The Agentic Shelf: a measurement framework for autonomous AI shopping journeys

A new commerce surface is forming. We call it the Agentic Shelf: the surface on which autonomous AI buyers — frontier-model shopping agents, retailer-owned assistants, and third-party orchestration tools — select products on a consumer’s behalf, with little or no human intervention. Within twelve months the Agentic Shelf will be a measurable distribution channel in major consumer categories. Within thirty-six months it will be the dominant first-touch surface for AI-mediated discovery in beauty, electronics and grocery. Today no major brand can answer the simple question: how do we perform on the Agentic Shelf? No major retailer building its own shopping agent can answer the parallel question: does our agent perform correctly for our customers? No frontier-model platform exposing a shopping experience can defensibly demonstrate that its agent honours the constraints its users state. This paper makes four claims, with measurement evidence behind each. 1 Agentic shopping is not a more efficient version of human shopping. It is a structurally different journey, with different failure modes, different decision drivers, and different commercial implications. Tools built to measure the digital shelf do not measure the Agentic Shelf. 2 Brands, retailers, and platforms each have a distinct blind spot. Brands cannot see what agents recommend. Retailers building agents cannot independently verify what their agents do for customers. Platforms operate the agents but cannot demonstrate that their agents honour customer constraints. Each blind spot is commercially material. 3 Persona and intent are not flavour text; they are first-class drivers of outcome. Identical query, different persona produces different recommendations. Identical persona, different intent produces materially different journey performance. A measurement framework that ignores either is not measuring the Agentic Shelf. 4 Constraint retention — the proportion of customer-stated constraints the agent actually honours through to selection — is the most under-discussed failure mode in agentic commerce. It is also the most consequential. A high recommendation score with a low retention score is a brand promise broken silently