What actually keeps you locked into one AI shopping agent — and what would make switching to another one easy?

This explores what would actually keep you stuck with one AI agent platform today, and what would make leaving it easy or hard. The short version is that the collection mostly argues about switching costs as a *future design choice*. It doesn't measure what consumers face right now. Still, it points to three places where lock-in can build up: your data, your preferences, and which merchants and services an agent can reach.

Data is the most concrete one. The UK competition regulator treats data portability as basic market plumbing, not a nice extra. An agent that can't take your purchase history, saved preferences and account links with it can't act well on your behalf. Without portability, switching agents means starting from zero, and the incumbent wins by default Should data portability be infrastructure for agent-driven markets?. A subtler cost sits on top of that. Transactional agents don't just store what you like. Through repeated back-and-forth they help *shape* what you like, so your preferences form in a loop between you and the agent Do AI shopping agents reshape what consumers actually want?. If that's right, switching means more than exporting a file. You'd be leaving a relationship that has been co-tuning your tastes, and that history doesn't fit neatly in a data export.

The second cost is reach: which agents are allowed to shop where. Waller expects agentic commerce to split the way payment networks did. Some systems will be open to any agent and others will require their own proprietary one Will agentic commerce platforms adopt open or closed models?. Rothschild and colleagues make a similar case. Technical standards let agents talk to each other, but the platforms that control those connections (Apple, Google, Meta, Microsoft) could recreate app-store-style gatekeeping Will agent economies develop as open webs or closed gardens?. In a closed world, switching agents could mean losing access to certain stores or services altogether. The protocols in play today (ACP, AP2, UCP, Visa's TAP, Mastercard's Web Bot Auth) work more like layers that merchants stack together than like rival camps Do agentic commerce protocols compete or complement each other?. That's mildly good news for portability, because a consumer's agent isn't obviously tied to a single protocol.

There's a counterweight. The same Microsoft researchers argue that agents could *lower* switching costs overall. Agents cut the friction of comparing, asking and negotiating, which makes it easier to move between *businesses* Could agent-to-agent communication reshape market power?. The less obvious point is that agents can make it cheap to switch stores while making it expensive to switch agents. The lock-in moves up a level, from the retailer to the agent platform. That fits with predictions that services will compete for the agent's attention rather than yours Will agents compete for attention just like users do?. It also fits with consumers telling Accenture they'd prefer retailer-built agents Will brands succeed by building their own AI shopping agents?. If that preference holds, you'd end up juggling several agents, one per store, which is fragmentation more than lock-in.

One more angle is worth knowing. A formal model shows that platforms benefit from agentic search before consumers rationally should. Consumers carry the cost of learning and trusting a new way to search, and platforms don't Why do consumers delay adopting agentic search?. That learning cost is itself a switching cost, and it applies to every new agent you try. What the collection doesn't have is evidence of what real consumers pay to switch today: no studies of export tools, account migration or retention. The honest answer is that lock-in is still being designed, and the open-versus-closed choices being made now will decide how high these costs get.