A year ago almost nobody said "agentic commerce". Now it turns up in every ecommerce conference talk. The idea behind it is simple: an AI that doesn't just suggest products but actually does the buying for you.

What has happened in practice is less simple. There have been big launches, at least one notable retreat and some fairly clear data on what people are willing to hand over to an AI. If you run an online store, it's worth knowing the story before deciding whether any of it affects you.

What it actually means

An AI agent is an assistant that acts as well as answers. When shopping, that can mean searching several stores, comparing options, filling a basket and, at the far end, paying.

It helps to picture three steps:

  • It answers. You ask about a product and it explains it.
  • It recommends. You describe what you need and it suggests options.
  • It buys. You say what you want and how much you'll spend, and it handles the rest.

Almost everything called "AI shopping" until now has sat on the first two steps. Agentic commerce is the third: the shopper hands over the decision, or at least the admin.

What has happened in the past year

In September 2025, OpenAI started letting people buy inside ChatGPT without leaving the conversation. It called this Instant Checkout and paired it with an open standard, the Agentic Commerce Protocol, built with Stripe so that stores could plug in.

In January 2026, Google announced its own standard, the Universal Commerce Protocol, developed with Shopify, Etsy, Wayfair, Target and Walmart. Through it, shoppers in Google's AI Mode and in Gemini can pay some US retailers without leaving Google, using the payment and delivery details saved in Google Wallet.

Two months later came the first setback. Walmart, which had put around 200,000 products on sale inside ChatGPT, told WIRED that those purchases converted at a third of the rate it saw when shoppers clicked through to its own site. Daniel Danker, one of its executives, gave a very down-to-earth reason: people worried that if every item was paid for separately, five boxes would turn up instead of one order.

A few days after that, OpenAI changed course. It said the first version of Instant Checkout didn't offer the flexibility it wanted to provide, let stores use their own checkout, and turned its attention to product discovery. Since then ChatGPT has been mainly a place to compare and discover, and stores can close the sale through their own checkout.

Google, meanwhile, has pressed on. In May it announced a basket that gathers items from different stores, flags price drops and lets people pay through Google or move over to each store's site. Google said it would reach the US from this summer, in Search and Gemini, with Canada and Australia in the following months and the UK later. For payments the agent makes on its own, the shopper can set limits: which brands, which products and how much it can spend.

What shoppers think

In a survey of just over 2,000 US adults carried out by The Harris Poll for Visa in May 2026, 72% had already tried an AI assistant, yet only 23% would trust generative AI to make a payment by itself.

A separate survey by Checkout.com, across six markets, asked what it would take for people to feel comfortable letting an agent shop for them. The most common answers were a spending cap (30%), being able to withdraw permission instantly (29%) and easy cancellation (28%). One in four said they would never let an AI make purchases for them.

Put together, the pattern is fairly clear: most people have already tried AI assistants, but very few would let one pay on its own. It's what most of us would do with someone we'd only just met: take their advice, but keep hold of the card.

What it means for an online store

First, some perspective. Nearly everything that has launched works only in the US for now. And the Walmart case shows that putting checkout inside a chat doesn't guarantee more sales. None of this means you need to change your store tomorrow.

What does make sense now, because it pays off whichever way things go:

  • Look after your product data. An agent cares little about your design; it relies mostly on the data in your product pages. If sizes, materials or compatibility details are missing, it can't recommend your products with confidence. Search engines and assistants need exactly the same thing, as covered in how to prepare your catalogue for AI.
  • Don't neglect your own site. The lesson of the past year is that outside assistants help people discover, but purchases still mostly end on the store's own site. A clear, fast site with a checkout that holds no surprises is still what converts.
  • See what your platform offers before building anything yourself. Shopify helped build Google's standard and its stores are among those the new basket supports, so much of this is likely to reach you through the platform, though only in the US for now. If you use something else, wait to see what it announces before investing in your own integrations.

The shopper who does come to your store

While outside agents mature, most of your customers still arrive on your site, browse and get stuck on a question: whether that size will fit, which of two models suits them, whether a product will do the job they have in mind.

Nobody needs to buy on their behalf at that point. They need someone to help them choose, the way a good shop assistant would, while the decision and the payment stay theirs. It's help that doesn't ask anyone to hand over payment, which is exactly what holds shoppers back in the surveys, and it's what SmartShop AI focuses on: an assistant inside your store that answers and recommends from your catalogue, without buying for anyone.

Agentic commerce will probably find its place, especially for repeat or very simple purchases. But on what we know today, the advantage goes to stores with good product information and a site that helps people decide, wherever the purchase ends up happening.