When a platform you do not use moves at the protocol level, you still have to read the move carefully. Shopify’s agentic commerce rollout is one of those moves. Most e-commerce directors I speak to have filed it under “AI storefront noise” and gone back to their roadmaps. That is a mistake, because the mechanics are not noisy. They are deliberate, and they rewrite the terms of discovery for every merchant on every other platform, including WooCommerce.
Two things pulled me into this story. The figures Shopify’s president, Harley Finkelstein, has put on earnings calls this year are large as multipliers and small as absolute revenue, and the gap between those two readings is where the honest conversation lives. And the mechanism Shopify shipped is an open protocol, not a product, which means a WooCommerce merchant has every right to adopt the same shape without touching Shopify at all.
This is not a vendor beating a drum. It is a vendor publishing its own operating thesis and then publishing figures that both support it and quietly undercut it. If you are weighing whether agentic commerce deserves a budget line in 2027, the Shopify arc is the file to read.
Shopify Is the Brand, Not the Store
Shopify’s position is unusual. It is not a retailer, it is the platform behind a long tail of them, which means every bet it places on buyer behaviour is a bet on how it thinks a shopper will choose to buy in two years’ time. On 11 January 2026 the company announced the Universal Commerce Protocol, an open standard it co-developed with Google, alongside two product moves: Agentic Storefronts, which makes a Shopify catalogue natively addressable from ChatGPT, Microsoft Copilot, Google AI Mode and Gemini, and an Agentic plan, which opens Shopify Catalog to brands that do not run their online store on Shopify at all. Vanessa Lee, Vice President of Product at Shopify, framed it as a continuation of the company’s checkout history: “Shopify has a history of building checkouts for millions of unique retail businesses.”
That last line is the one that tells you what Shopify thinks it is doing. It is not selling an AI feature to the merchants it already has. It is trying to be the merchandising layer behind every AI shopping surface, whether the brand on the other end pays Shopify a monthly subscription or not. The Agentic plan is explicit about that: list products once in Shopify Catalog, and they appear in the AI channels Shopify has signed up, including surfaces it does not control.
The Scrape-and-Hope Baseline Was Nobody’s Idea of Progress
The uncomfortable before-state is easy to describe because every store that is not on Shopify is still in it. An AI agent asked to buy a jacket on behalf of a user has, until very recently, two options. It can scrape the brand’s public site, hope the price it reads is current, hope the colour it picked is in stock, and hand a URL back to the user to finish the job themselves. Or it can try to drive a browser session on the user’s behalf and resolve the checkout the way a human would, with all the authentication, consent, promo code and variant selection that implies.
Both options were, in Finkelstein’s framing to Modern Retail’s Allison Smith on 22 January 2026, dead ends. “Merchandising is a real thing,” he told the briefing. “You can’t just scrape information and expect a great experience.” He is right. A merchandising decision is a judgement about which variant to recommend to which customer in which market under which promotion, and the data to make that judgement is not on the product page. It is in the merchant’s own systems.
UCP Was the Only Move That Would Actually Scale
What Shopify changed, and why that and not something else, matters more than any single feature in the announcement. The company could have shipped a proprietary agent connector, charged for it, locked it to Shopify merchants and let the market pay a Shopify tax to be reachable by AI. It did the opposite. UCP speaks REST, Model Context Protocol, Agent Payments Protocol and Agent2Agent, works with any payment processor, and handles discount codes, loyalty credentials, subscription cadences and final-sale terms as structured fields rather than text an agent has to parse off a page.
Ashish Gupta, Vice President and General Manager of Merchant Shopping at Google, put the case for an open protocol plainly in the joint announcement: “The shift to agentic commerce will require a shared language across the ecosystem.” Nayna Sheth, Head of Product for Agentic Payments at Microsoft, added the demand-side framing: “Today’s shoppers expect to go from search to purchase in a single conversation.” Two things to take from that pairing. First, this is not a Shopify-only standard, because Google and Microsoft had to be in the room for it to work. Second, the merchants listed as live on the day the announcement went out were a mix of Shopify (Monos, KEEN, Pura Vida) and brands coming to the Agentic plan without a Shopify online store (Gymshark, Everlane are listed as coming soon).
Victor Tam, CEO and Co-Founder of Monos, kept the merchant case grounded in a single sentence: “It enables us to meet customers where they already are.” That is the whole argument for not fighting the medium. If your buyer is in Gemini asking for a weekender bag that fits a cabin dimension, you want to be the answer, not the thirteenth tab they open after the answer.
Big Multipliers on a Small Base
The measured results are the part most coverage either exaggerates or dismisses. Both readings miss the shape. On the January 2026 Modern Retail briefing Finkelstein cited Shopify data showing that “merchants have seen 14x more orders coming from agents compared to last year at the same time,” reading the twelve months from January 2025 to January 2026. Six months later, on the Q2 2026 earnings call reported by Digital Commerce 360 on 6 August 2026, he reported that AI-driven traffic and orders had both tripled year over year in Q2 alone.
Four more figures from the same Q2 report are worth putting on a slide: new buyer orders from AI channels were coming in at nearly twice the rate of other channels; AI searches powered by Shopify Catalog converted at twice the rate of searches running off scraped data; roughly half of AI-referred sessions landed directly on a product detail page, 2.5 times the rate for traditional search; and 75 percent of AI-attributable orders in Q2 came from outside Shopify’s top 100 product categories. The last of those is the most interesting line in the file, because it says the agents are routing demand to the long tail, not the headline SKUs.
Here is the honest part. On the same call Finkelstein also said, cleanly for the record, that “the volume from agentic commerce is still small relative to our massive GMV, the growth trends are impressive.” That is a quote worth printing next to the multipliers rather than instead of them. Three times a small number is a small number. Fourteen times a small number is a small number. What those multipliers tell you is the direction of travel, not the current revenue line, and a mid-market board that treats them as either is going to make the wrong call.
What Shopify Did Not Solve, and What That Means for You
Every honest case study has an unsolved column. Shopify’s has at least three worth naming before anyone commits budget.
First, attribution. When a user asks Gemini for a cabin bag, picks one from a UCP response, and the purchase posts to the merchant’s checkout, the referrer chain the merchant’s analytics sees is at best fragile and at worst empty. Shopify’s own numbers are “AI-attributable” with a definition the company controls. On a WooCommerce estate running GA4, the equivalent is not instrumented out of the box. The question of what counts as an AI order, measured by whom, with what referrer header, is still open.
Second, the long-tail skew is a feature from a Shopify-president’s seat and a problem from a brand-director’s. 75 percent of AI-attributable orders landing outside the top 100 categories is good news if you sell niche cabin bags; it is a warning if you sit at the top of a mainstream category, because it tells you the agent surfaces are not yet moving your biggest line. In the beauty, apparel and home categories where paid search still rules, agents are routing around you, not through you.
Third, who pays for the agent call, and who gets paid for the demand. Agent Payments Protocol handles the money, but the economics of referral revenue (does the AI surface take a cut, is there a sponsored placement in-chat, does the merchant discount for the agent) are not written down anywhere a procurement team can plan against. The quiet guidance I give clients is that the next twelve months are for infrastructure, not margin modelling. Build the pipe. The invoice follows.
What a WooCommerce Build Section Looks Like at Protocol Parity
For a mid-market WooCommerce estate that wants to be reachable the way Shopify Catalog is reachable, the work splits into five pieces, in roughly this order.
One, a clean product schema layer. The agent cannot read your merchandising metadata if it is not there. The baseline is schema.org Product with Offer, PriceSpecification, availability, GTIN or MPN, and a complete variation axis with image and price per variation. The audit is unglamorous and takes a sprint on a store with two hundred SKUs, longer on a catalogue with real complexity. If your product descriptions still carry copy that reads as marketing rather than attributes, that is where the work starts.
Two, a public, agent-readable product endpoint. WooCommerce ships a REST API, but the standard product endpoint is authenticated and shaped for store admin, not for an agent. The practical build is a thin read-only MCP server or UCP-compatible endpoint that reads from WooCommerce’s own tables and emits the fields an agent needs: canonical URL, structured attributes, inventory availability, promotional terms, loyalty eligibility, final-sale status. Allow two to three sprints on a store with the schema already clean, more if the schema work is still in progress.
Three, a prepared-cart checkout handoff. Agents that cannot complete the checkout themselves want to hand a user a URL that drops them into a cart with the right variant, quantity, discount code and shipping option pre-selected. WooCommerce handles that with the Store API and a few lines of a cart URL handler. Build it once, document the parameters, test it against the two or three agent surfaces you care about. A sprint.
Four, agent-aware analytics. GA4 does not know what to do with a ChatGPT, Perplexity, Gemini or Copilot referral out of the box. The fix is a few hours of channel group configuration (custom source and medium rules), a dimension for the agent that posted the order, and a quarterly review of what the referral strings actually look like. Without this step your board will see your Shopify competitor publishing multipliers and will see nothing comparable in your own analytics even if you are earning the orders.
Five, merchandising policy as structured data. Discount codes, loyalty tiers, subscription cadences and promotional rules all need to live as structured post meta on the products and the orders they attach to, not in human-readable banners the agent will mis-interpret. This is the part that benefits most from a WordPress-native build, because the custom post meta plus REST exposure layer is already there. For a brand we would scope this at four to six weeks on top of the three sprints above.
Total protocol-parity build for a mid-market store, honestly scoped: eight to twelve weeks of work plus a quarter of measurement before any of it tells you anything useful. That is not a weekend plug-in and it is not a boil-the-ocean replatform. It is the shape of work that WooCommerce is actually good at, because the pieces live in the same database and the same cron.
What I Would Tell a Board Looking at This Today
Read the Shopify numbers twice. Once as the growth story, which is real, and once as the base-rate reminder, which the company’s own president keeps putting on the record. Treating “we are on WooCommerce, not Shopify” as a reason to opt out is the wrong read: UCP, MCP and Agent Payments Protocol do not care which store software sits behind the endpoint. Spend the next two quarters on the five pieces of the build above, measured against a baseline you capture now, and resist the temptation to pay for a managed agent connector before the standards settle. This is the kind of groundwork that is worth doing calmly rather than on a 2027 panic cycle, and it is the sort of work we are often handed when the market mood turns. If you want a second pair of eyes on what that looks like on your estate, you know where to find us.
Last modified: October 30, 2026
