Agentic commerce’s dirty little secret.

And every one of these in one way or another expresses an urgency to get merchants ready: adopt the protocols, publish the feeds, make your catalog legible to ChatGPT and Gemini and the rest. All of it is real and necessary. But almost no one is asking the uncomfortable question underneath it.

When discovery moves from the storefront to the agent, who actually wins? Because by default, it’s not the merchant.

𝗪𝗵𝗼 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝗰𝗼𝗺𝗺𝗲𝗿𝗰𝗲 𝗿𝗲𝗮𝗹𝗹𝘆 𝗯𝗲𝗻𝗲𝗳𝗶𝘁𝘀

Pretty sure the ACP (Agentic Commerce Police) are going to be knocking on my door after this one.

Follow a single agentic purchase. The shopper no longer starts on a brand’s site. They start with an AI assistant, describe what they want, and let it find, compare, and increasingly transact on their behalf. Two parties are structurally advantaged by that shift. The first is the AI platform, which now owns the top of the funnel, the shopper relationship, and the ad and placement economics forming on top of it. The second is the large marketplace, which gives the agent exactly what it wants: an enormous, uniformly structured, always-available catalog it can trust and pull from in one call. Scale and structure are what an agent rewards, and marketplaces are built out of both. This is an inevitable and completely logical migration for marketplaces, commerce platforms and providers.

However, an individual merchant enters that system as one source among millions. Even fully agentified, armed with every protocol and a clean outbound feed, a single store is a needle competing against a haystack with billions of products for a slot in an answer whose ranking logic it does not control and cannot see.

𝗧𝗵𝗲 𝗻𝗲𝗲𝗱𝗹𝗲 𝗶𝗻 𝗮 𝗯𝗶𝗹𝗹𝗶𝗼𝗻-𝗽𝗿𝗼𝗱𝘂𝗰𝘁 𝗵𝗮𝘆𝘀𝘁𝗮𝗰𝗸

Picture the actual query. Someone asks ChatGPT for “a nice, knee-length, belted, lightweight trench coat for summer business wear in Seattle.” That is a rich, specific, high-intent request, exactly the kind agents are good at. Perfect. So what happens next? The agent scans a massive pool (millions) of candidate products and ranks them by relevance, structured-attribute match, availability, authority, price, and whatever commercial arrangements sit in the background, then returns a short list. How many of those candidates are individual DTC catalogs, and how many are marketplace listings?

For a store like Cassie’s Kurations to surface in that answer, its data has to be not just clean but better-matched than everything else in the pool, and the ranking still has to choose to show that result instead of a marketplace row selling something close enough. The odds are not zero. They are also not something to build a business on.

That is the part no one says out loud. Agent-side discovery is a lottery whose house is the platform and whose most reliable winners are the aggregators, who always get a cut of each transaction.

𝗕𝗲𝗶𝗻𝗴 𝗮𝗴𝗲𝗻𝘁-𝗿𝗲𝗮𝗱𝘆 𝗶𝘀 𝗻𝗲𝗰𝗲𝘀𝘀𝗮𝗿𝘆, 𝗻𝗼𝘁 𝘀𝘂𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁

None of this is an argument against agentic commerce. Every merchant should have clean, enriched product data and universal agentic feeds. If your catalog is not legible to agents, you are not in the lottery at all, and staying out is not a strategy. Legibility is table stakes.

But let’s at least be honest about what that ticket buys. It gives the merchant a chance to be surfaced by someone else’s algorithm, into someone else’s interface, often as a listing inside someone else’s marketplace, with someone else’s take rate attached. Play only that game and you have volunteered to have your discovery bled off the top and handed downstream, to the platforms first and to you second, if you are lucky. The brand you built becomes a drop ship supplier.

𝗣𝗿𝗼𝘁𝗲𝗰𝘁 𝘁𝗵𝗲 𝗰𝗮𝘀𝘁𝗹𝗲 𝘆𝗼𝘂 𝗯𝘂𝗶𝗹𝘁

The counter-move is not to opt out of agents. It is to run a multi-pronged discovery plan and keep as much discovery as possible, on ground you own. 𝗧𝗵𝗲𝗿𝗲 𝗶𝘀 𝗲𝘅𝗮𝗰𝘁𝗹𝘆 𝗼𝗻𝗲 𝗽𝗹𝗮𝗰𝗲 𝘄𝗵𝗲𝗿𝗲 𝘁𝗵𝗲 𝗮𝗻𝘀𝘄𝗲𝗿 𝘁𝗼 𝗮 𝘀𝗵𝗼𝗽𝗽𝗲𝗿’𝘀 𝗾𝘂𝗲𝗿𝘆 𝗶𝘀 𝗴𝘂𝗮𝗿𝗮𝗻𝘁𝗲𝗲𝗱 𝘁𝗼 𝗯𝗲 𝗼𝗻𝗲 𝗼𝗳 𝘆𝗼𝘂𝗿 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝘀: 𝘆𝗼𝘂𝗿 𝗼𝘄𝗻 𝘀𝘁𝗼𝗿𝗲𝗳𝗿𝗼𝗻𝘁. Onsite, you are not competing against billions of items. You are matching a shopper’s intent against your catalog, and every result is a sale you keep in full, with the margin, the data, and the customer relationship intact.

So the plan has two prongs, not one. Be legible offsite so agents can find you. And win onsite so that every shopper who arrives, whether from an agent, an ad, a social post, or straight to your domain, meets a storefront that can actually answer them instead of a keyword box that sends them back to the marketplace to finish the job.

𝗪𝗵𝗮𝘁 𝗼𝗻𝘀𝗶𝘁𝗲 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝘀

Meeting an AI-shaped shopper on your own site takes more than a search bar. It takes the same describe-show-ask experience they just had with the agent, running against your inventory. That is the onsite side of what we build at visualAI, and it rests on a few capabilities working together.

The foundation is data quality and enrichment: accurate attributes, consistent taxonomy, complete descriptions. Clean data is what makes onsite search return the right products, and it is the same asset that makes you legible offsite, so the work pays twice.

On top of it sits trimodal discovery: natural language, precise color, and image similarity in one flow. A shopper who types the Seattle trench-coat request, drops in an inspiration photo, or specifies an exact color shade gets matched to what you actually carry, in seconds, with no dead-end “no results” page. That is where onsite conversion is won or lost, because intent that is not met on arrival leaks straight back out.

Alongside it, virtual try-on lets a shopper see the piece on themselves before they commit. In high-consideration categories that raises confidence at the moment of decision, which lifts conversion, improves satisfaction, and cuts the fit-and-sizing returns that quietly erode apparel margin.

Because these run as one integrated engine rather than a stack of stitched-together point tools, the technical burden on the merchant stays low, the checkout path stays clean and accurate, and the experience is consistent from first query to cart. The shopper gets an answer, on your site, that has the best chance of resulting in the sale of one of your products.

𝗥𝗲𝗮𝗱𝗶𝗻𝗲𝘀𝘀 𝗶𝘀 𝘁𝗵𝗲 𝗳𝗹𝗼𝗼𝗿, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝘆

Agentic commerce is coming whether any single merchant likes it or not, and being ready for it is not optional. To be clear, we love it. It is one of the most important advances in product discovery in twenty years, which is exactly why 𝘃𝗶𝘀𝘂𝗮𝗹𝗔𝗜 𝗯𝘂𝗶𝗹𝗱𝘀 𝘀𝘂𝗽𝗽𝗼𝗿𝘁 𝗳𝗼𝗿 𝗲𝘃𝗲𝗿𝘆 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝗽𝗿𝗼𝘁𝗼𝗰𝗼𝗹 directly into our APIs, so your catalog is ready for whatever surface an agent reaches for next. But readiness offsite is a floor, not a growth strategy. The merchants who come out of this era ahead are the ones who show up wherever an agent might look and still refuse to hand their storefront’s core job to anyone else. Feed the agents, all in. But keep control of the door to the castle.