
- Today: buyers ask an AI assistant for options. Agents read product pages, structured data and APIs to decide who makes the shortlist.
- Next: procurement agents place orders against approved vendors, contract prices and spend limits. A person approves the exceptions.
Standards that are emerging
Three open standards come up most. MCP (Model Context Protocol) connects AI models to tools and data. ACP (Agentic Commerce Protocol) was co-developed by OpenAI and Stripe for agent checkout. UCP is Google’s commerce protocol. Support is still early and changing, so a readable site and a checkout an agent can reach remain the foundation.
Why B2B agentic buying is harder
Agent checkout built for one shopper paying by card does not cover business buying. An agent can only buy from a business if it can see and use these eight things:
Contract pricing
The price an agent sees has to match what its company negotiated, not the list price.
Units and minimums
Each, box, case and pallet, minimum order quantities and quantity breaks must be unambiguous.
Account catalogs
Approved items, entitlements and ship-to rules differ by customer and by location.
Approvals and limits
Orders above a threshold need sign-off. An agent needs a way to request and track it.
POs and credit terms
Net terms, credit limits and purchase order numbers replace a card payment at checkout.
PunchOut, EDI, ERP
Orders have to flow into the buyer’s procurement and finance systems, not stop at a cart.
Stock by warehouse
Availability depends on location, allocation and ship date, not on a single number.
Tax and compliance
Exemption certificates, restricted goods and invoices have to be handled correctly.
What sellers should do
Make prices, stock and product data readable by software, let declared agents through bot protection, keep search and checkout usable without friction, and expose the business rules your B2B buyers depend on. The Axidort Index shows how 50 leading sites do today, and the scanner below scores your own.
Frequently asked questions
01What is agentic commerce?
Agentic commerce is buying done through AI agents that research, compare and purchase on behalf of a person or a company. The agent reads product pages, structured data and APIs, and completes the purchase within rules the buyer sets.
02How do AI agents buy products?
They interpret the request, search or browse for matching products, read price and availability, compare options, choose within the buyer’s limits and then complete checkout or ask a person to approve. Each step depends on the site being readable and usable by software.
03What are MCP, ACP and UCP?
They are emerging open standards for agents and commerce. MCP (Model Context Protocol) connects AI models to tools and data. ACP (Agentic Commerce Protocol) was co-developed by OpenAI and Stripe for agent checkout. UCP is Google’s commerce protocol. Support is still early, so a readable site and a working checkout remain the foundation.
04Is B2B agentic buying different from consumer agentic shopping?
Yes. Business orders involve contract prices, units and minimums, approvals and spend limits, purchase orders and credit terms, and integration with procurement and finance systems. Checkout designed for one shopper paying by card does not cover them.