What Is Agentic Commerce? A Builder’s Guide

Agentic commerce is shopping in which an AI agent finds, compares and sometimes buys for a person. This guide explains the layers an agent works through and the data each one needs, with a step-by-step demo on real prices.

Affiliate.com Affiliate.com what is agentic commerce graphic

Agentic commerce is shopping in which an AI agent does some or all of the work for a person. The agent finds products, compares them, checks whether a price is a good one and, in some cases, completes the purchase. The person sets the goal, such as a gift under a budget or the best price on a named product, and the agent handles the steps in between. This guide explains what that means in practice, the layers an agent works through and what each layer needs from the systems around it. The walk-through below uses real prices last pulled November 19, 2026.

What agentic commerce means

The term covers a range, and it helps to say where on the range a product sits. At one end is an assistant that helps a shopper decide: it searches, compares and explains, and the shopper does the buying. At the other end is an agent that can complete a purchase on the shopper’s behalf, within a budget and rules the shopper has set. Most of what exists today is toward the first end, and the second end is where much of the new engineering is going.

LevelWhat the agent doesWhat the person does
Agent-assistedFinds products, compares prices, explains the differences and builds a shortlist or a cart.Reviews the answer and buys at the store.
Agent-ledDoes all of the above and starts or completes the purchase using credentials the shopper has delegated.Sets the goal, the budget and the rules, and approves what the agent is allowed to do.

Both levels depend on the same foundation. An agent can only buy a good product at a good price if it can find the product, tell it apart from similar ones, see what it costs in each store and know whether it is in stock. The rest of this guide is about that foundation.

The layers an agent works through

Answering one shopping request takes a handful of steps, and each one needs something different from the world outside the model.

LayerWhat the agent doesWhat it needs
1. AskWorks out what the shopper wants.The model, and a clear request.
2. FindSearches for the product and identifies it.A product catalog it can search, and a stable identifier such as a barcode.
3. CompareLooks at every store that sells it.Today’s price and stock for each store, with the currency.
4. CheckTests whether the price is a good one.The price history for the same product.
5. AnswerWrites the recommendation.The figures from the earlier layers, so nothing is guessed.
6. Hand offSends the shopper to the store, or completes the purchase.A link, or a checkout and payment protocol the store supports.

Layers 2 to 4 are product data. Layer 6 is checkout and payment. Several protocols have been published for that last layer, including the Agentic Commerce Protocol from OpenAI and Stripe and Google’s Agent Payments Protocol, and the Model Context Protocol is a general way of connecting an agent to tools and data, not only for shopping. They solve different problems and a builder can use more than one.

Walk through one request

The demo below follows a single request through the six layers. Pick a product, then step through. Each layer shows what the agent does and the real data it needs, taken from one Price History API response for that product’s barcode. The page holds a rotating set of products that is rebuilt every night, and it offers three of them, a different three on most visits.

Two things are worth noticing as you step through. The first is that almost everything the agent says in layer 5 was fetched, not remembered: the store, the price, the stock and the range all come from the data. The second is that layer 6 depends on the earlier ones. A smooth checkout does not help if the agent sent the shopper to the wrong product or a price that had changed.

Who builds with it

Teams building agentic commerce come from several directions. Shopping assistants and comparison sites add an agent to answer product questions. General assistants add product answers so a shopper does not have to leave the conversation. Publishers add an agent to a gift guide or a review so that readers can ask follow-up questions. Retailers and brands build agents that answer for their own catalogs.

They share one problem, which is getting product data an agent can rely on. Our post on choosing between an MCP layer and an API covers how to connect that data, and LLM product data API covers how to hand it to a model.

What the data layer has to get right

A shopper judges an agent by its answers, and its answers are only as good as the data behind them. Four requirements come up repeatedly.

RequirementWhy it mattersMore
Product identitySimilar names, colors and bundles are different products. A barcode keeps them apart.How LLMs fail on product identity
Current price and stockA price that was right last week is a wrong answer today.Handling out-of-stock and price changes
Every store, in one responseComparing means seeing all the stores that sell the barcode, not one store’s page.Barcode search
A way to connectThe same data should be reachable as a tool by an agent and as a request by an application.MCP integration for AI shopping

What the November 12, 2026 pull shows

The comparison layer is where an agent earns its keep, so we looked at it. On November 12, 2026 we pulled every store that sells each of the five products in that day’s set. Two of the five products with two or more stores in stock had a gap of at least ten percent between the cheapest and the most expensive store. Every store shown had the product in stock that day.

ProductStoresIn stockCheapest in stockHighest in stockGap
Weber Q 2800+ gas grill66$449.00$449.000%
Google Nest Learning Thermostat66$225.98$279.9924%
Tempur-Pedic Ergo neck pillow33$134.00$134.000%
Bissell Little Green Mini cleaner66$79.99$85.977%
Netgear Orbi 370 mesh Wi-Fi, 2-pack33$199.99$249.9925%

This table is fixed to November 12, 2026. The walk-through above is rebuilt every night with a different set of products, so its figures will change. Treat the table as a record of one day.

A handful of products is a small sample and not a measure of any market. Each figure is a store’s current price in US dollars on the day of the pull, and the gap is the difference between the cheapest and the most expensive in-stock store, as a percentage of the cheapest.

Terms builders use

TermWhat it means here
AgentA program built on a language model that can take steps, call tools and use the results, instead of only answering.
Tool callA request the agent makes to a function or service, with arguments, to get something it does not know.
MCPThe Model Context Protocol, a standard way of exposing tools and data to an agent.
GroundingBasing an answer on data that was fetched, not on what the model remembers.
CatalogThe set of products an agent can search, with their prices, stock and identifiers.
BarcodeA product’s UPC, EAN or GTIN. It identifies the product itself, regardless of which store sells it.

Getting started in three steps

If you are building the data layer of an agent, three steps get a first version working.

StepWhat to do
1. Get a keyRequest an API key from Affiliate.com. REST requests send it as a Bearer token in the Authorization header, and an MCP client is configured with the same key.
2. Find, then identifySearch for the product by name, keep the barcode of the result you choose and use that barcode for everything after. Assistant-ready product search goes deeper.
3. Compare and checkMake one price history request for the barcode. It returns every store with its price, stock and range, which is what layers 3 to 5 need.
One API call
The data layer for your agent
  • ✓Search, compare and check a product by barcode
  • ✓Every store, with today’s price, stock and range
  • ✓The same data as an MCP tool or a REST request
Get API access →
Request
{
  "target_type": "barcode",
  "target_ids": [
    "606449172560"
  ],
  "currency": "USD",
  "include_series": false,
  "include_stats": true,
  "include_product": true,
  "per_page": 25
}