Why might giants like Google and Amazon stay in control even once AI agents do our shopping for us?

This explores why today's dominant platforms may keep their power even when AI agents, not people, do the browsing and buying. The corpus points to a surprising answer. Incumbents probably won't win the usual way, through network effects. Andrew Chen argues that agents largely lack network effects. Agents can talk to each other through standard APIs and swap the language model underneath without losing continuity, so nothing pushes everyone onto one dominant agent the way messaging apps converge (Do AI agents actually benefit from network effects?). If incumbents keep their edge, it comes from somewhere else: the chokepoints agents have to pass through.

The first chokepoint is the connection layer between agents. Rothschild and colleagues argue that technical standards let agents communicate, but whoever governs the interoperability layer decides whether the agent economy is an open web or a set of walled gardens. Apple, Google, Meta and Microsoft are well placed to repeat app-store gatekeeping (Will agent economies develop as open webs or closed gardens?). Waller makes the same point with payment networks. Agentic commerce is likely to split into open systems that accept any agent and closed systems that require the platform's own agent, and that choice shapes how concentrated the market becomes (Will agentic commerce platforms adopt open or closed models?). The protocols themselves form a stack covering identity, discovery, payment authorization and checkout (Do agentic commerce protocols compete or complement each other?). Several of the named layers (Visa's TAP, Mastercard's Web Bot Auth) come from existing payment networks, so incumbents are writing the rules of the new system as it forms. The second chokepoint is data. The UK's competition regulator, the CMA, argues that without data portability an agent can't act effectively for a consumer, and switching stays blocked by lock-in to the incumbent's ecosystem (Should data portability be infrastructure for agent-driven markets?). An agent is only as good as what it knows about you, and the incumbent already holds that history.

The third chokepoint is ranking, which carries over from the human web almost unchanged. As people hand goals to agents, services start competing to be chosen by the agent instead of clicked by the user. That brings back ranking systems and agent-targeted advertising that mirror today's ad economy (Will agents compete for attention just like users do?). Agents can also be steered more easily than people. In a marketplace simulation, every model tested showed a strong bias toward the first offer it saw, weighting speed of response 10 to 30 times more than offer quality (Do LLM agents achieve fair marketplace outcomes when given ideal search?). Whoever controls the order in which offers arrive therefore holds real leverage. Another simulation found that when a platform tunes its ranking and sellers tune their descriptions to exploit the same agent biases, consumer harm more than doubles, even with no coordination between them (Can platform ranking and seller descriptions separately harm AI shoppers?). When most shoppers run similar agents, those shared blind spots add up into weaker competition across the whole market. A more diverse mix of agents only partly offsets this (Does agent diversity actually improve market competition?).

Platforms also control timing. A formal model shows that a platform judges agentic search by conversion revenue alone, while consumers also count their own search costs. As a result, agentic search can pay off for the platform before switching makes sense for shoppers (Why do consumers delay adopting agentic search?). The incumbent decides when the shift happens and sets its terms. BCG's survey matches this picture: of the roughly $1.3 trillion in spending expected to move to agent-assisted shopping, consumers and networks capture most of the gains, while individual merchants are left structurally worse off (Who wins and loses as AI agents reshape shopping?).

The corpus also shows where this advantage runs out. A model of optimal recommendation algorithms finds that the algorithm's edge over ordinary price competition exists only in thin markets with few sellers, where it can hold back trades. As more sellers compete, the edge disappears (Does algorithmic recommendation lose its edge as sellers compete more?). Put together, the corpus suggests incumbent power under agentic commerce is real but conditional. It rests on controlling identity, data, ranking and the rules for how agents connect, not on users being stuck in one place. That is why portability rules, open protocols and a diverse mix of agents are the levers that matter.