# Who Gets to Do the Finding Agentic commerce is no longer theoretical. It is being written into the infrastructure layer of the internet right now, and most people in e-commerce are not asking the right questions about it. The [Universal Commerce Protocol](https://ucp.dev/) is an open standard that lets AI agents complete purchases on your behalf, across any retailer, without you ever leaving the AI's interface. You ask Gemini to find you a jacket, it finds one, and the transaction completes inside Google's UI. That is the whole thing. The framing you will encounter everywhere is that this is a friction-reduction story. Lower cart abandonment, better conversion, seamless checkout. That framing is technically accurate but strategically incomplete. **The real question UCP forces every platform to answer is not "how do I integrate?" It is "what business am I actually in?"** Disclosure that I am taking a fashion-specific lens here. I have spent years working in the Shop team at ABOUT YOU, one of the largest & fastest growing fashion e-commerce platforms in Europe. A lot of what I am going to argue is specific to discovery-first commerce, and does not necessarily apply to a platform that sells commodities at scale. Keep that in mind. ## The DoorDash Problem [Nilay Patel](https://www.theverge.com/podcast/823909/the-doordash-problem-ai-agents-web-amazon-perplexity-lawsuit), one of my favorite tech journalists, has a frame for this he calls the 'DoorDash problem'. When restaurants joined DoorDash, they did it for reach. More customers, more orders. What they did not fully account for was that DoorDash would absorb the customer relationship in the process. The restaurant became a kitchen. DoorDash became the brand the customer was loyal to. The platform that was supposed to be a distribution channel became the destination, and the restaurant was left optimizing for someone else's growth. UCP has the same dynamic, except it is happening at the protocol level rather than the application layer. The spec is open source. Apache 2.0. Anyone can build against it. But today, there is exactly one live consumer surface running UCP, and it is Google's - AI Mode in Search and the Gemini app. So while the documentation describes a vendor-agnostic standard, the reality of integrating with UCP today is that you are selling inside Google's UI. **The protocol is open. The gravity well is not.** Zalando, one of the largest fashion platforms in Europe with genuinely strong product instincts (and recent owner of ABOUT YOU), has also signed up to integrate UCP. I am honestly not sure why, or whether they have a clear answer to the question I am about to raise. My best read is that they are betting the volume upside of Google's surface outweighs the strategic cost. That might be the right bet. But I think it deserves more scrutiny than it is getting. ## The Session Is the Business Model Most users do not arrive at ABOUT YOU with a purchase decision already made. They are not coming to execute a transaction. They are coming to figure out what they want. **They are building a look, not procuring a product.** And that distinction has enormous consequences for how you should think about UCP. The entire commercial architecture of fashion ecom is built around that exploratory session. Sponsored placements only generate value because the user is actively browsing and open to influence. Complementary recommendations work because the user's consideration set is still open. Editorial content, curated assortment and brand storytelling are not decorative. They are where the margin comes from. When an AI agent routes around all of that to reach checkout, it is not removing friction. **It is removing the business model.** This is why the DoorDash analogy actually undersells the risk for fashion specifically. A restaurant session has no monetization layer beyond the order value. Fashion ecom does, and it only works because the platform owns the session. ## Not All GMV Is Created Equal Intellectual honesty requires making a distinction here. Not all of a fashion platform's GMV comes from that exploratory session state. A good portion of transactions come from users who have already decided. They know they want a Nike Air Force 1 in size 40 and white color. For that user, the session adds nothing. The platform was probably already losing them to Google Shopping, price comparison sites or direct brand search anyway. UCP does not create that leak. It just makes it more efficient. **The question every platform / product leadership team should be asking is: what percentage of our GMV is transactional intent versus discovery intent?** That ratio is your actual exposure (or risk) to UCP. And it should be the first number you reach for before deciding if you want to integrate UCP. ## The Potential Upside Here is where the analysis gets more interesting. For protocols like MCP, which lets AI agents interact with external tools and services, genuine ecosystem adoption happened because multiple competing surfaces emerged. The protocol did not stay captured inside one company's interface. Any agent, any surface, any developer can build on it. If UCP follows that same path, the picture changes completely for fashion platforms. But there is a real chance it does not. Unlike MCP, UCP launched inside Google's existing distribution infrastructure, with Google's Shopping Graph powering discovery and Google Pay handling payments. The gravity well around Google is already much stronger from day one. Competing surfaces can technically exist. Whether the market actually produces them, against that kind of structural headstart, is a genuinely open question. If the competition does emerge, the strategic move is not to be the best-stocked inventory node. **The race becomes building the best taste layer.** A fashion platform could build a stylist agent that understands personal aesthetic preferences at a granular level, assembles complete looks from products across the entire connected merchant ecosystem, and executes multi-merchant transactions in one session. The moat shifts from "we have the products" to "we understand your style better than anyone." Some users would like ABOUT YOU's fashion stylist better, others would like Zalando's and someone else still Google's. Companies with years of fashion-specific behavioral data, strong recommendation infrastructure, and real brand trust with their users are not poorly positioned for that world. They might actually be the best positioned for it. Time will tell if that will happen, but if it does this industry is headed for maybe the biggest fundamental shift in decades. ## The Etsy Puzzle One thing I keep returning to: Etsy signed up to integrate UCP as well. Etsy's whole value proposition is the serendipity of finding something you did not know you were looking for, from a maker you would never have found through a normal search. That is almost the opposite of what a transactional AI agent is built for. Either Etsy believes that a good LLM can replicate that open-ended discovery better than keyword search ever could, which is genuinely plausible, or they are making a category error about what their product actually is. I do not know which one it is. And I think that ambiguity is **the most honest signal we have about how uncertain the outcomes here still are.** ## So What Do You Do? The platforms that will hurt the most are the ones that integrate without asking what they are giving up. The ones who see "reduced cart abandonment" and stop thinking there. The platforms that could win are the ones who look at UCP and see an opportunity that the consensus analysis is missing entirely. Everyone is asking how to be found. The better question is who gets to do the finding. The session is not the friction. **The session is the asset.**