The Maze: Agentic commerce is not one new shelf. It is four different shelves, each with a different rent. Ankit Minocha's framework splits the market by two questions: does the brand own a native app inside the agent surface, and is the shopper asking with direct intent or still exploring? The uncomfortable answer for most brands is simple. They will not build a custom app inside ChatGPT next quarter. Their first agentic-commerce move is less glamorous: make the catalog good enough for machines to trust.
Native apps are the premium shelf, but they are expensive real estate. The direct-intent lane puts Walmart, Target, Instacart, DoorDash, Uber Eats, and Swiggy in the hardest box. A shopper can ask for a specific item and expect the agent to close the job. That requires more than a nice storefront. It needs real-time inventory, fulfillment confidence, and a custom in-chat transaction path. OpenAI's commerce docs frame the same plumbing problem: ACP helps ChatGPT ingest structured catalog data, understand inventory, and surface products in context. If the promise is "delivered tomorrow," the agent needs confidence before the user ever sees a recommendation.
Discovery apps are category media before they are checkout rails. Sephora and ASOS sit in the native-app plus exploration lane because the shopper is forming intent. The job is not "buy this SKU." It is "help me choose." That changes the economic logic. The brand wins by shaping the category conversation, teaching the agent what matters, and pushing the user back to owned checkout when the decision is ready. This is useful for categories where expertise is defensible. Beauty can do it. Fashion can do it. A commodity SKU with weak differentiation has a harder time pretending it is a destination.
The sleeper lane is discovery feed plus direct intent. Home Depot, Lowe's, and Best Buy appear where shoppers ask for specific attributes: voltage, motor type, price, compatibility, availability. There is no native app advantage in the source framework. The agent reads the feed and matches the request. OpenAI's product spec is blunt about what that means: structured feeds include price, availability, variants, dimensions, weight, images, shipping, and product attributes. In old ecommerce, a weak attribute table hurt filtering. In agentic commerce, it can make a product functionally invisible.
Browse-driven products need richer context, not just more copy. Nordstrom, Wayfair, and Etsy sit in the open-exploration feed lane. The query is softer: a gift for a child, a room style, an occasion, a budget, a taste profile. That world rewards descriptions, taxonomy, use-case tags, and variant-level accuracy. OpenAI's best-practices page pushes the same discipline: factual descriptions, stable optional fields, variant-specific data, and durable seller links. The agent is not browsing like a human. It is compressing metadata into a shortlist.
The real sequencing question is delegation posture. The strongest comment under the post adds the missing operator lens: product type suggests the right lane, but consumer willingness to delegate decides whether the investment compounds. Direct-intent shoppers are ready to hand off execution. Exploration shoppers are still forming the brief. Brands should not start with "we need an AI app." They should start with "what decisions do customers trust an agent to make for this category?"
Why it matters: Retailers spent two decades optimizing pixels: hero banners, product tiles, filters, reviews, checkout buttons. Agentic commerce moves the bottleneck upstream. The new merchandising layer is the data that tells an agent what a product is, when it is available, who it fits, and why it should be recommended. The winners will not be the brands with the loudest product pages. They will be the brands whose catalogs can answer a messy human request without making the agent squint.
