Shopping as an Operating System
Stuart Kearney, co-founder of Vetted, on AI agents in shopping
The core implication is that shopping looks less like generic search and more like a tightly managed operating system for choice, trust, and transaction. A shopper is not just asking for facts. They are comparing variants, checking price, delivery speed, returns, seller quality, and social proof, often across several turns. That is why broad search products can surface options, but a focused product can still win by making the whole path feel consistent and dependable.
-
Different product categories break in different ways. Fashion needs taste, sizing, and style context. Commodity purchases need fast comparison on pack size, materials, shipping, and total price. Vetted describes this as a routing problem, where the system has to know what kind of buying decision is happening before it can present the right interface and sources.
-
Google historically had reach, but Amazon won shopping behavior by making the experience predictable. Amazon trained users to expect familiar reviews, fast shipping, and clean fulfillment. Google often sent users into a patchwork of unknown merchants and changing layouts, which made product discovery feel less trustworthy even when selection was broad.
-
OpenAI has already added shopping recommendations to ChatGPT, and later added shopping research and direct checkout experiments. That confirms the opportunity, but it also shows how much infrastructure is required. The product has to keep price, availability, merchant data, and purchase flow accurate enough that one bad handoff does not break trust.
The market is heading toward a split. General assistants will become the default place to start, while specialized shopping systems will keep winning where the job is messy and high intent. Over time, the leaders will be the ones that combine conversational guidance with retailer grade reliability, so the assistant can move from advice to purchase without introducing doubt.