COM-330 · Lesson 1 of 12 · Module 1: How AI Agents Shop · 3:15

How an agent picks

AI shopping comes in three kinds: an AI answer names a few products, an assistant compares them and sends the buyer to a shop, and an agent buys within the limits the buyer set. An agent turns the request into must-haves, reads each candidate’s facts, checks every must-have, drops any product that fails one or whose fact it can’t find, ranks the rest by price, reviews or delivery, and stops to show the buyer its pick and the full total before any money moves. A better product with one unreadable fact loses to a weaker one whose facts are complete. Worked example: a buyer asked for a 12-inch induction pan under $150 delivered by Friday; Brindlewood’s skillet met every must-have, but its page said only “Ships fast!”, so the agent picked a rival’s pan; Brindlewood was picked in one of ten agent tests.

Frame from lesson 1 of Sophyx Certified Agentic Commerce – Specialty, "How an agent picks"

What you will learn

  • Tell an AI answer, an assistant and an agent apart
  • Trace how an agent checks a buyer’s must-haves
  • Explain why an unreadable fact drops a product

The one thing to remember

Agents pick what they can check.

Common mistake: Assuming agents pick the best product

An agent picks the product it can check, so a better product with one missing fact loses to a weaker one whose facts are complete.

The fix: Make every likely must-have readable on your product pages.

Key terms

Shopping agent
AI that compares products and can buy for a person.
Must-have
A condition the buyer sets that every pick must meet.
Buyer approval
The buyer’s yes to the exact item and total, before paying.

Transcript

The brands in the worked examples are fictional, and the numbers are illustrative.

A home cook asked her AI assistant for a twelve-inch steel pan for induction, under a hundred and fifty dollars, by Friday.

Brindlewood's skillet fit every wish. The agent left it out.

Welcome to Agentic Commerce. Lesson one is about how an agent picks.

You'll learn three kinds of AI shopping, how an agent checks must-haves, and why one unreadable fact drops a product.

First, a quick recall from the Answer Engine Specialist course. What does AI trust when it builds an answer?

Agreement, the same facts in many places, and a shopping agent checks facts even more strictly.

AI shopping comes in three kinds.

An AI answer names a few products. An assistant compares them and sends you to a shop.

A shopping agent goes one step further, and buys for you within the limits you set.

So how does an agent pick? It turns the request into must-haves, like a size, a price and a delivery date.

Then it checks every must-have against each product's facts, and drops a product that fails one or can't be checked.

The third shop's nonstick pan failed on induction, and Pellmont's pan passed all five.

Brindlewood's skillet passed four, but its page only said ships fast, so the agent couldn't check Friday.

Whatever is left, it ranks by price, reviews or delivery.

Here's a quick check. Why did the agent drop Brindlewood's skillet?

It couldn't check the delivery date, because ships fast gave it no date to hold against Friday.

A careful agent shows you its pick and the full total with tax, then waits for your approval.

She taps yes, and only then does money move.

So your facts have to win before that moment, because the buyer only sees what the agent kept.

Theo runs Brindlewood's online shop. He sent that request ten times, and his skillet was picked once.

It ships the same day and would have arrived Thursday, but the date wasn't on the page.

The agent didn't choose the better pan. It chose the pan it could check.

A common mistake is assuming that agents pick the best product.

Instead, make every likely must-have readable on your product pages.

This week, write three real buyer requests with must-haves, and run each in two AI assistants.

Then note the must-haves your product can't prove yet.

Before we wrap up, can you name the four steps an agent takes before it buys?

It writes must-haves, checks each fact, drops what fails or can't be checked, and waits for your yes.

So remember, agents pick what they can check.

Now try the practice quiz, then join Theo in lesson two.

Part of Agentic Commerce Specialty (COM-330)

For e-commerce and product teams: get products picked when AI agents shop, with clear facts, clean feeds and a checkout an agent can finish.

12 lessons, practice quizzes, a final exam and the Sophyx Certified Agentic Commerce – Specialty certificate. Regular price USD 89. Your first certification is free. Free with any paid Sophyx plan.

Start the course free
Book a Demo