Blog
Agentic commerce is coming, here is how a storefront prepares
AI agents are starting to shop on people’s behalf. Most storefronts are not built for that yet.

Shoppers are already letting AI shop for them
Ask around and most people will say they would never let a piece of software buy something on their behalf. The data says otherwise. Recent industry surveys put a majority of online shoppers already using an AI assistant somewhere in the purchase journey, comparing prices, summarising reviews, narrowing down options, and a meaningfully large minority already say they are comfortable letting an agent complete the purchase itself once it has done that work.
Amazon’s own shopping assistant, Rufus, now serves hundreds of millions of shoppers and has been credited with billions of dollars in incremental sales in its first full year alone. That is not an experiment anymore, it is a live distribution channel that most retailers have no visibility into at all.
The emerging standards behind it
Late in 2025, Stripe launched an Agentic Commerce Suite that lets any compatible AI agent complete a purchase directly with a merchant, and retailers including Anthropologie, Free People, Coach and Kate Spade went live on it almost immediately. A parallel open standard, generally referred to as the Agentic Commerce Protocol, is maturing quickly, with support for multi item carts arriving as one of its first serious upgrades.
None of this requires a retailer to build anything exotic. It requires a catalogue and a checkout that speak a language an agent can parse reliably, the same clean, unambiguous structure that makes a page legible to a person skimming quickly.
What a merchant actually needs to get right
An agent cannot infer what a page does not say. If a product’s availability, price or shipping window is only obvious from a visual layout, an agent reading the underlying markup may simply get it wrong, or skip the product. Clean product data and a responsive, well documented storefront matter more here than clever visual design, because the first audience reading the page is now frequently a machine, not a person.
The retailers already live on agentic checkout share one thing in common, an unusually disciplined data layer behind an otherwise ordinary looking storefront. The agent never sees the design, it only sees the data underneath it.
Where we start
When we build on Salesforce Commerce Cloud, we structure the catalogue, the policies and the content so both a person and an agent can read them without ambiguity, the same discipline behind our answer engine optimisation work, applied directly to a live storefront. Merchants who wait for agentic commerce to fully arrive before preparing for it will be doing this same structural work later anyway, under pressure, on someone else’s timeline.