AI shopping agents usually belong to the store, not the shopper — why does almost nobody build one that works for you instead?

This explores why AI shopping agents mostly belong to the platform or retailer instead of the shopper, and what follows from that. The corpus doesn't count how many agents are platform-owned versus independent. What it does explain well is why platforms have strong reasons to keep agents in-house. The plainest reason is incentives. Whoever builds and maintains an agent sets its goals. MIT's 'Coasean singularity' work calls platform agents 'bowling-shoe agents': rented equipment that serves the house. By default they optimize for seller outcomes, while an agent the consumer owns could be tuned to the buyer's preferences Does agent ownership determine whose interests get optimized?. A formal model of agentic search sharpens the point. Platforms judge agentic search by conversion revenue alone, so it becomes worth deploying for them before it is worth adopting for shoppers, who also bear the costs of searching Why do consumers delay adopting agentic search?. Platforms get the agent built to their own timetable and their own goals.

The second reason is less obvious: open agents run into missing infrastructure. Amazon blocking Meta's Muse agent shows that retailers have no standard way to check who an outside agent is or whether a shopper actually authorized it. So they fall back on terms-of-service bans What identity standards do shopping agents actually need?. Payments add to the problem. Fraud systems were built to read human signals like typing rhythm and mouse movement, and those signals disappear when an agent checks out Does agent-initiated commerce break existing payment infrastructure?. Under every proposed protocol, the merchant still carries the fraud liability Does agentic commerce shift fraud liability onto merchants?. When you can't verify a stranger's agent and you'll pay for its mistakes, building your own agent is partly a way to manage risk, not only to grab control.

Third, closed agents rely on data lock-in. The UK's competition regulator (CMA) argues that an agent can't act well for a consumer without portable data. When data stays inside an incumbent's ecosystem, switching to a rival agent gets blocked Should data portability be infrastructure for agent-driven markets?. Rothschild and colleagues point out that technical standards alone won't open the market. Apple, Google, Meta and Microsoft are positioned to rebuild app-store-style gatekeeping at the layer where agents connect to each other Will agent economies develop as open webs or closed gardens?. Waller expects the market to split the way card networks did, between open systems that accept many agents and closed ones that require their own Will agentic commerce platforms adopt open or closed models?.

There is a demand-side twist. Shoppers say they want this arrangement, at least for now. In Accenture's survey, consumers preferred agents built by retailers. That is survey evidence, though, and doesn't show the agents work well at scale Will brands succeed by building their own AI shopping agents?. BCG expects consumers and networks, not merchants, to gain most as about $1.3 trillion of spending moves toward agents Who wins and loses as AI agents reshape shopping?. That gives brands a defensive reason to own the agent before someone else's agent becomes the gatekeeper.

The takeaway you might not expect: platform ownership may be a temporary result of missing standards, not a permanent structure. The open-agent stack is being assembled in layers. Visa's TAP and Mastercard's Web Bot Auth handle agent identity, Google's AP2 handles payment authorization, and protocols like OpenAI's ACP and Google's UCP handle checkout and discovery Do agentic commerce protocols compete or complement each other?. Independent agent wallets are already in habitual use for small payments, with time from funding to first spend falling from 73 days to 18 minutes Are AI agent wallets becoming genuinely operational instead of experimental?. If merchants get identity and liability standards they can trust, much of the risk-management case for closed agents goes away. What remains is the incentive case, and that is a question for competition policy, not for engineering.