A consumer wants organic wheat flour. It doesn’t open your app. He typed nothing into Google. You ask the AI assistant, it compares the options, and it drops one into the cart. He’s never seen your storefront, your merchandising, your carefully crafted brand page, or the promotion you spent all quarter planning. He saw the answer.
That’s agent marketing, and it’s reshaping the one thing every sales CMO has long thought it was theirs to maintain: ownership when the buyer decides.
For most of the last decade, your site and your app were your destination. He spent a small amount of money to drive cars there, and when the buyer arrived he controlled everything they saw. AI agents completely disrupt this interaction and experience. They sit between the consumer and the product, perform the search, and return recommendations. The seller becomes one option within the agent’s response rather than the buyer’s destination. In plain words, the agent is starting to do the job that he used to do in the shop front.
There is a word for the middle man being taken out of the transaction: disintermediation. It happened in travel agents, recording shops, bank branches. The uncomfortable part is that this time the person in the middle of the accident is you.
The pipes are already there
It would be easy to file this under “someday.” But the day has come. Infrastructure that allows an agent to make purchases on behalf of a customer is already being deployed.
At its Ship 2026 conference, Vercel went with a partnership with Shopify that built on this idea. Shopify is building what it calls Catalog, a product search API that aggregates billions of products from millions of sellers. It solves a really difficult problem in commercial data: taking the same shirt sold by four different sellers at different prices, in every color and size, and organizing it into clean, structured data that an agent can actually read. It is open, and it is not closed after login.
Let’s sit with that for a while. Any developer, any person really, who defines what they want in simple language, can now build a shopping experience on top of a universal product catalog. Vercel’s own demos included a run-club app that suggested gear based on the weather and nearby shops, and a voice assistant that found a kid’s game with just a conversation. Commerce ceases to be a place you visit. It becomes a built-in feature of apps that have nothing to do with shopping, powered by agents pulling from a shared catalog.
Here is the part the CMO should not miss. When an agent buys a global catalog, your products are competing on the quality of your data, not the quality of your website. A nice storefront sounds nice but it doesn’t really do anything for you. The server does it all.
Incomplete data is a withdrawal, not a downgrade
We’ve spent the last year stress-testing how agents actually choose products, and the findings are less favorable than most brands expect.
In general search, weak data costs your rating. You go from third place to eighth place, and you lose traffic quickly. Agent trading doesn’t work that way. The “small bag whole wheat flour” filter agent requires that feature to be present in your feed. If not, you are underrated. He was completely removed from the set. The agent doesn’t consider you, and you don’t get why.
The numbers we see prove this. Brands with a clean, enriched feed see about a 23% lift in AI-driven acquisitions. Push the catalog to almost a full attribute, which we call the Golden Record of about 99.9%, and visibility in AI recommendations can be four times higher than the basic, search-oriented feed. Go the other way and incorrect price or availability data can result in a 42% potential profit loss, because the agent who sees the old price immediately closes the product.
Reaching that gold record is never exciting, but it comes down to four things:
- Kill empty values. Agents filter incomplete records to avoid bad recommendations. Every empty feature is a door that closes itself.
- Write for purpose, not keywords. Semantic titles that correspond to the way the consumer defines the problem (“protein-rich,” “baking with children,” “very low calorie”) are matched more often than the product copy written in the search box in 2015.
- Keep price and availability accurate in real time. Agents are actively checking. One wrong price and you’re out, so accurate feeds require API-level updates, not nightly batch updates.
- Make the images and text compatible. Multimodal branding, where the image and words tell the same story, creates a reward for agents to agree on common availability.
None of this requires a moonshot. It requires managing your brand feed as a marketing asset with a named owner and budget, which is not how most organizations manage it today.
In the agency world, accuracy is how you earn trust
The Vercel session kept circling back to one idea, and it’s the one that’s worth taking with you. As the code approaches liberalization, the panel argued, the benefits are long-lasting to trust. Commerce has always been about trust: I buy from you because someone I trust has confirmed it. Shopify’s entire model works because it puts its trust in the sellers above it.
Extend that logic to agents and it gets sharper. When a person delegates a decision to an assistant, he trusts the assistant to choose the best option for him. Assistant has not tested any of these wheat flours, but knows to trust the data. A product that provides accurate, complete, consistent information becomes a safe recommendation. A brand that offers old prices and missing qualities becomes a liability that the agent learns to use. Brand trust was often built with consumers over the years of advertising. A piece of that sound is now created, or lost, by the machine that stands between you and the consumer.
Visibility of the financial agent as the right channel
The old model: eCommerce was at the bottom of the funnel, local demand was captured after your brand’s work had done its work at the top.
Agents suppress that. An AI assistant that answers “the best budget grocery delivery near me” or “what I need for a Sunday roast” shapes the decision long before the consumer gets to checkout. Winning that response drives demand for your search, your app, and your stores. So the visibility of the agent is not a good play of the lower funnel. It behaves like generating high funnel demand, and your measurement should eventually be able to see it in your hybrid model, not treat it like a rounding error.
The competition clock is already ticking. Walmart has Sparky. Target and Instacart are using their assistant experience, and many major retailers have helped shape new commercial deals for agents to work on. Some species were in the house. Others will spend the next two years reaching a level they had no input in writing. If your feed and agent strategy isn’t on the road map, you’re already behind the marketers who handled this as a channel last year.
What you should do with it, starting this quarter
You don’t need to be reorganized. You need to remove a few items from the “someday” list:
- Check your feed against agent readiness, not search. Perfection of result, quality of subject, integrity of category, consistency of image, and accuracy of price. Find out how many of your products can survive the agent filter today.
- Provide feed to owner and budget. Brand feeds with no accountability are one of the most common reasons brands are not visible to agents. Treat it like a storefront.
- Track how contributors respond to your category. Ask ChatGPT, Gemini, and sales assistants the questions your buyers ask. See where you are from, where your competitor is from, and when no one has won.
- Set a realistic goal and prove the lift. Identify a 98%-plus feed approval rate and A/B test attribute changes so you can put a number on the right enrichment.
The storefront is changing shape. In the next few years, the brands that win agent trade will be those whose product data an agent can trust when they see it. Getting there is tedious, but the task is doable, and it’s open to anyone willing to start now. Start early and you’ll be the agent’s go-to recommendation, while your competitors are still redesigning a homepage that will never open.
