Why does manual search fail as product choices become complex?

A formal economic model of agentic commerce (arXiv 2608.08395, econ.TH) compares manual search, which "accurately evaluates a limited product set," against agentic search, which "screens a broad catalog using noisy representations of preferences and products." Its first main finding: "as preference complexity grows, manual search eventually collapses: inspection ceases, mismatch returns to its no-search level, and platform revenue falls to zero." Agentic search "avoids a finite complexity collapse" — as complexity rises, "the marginal value of refinement" keeps increasing, so once refinement becomes worthwhile it "remains worthwhile at every higher level," and "mismatch remains below the no-search level and revenue remains positive."

The model places products and a consumer's latent "ideal point" in a k-dimensional characteristic space, where k is "preference complexity" — the number of satisfaction-relevant dimensions that are "difficult to articulate or query ex ante but readily evaluated through rich listing inspection" (a hotel's "room feel, neighborhood character, noise, view" versus a phone case's color and bulkiness). A finite attention budget forces the consumer to split effort between manual search intensity (how many products to inspect) and agentic search intensity (how deeply to refine the agent's representation of her preferences). Manual search can only ever widen the set inspected, so it hits a ceiling as k grows; refinement effort targets the dimensions driving mismatch directly, so its payoff keeps rising with complexity instead of exhausting itself.

This sharpens the scarcity question in Will agents compete for attention just like users do?: that note treats agent-mediated commerce as shifting competition from clicks to agent invocation, while this model specifies what the consumer must supply to make that invocation pay off — refined preference signal, not more browsing. It also parallels Can user preferences be learned from just ten questions?, where active-learning queries substitute for exhaustive preference elicitation; both treat the cost of articulating preferences, not catalog size, as the binding constraint on personalization.

The model is a theoretical abstraction — a Lancasterian characteristic space with Euclidean mismatch and no empirical calibration — comparing only two canonical regimes, pure manual browsing and pure agentic delegation, while the paper itself notes "shopping journeys are often hybrid." It does not measure real consumer behavior or test the collapse mechanism against purchase data; the adoption figures it cites (Adobe, Worldpay, Contentsquare surveys) describe current usage, not validation of the model. The implication the theory supports, at that strength, is that commerce platforms investing in agentic shopping should expect diminishing returns from expanding catalog scope and rising returns from tools that make preference articulation cheaper for the consumer.