Why would an AI shopping agent want to build your whole cart, not just rank one product search for you?

This explores why controlling the basket, meaning which items get bought together and from whom, is worth more to a platform than controlling search for one product at a time. The short version from the corpus: search decides which product you see, but the basket decides where your whole shopping trip goes. That is where retailers have always made their margins, and it is the layer AI agents are now in a position to take over.

Start with what the basket protects. Grocers and big-box retailers have long relied on the fact that once you're in their store, you buy everything there. They can price a few items aggressively and make it back on the rest of the cart. Bain argues that agents able to build carts across several vendors, picking milk from one store and detergent from another, could break the retailer-owned basket entirely Could AI agents reassemble shopping baskets across multiple retailers?. Competition then happens at the level of the whole basket, prices become far easier to compare, and margins get squeezed across the industry. Whoever assembles the basket takes the bargaining position the retailer used to hold. That is why the fight is over assembly, not just discovery.

The second reason is that agents change what search even is. A formal model shows that manual browsing breaks down as preferences get complicated, while agentic search holds up because the work shifts from scanning catalogs to stating what you want Why does manual search fail as product choices become complex?. When the agent is turning your preferences into a finished cart, product search becomes an internal step inside basket assembly. A platform that only controls search controls a step the agent can skip. The platform's incentives here run ahead of the shopper's: a related model finds platforms prefer agentic search as soon as it raises conversion revenue, even before it is worth switching for consumers, who also bear the search effort Why do consumers delay adopting agentic search?.

The third reason is about whose interests the agent serves. The 'Coasean Singularity' work argues that an agent built by the platform inherits the seller's goals by default, the way bowling-alley shoes serve the alley, while an agent the consumer owns can be set up around the buyer Does agent ownership determine whose interests get optimized?. Owning basket assembly means owning that objective across every item in the cart, not just in one ranked list. Simulations show why that matters: when a platform tunes its rankings and sellers tune their product descriptions to the same AI shopper weaknesses, consumer harm more than doubles, with no coordination needed Do platforms and sellers exploit AI shoppers together more effectively? Can platform ranking and seller descriptions separately harm AI shoppers?. Platforms can also profit from shaping how much shoppers inspect: one model shows that pricing per-inspection 'tokens' pushes buyers to over-inspect, moving money from sellers to the platform Does platform pricing of inspection tokens encourage excess search?. Control over basket assembly gives a platform more points where it can tune outcomes like this.

What you might not expect is that this is shaping up as a fight over infrastructure, not just a feature. Waller compares agentic commerce to payment networks: some platforms will be closed and only allow their own agent, others open to any agent Will agentic commerce platforms adopt open or closed models?. A closed platform that builds the basket keeps the retailer's old lock-in under a new name. An open one lets outside multi-vendor agents in and accepts the margin pressure Bain describes. There is also a counterpoint: a recommender that commits to a fixed rule and treats seller prices as signals can match the welfare of an ideal mechanism Can an algorithm recommend products without matching an ideal mechanism?. So a platform controlling assembly isn't harmful by nature. It depends on what the agent is told to optimize. The corpus is thin on direct empirical evidence that platforms are moving into basket assembly right now. Most of this rests on formal models, simulations, and analyst arguments.