For most of e-commerce history the job was legible: get a human to your site, and make the site good enough that they buy. Every dollar of SEO, every conversion-rate tweak, every abandoned-cart email served that one motion — human, to storefront, to checkout. That motion is being quietly replaced, and the businesses that notice late will find their best customer is a piece of software that never loads their homepage.
Two numbers from one holiday
Adobe's analytics on the 2025 holiday season are the clearest signal I have seen. Traffic to US retail sites coming from generative-AI tools rose 693% year over year — up 769% in November and 673% in December.1 And this traffic stopped being tyre-kickers: AI referrals converted about 31% better than other sources and drove revenue-per-visit up 254%.1 Salesforce, looking at the same season, estimated that AI agents and generative tools influenced more than 20% of all online retail sales globally.1 One in five online purchases, already passing through an AI recommendation. The discovery layer of commerce has moved, and it moved fast.
And then the checkout left your website
Discovery moving is survivable; you can still hope the agent sends the shopper to you. What changed the game is that the purchase stopped requiring your site at all. In late 2025, OpenAI and Stripe shipped Instant Checkout in ChatGPT and open-sourced the Agentic Commerce Protocol underneath it — a standard that lets a buyer complete a purchase inside the chat, starting with US Etsy sellers and extending to over a million Shopify merchants. A merchant already on Stripe can turn it on in about one line of code, and still owns acceptance, tax, fulfilment and returns.2
It is not one company's land-grab, either. Days earlier Google had announced the Agent Payments Protocol with sixty-plus partners — Mastercard, PayPal, American Express, Coinbase, Salesforce — using signed Intent, Cart and Payment mandates so an agent can pay across cards, bank rails and stablecoins, and has since donated it to the FIDO Alliance to keep it neutral.3 When the checkout button, the payment rails and the recommendation all live inside someone else's agent, the merchant's website is no longer where commerce happens. It is just one possible fulfilment endpoint.
So the catalogue becomes the product surface
Here is the reframe that follows. If a human shops your storefront, the storefront is the thing you polish. If an agent shops your catalogue, the catalogue is the thing you polish — because structured, complete, accurate product data is the raw material the agent reads, trusts, and transacts against. "Everywhere AI recommends products," which is how the GigaCommerce venture frames its whole reason to exist, is really a statement about legibility: you win by being the option an agent can confidently understand and buy, on every surface where the buying now happens.
This is the same discipline I wrote about for production agents, pointed at the demand side. Garbage catalogue in, wrong recommendation out — except now the wrong recommendation is a competitor's product the agent picked because its data was cleaner than yours.
What you actually have to build
An AI-ready catalogue. Not a prettier product page — a machine-legible feed: complete attributes, accurate availability, unambiguous variants, structured enough that an agent can answer "is this the right thing for me" without guessing. This is unglamorous data plumbing, and it is now the storefront.
Brand agents. Something that represents the brand on the agent's surface — answering the questions, carrying the positioning, defending the sale — because the shopper is no longer reading your carefully written PDP, they are asking an intermediary about you.
Checkout on the rails. Being enabled on the agentic-commerce and agent-payment protocols so the purchase can complete where the buyer already is, rather than bouncing them to a site they were never going to visit. GigaCommerce packages these as Brand Agents, Copilot Checkout and AI-ready catalogues for mid-market Shopify and Amazon merchants — the shops big enough to have real demand and too small to build this themselves.
What "AI-ready" actually means, concretely, is dull and decisive: the catalogue stops being a set of pages designed to persuade a human and becomes a clean, complete, structured feed an agent can reason over without guessing — every variant unambiguous, availability accurate to the minute, the attributes a shopper would ask about present as data rather than buried in marketing copy or trapped inside a product image. Most mid-market catalogues are not this. They are optimised for a person browsing, which means they are full of exactly the ambiguity an agent resolves by picking the competitor whose data was cleaner. The unglamorous work of making a catalogue legible — and then present, as a brand agent, on the surfaces where the buying now happens — is the whole job. It does not photograph well. It just decides whether the agent recommends you.
Why mid-market, and why now
The rails are being laid by giants — OpenAI, Stripe, Google, Shopify, Amazon. What they do not do is walk into a specific mid-market merchant and make that merchant's messy catalogue legible, its brand present on the agent surfaces, and its checkout wired to the new protocols. That is the same tail I keep returning to: too small for the platform owners to hand-hold, too consequential to leave un-served. And the timing is unusually sharp — the protocols standardised in 2025, the traffic inflected over one holiday, and the merchants who become agent-legible first are the ones the agents will keep choosing while everyone else is still A/B-testing a homepage nobody visits.
What this means if you are building one
Three sentences for anyone selling anything online. Treat your catalogue as an API for agents, because that is now literally what it is. Get onto the checkout and payment rails before the moment an agent has to choose between you and a competitor, because "not integrated" reads to the agent as "not available." And do not mistake a polished website for a healthy business — the site is no longer where the sale is won or lost.
The honest caveat, and it is a real one: this shift hands enormous leverage to whoever owns the agent. When discovery, recommendation and payment all sit inside ChatGPT or Google, the merchant risks being demoted to an interchangeable feed — a supplier the agent selects on price and data quality, stripped of the brand relationship the storefront used to protect. Getting agent-ready is table stakes for staying visible; it is not, by itself, a moat. The harder, longer question underneath all of this is how a brand keeps a direct relationship with a customer it now reaches through an intermediary that would quite like to own that customer itself.