A shopper wants organic whole wheat flour. She doesn’t open your app. She doesn’t type anything into Google. She asks an AI assistant, it compares the options for her, and it drops one into the cart. She never saw your storefront, your merchandising, your carefully built brand page, or the promotion you spent all quarter planning. She saw an answer.
That’s agentic commerce, and it rearranges the one thing every retail CMO has long assumed was theirs to keep: ownership of the moment a shopper decides.
For most of the last decade, your site and your app were the destination. You spent a not insignificant amount of money to drive traffic there, and once a shopper arrived you controlled everything they saw. AI agents completely disrupt this interaction and experience. They sit between the shopper and the product, do the searching, and hand back a recommendation. The retailer becomes one option inside the agent’s answer rather than the place the shopper goes. In plain terms, the agent is starting to do the job your storefront used to do.
There’s a word for a middleman being pushed out of a transaction: disintermediation. It happened to travel agents, to record stores, to bank branches. The uncomfortable part is that this time the middleman at risk is you.
The plumbing already exists
It would be easy to file this under “someday.” But someday has arrived. The infrastructure that lets an agent shop on a customer’s behalf is already shipping.
At its 2026 Ship conference, Vercel walked through a partnership with Shopify built on exactly this idea. Shopify built what it calls the Catalog, a product search API covering billions of products across millions of merchants. It solves the genuinely hard problem in commerce data: taking the same shirt sold by four different retailers at four different prices, in every color and size, and organizing it into clean structured data an agent can actually read. It’s open, and it isn’t gated behind a login.
Let’s sit with that for a moment. Any developer, or really any person, describing what they want in plain language, can now build a shopping experience on top of a universal product catalog. Vercel’s own demos included a running-club app that suggests gear based on the weather and nearby stores, and a voice assistant that found a kids’ game by conversation alone. Commerce stops being a place you visit. It becomes a feature stitched into apps that have nothing to do with shopping, powered by agents pulling from a shared catalog.
Here’s the part a CMO should not miss. When an agent shops a universal catalog, your products compete on the quality of your data, not the quality of your website. A beautiful storefront feels great but in reality does nothing for you. The feed does everything.
Incomplete data is a disqualification, not a demotion
We’ve spent the last year pressure-testing how agents actually choose products, and the finding is blunter than most brands expect.
In traditional search, weak data costs you rank. You slip from position three to position eight, and you immediately lose traffic. Agentic commerce doesn’t work that way. An agent filtering for “small bag organic whole wheat flour” needs that attribute to exist in your feed. If it doesn’t, you aren’t ranked lower. You are removed from the set entirely. The agent never considers you, and you never find out why.
The numbers we’re seeing bear this out. Brands with clean, enriched feeds see roughly a 23% lift in eligibility for AI-driven discovery. Push a catalog to near-total attribute completion, what we call a Golden Record at around 99.9%, and visibility in AI recommendations can run up to four times higher than a basic, search-oriented feed. Go the other way and inaccurate price or availability data can leak up to 42% of potential revenue, because an agent that spots a stale price disqualifies the product on the spot.
Getting to that Golden Record isn’t exactly thrilling, but it comes down to four things:
- Kill the null values. Agents filter out incomplete records to avoid bad recommendations. Every empty attribute is a door you close on yourself.
- Write for intent, not keywords. Semantic titles that match how a shopper describes a problem (“protein-rich,” “for baking with kids,” “lowest calorie”) get matched far more often than product copy written for a 2015 search box.
- Keep price and availability true in real time. Agents check dynamically. One wrong price and you’re out, so accurate feeds need API-level refreshes, not overnight batch updates.
- Make the images and the text agree. Multimodal tagging, where the picture and the words tell the same story, builds the consistency agents reward with more frequent discovery.
None of that requires a moonshot. It requires treating your product feed as a marketing asset with a named owner and a budget, which is not how most organizations treat it today.
In an agentic world, accuracy is how you earn trust
The Vercel session kept circling back to one idea, and it’s the right one to carry with you. With code becoming close to free, the panel argued, the durable advantage is trust. Commerce has always been a relay of trust: I buy from you because someone I trust vouched for you. Shopify’s whole model works because it hands its trust to the merchants on top of it.
Extend that logic to agents and it gets sharp. When a person offloads the decision to an assistant, they are trusting the assistant to pick the best option for them. The assistant hasn’t actually tested any of these organic whole wheat flours, but it knows to trust the data. A brand that feeds it accurate, complete, consistent information becomes a safe recommendation. A brand that feeds it stale prices and missing attributes becomes a liability the agent learns to route around. Brand trust used to be built with shoppers through years of advertising. A meaningful slice of it now gets built, or lost, with the machine standing between you and the shopper.
Fund agent visibility like the channel it is
Old model: eCommerce was the bottom of the funnel, the place demand got captured after your brand work did its job higher up.
Agents compress that. An AI assistant answering “best budget grocery delivery near me” or “what do I need for a Sunday roast” is shaping the decision long before a shopper reaches a checkout. Winning that answer pulls demand into your search, your app, and your stores. So agent visibility isn’t a bottom-funnel efficiency play. It behaves like upper-funnel demand generation, and your measurement should eventually be able to see it in your mix model, not treat it as a rounding error.
The competitive clock is already running. Walmart has Sparky. Target and Instacart are running their own assistant experiences, and several of the largest retailers helped shape the new commerce protocols the agents will run on. Some brands were in the room. Others will spend the next two years catching up to a standard they had no hand in writing. If your feed and your agent strategy aren’t on a roadmap yet, you are already behind the retailers who treated this as a channel a year ago.
What to do about it, starting this quarter
You don’t need a reorganization. You need to move a few things off the “someday” list:
- Audit your feed against agent readiness, not search. Score completeness, title quality, category integrity, image consistency, and price accuracy. Find out how many of your products would survive an agent’s filter today.
- Give the feed an owner and a budget. A product feed with no accountable owner is the single most common reason brands are invisible to agents. Treat it as the storefront it has become.
- Track how the assistants answer your category. Ask ChatGPT, Gemini, and the retailer assistants the questions your shoppers ask. See where you show up, where a competitor shows up instead, and where nobody wins yet.
- Set a real target and prove the lift. Aim for a 98%-plus feed approval rate and A/B test attribute changes so you can put a number on what enrichment is worth.
The storefront is changing shape. Over the next few years, the brands that win agentic commerce will be the ones whose product data an agent can trust on sight. Getting there is tedious, but the work is winnable, and it’s open to anyone willing to start now. Start early and you’ll be the recommendation the agent reaches for, while your competitors are still redesigning a homepage it will never open.
