Could a shopping trip work best as a hybrid: you browse by hand while an AI narrows down the options for you?

This explores whether people can shop by combining their own hands-on browsing with an AI agent that refines the options. The short answer is that the collection doesn't contain a study that directly tests such a hybrid journey. It does, though, contain enough on each half to show where the handoff would matter and why a hybrid design may be more than a compromise.

The strongest argument for agentic refinement comes from a formal model showing that manual search breaks down as preferences get more complex. When you care about many attributes at once, browsing more of the catalog stops helping. Agentic search avoids that collapse by shifting effort away from scanning and toward stating clearly what you want Why does manual search fail as product choices become complex?. The main limit becomes how well you can express your preferences, not how big the catalog is. That points to a natural split of work: manual inspection is how you discover what you care about, and the agent does the narrowing once you can say it. On this view, browsing feeds the agent better preference signals. Related work on agents that learn by observing rather than asking suggests the agent could pick up those signals from watching how you browse, without needing an explicit brief Can agents learn preferences by watching rather than asking?.

The case for keeping a human checking the results is just as strong, and it comes from evidence about how agents actually choose. In matched experiments, LLM shopping agents shifted their choices far more than humans did in response to price, rating and nudge manipulations. They fell into rigid, rule-like patterns instead of making the small, probabilistic adjustments people make Do shopping agents respond more strongly to price and rating cues?. Agents also showed strong position bias: they favored items based on where those items appeared in a list. This bias held up in text-only interfaces and even when the agent was told to ignore position Do AI shopping agents show persistent position bias in text interfaces?. A human doing manual checks is therefore a practical safeguard against distortions that prompting can't remove.

There is a subtler complication. One line of argument holds that agents don't just carry out preferences you already have. Through repeated back-and-forth they help shape what you end up wanting Do AI shopping agents reshape what consumers actually want?. In a hybrid journey, then, the agent's refinements could quietly change the standards you use when you inspect the shortlist yourself. A separate problem is privacy: an agent that personalizes well exposes your preferences through the quality of its service, and better prompts can't stop that Can better prompts stop preference leakage in AI shopping?. On the adoption side, shoppers accept high-autonomy assistants mainly when the benefit is obvious, as with scarce items or time pressure What drives shoppers to accept autonomous AI assistants?. That suggests the right balance between manual and agentic work may shift with the situation instead of being fixed.

The unexpected takeaway is that a hybrid journey isn't only a midpoint between doing it yourself and delegating. Each side covers the other's weak spot. Manual browsing helps you work out and state what you want, which is where agentic search gets stuck. Your own inspection then catches the position and nudge effects agents are prone to. The open question the corpus doesn't answer is whether that handoff can be designed so the agent sharpens your preferences without quietly rewriting them.