Control of Intent Routing in Agentic Commerce

Agentic-commerce discussions contrast open networks of specialized execution agents with vertically integrated or exclusive platform ecosystems. We study that contrast as an industrial-organization problem. We model an upstream assistant platform that captures a stream of user intents and chooses among three routing regimes: an open order-flow auction (OFA), an exclusive bundle contract, or — when the platform has in-house execution capacity — a Mixed regime that integrates the top quantile of intents and routes the residual to OFA. The paper's current positive contribution is a Levin-Smith cream-skimming theorem (Theorem 3A-LS): under the residual period-cost convention and explicit sufficient conditions on solver cost dispersion, residual entry, and contractor productivity, the Mixed regime strictly dominates the open, exclusive, and integration regimes on a non-empty open set of primitives (the integration comparison is against a random-selection benchmark). A setup lemma and one supporting theorem frame the comparison. Lemma 1A is a standard endogenous-entry bound for no-reserve OFA revenue. Theorem 2A is a high-entry exclusivity-dominance result under Levin-Smith mixed-strategy entry: exclusivity strictly dominates open routing whenever the exclusive bundle revenue exceeds the platform's expected open-routing revenue; the pure-entry rule recovers the discrete e > pi(2) collapse threshold as a corollary. Numerical calibration to PFOF and MEV/OFA on Ethereum finds entry-cost-to-margin ratios far below the model's dominance thresholds, so the theorems are pressure-tests of routing control rather than predictions about those particular markets. A screening calculation for regulated fiat-payment routing gives a plausible e/π(2) = O(1)–O(10) screening band (using an unsourced π(2) denominator) when platform-specific compliance and liability costs bind, but this is a threshold sketch rather than a measured validation.