Abstract
Abstract
Artificial intelligence agents now search, evaluate, negotiate, and pay on behalf of consumers, while firms deploy agents that create content, personalize offers, and complete transactions. Agentic commerce therefore weakens a foundational assumption of marketing thought and marketing measurement: that the recipient of marketing activity is a human being. The evidence is scattered across marketing, information systems, computer science, economics, and law, and it has a largely forgotten first act in the shopbot research of the late 1990s. Following the SPAR-4-SLR protocol with PRISMA 2020 reporting, we screen 5,157 records from Scopus, Web of Science, and EBSCO Business Source and add verified snowballing to assemble 140 primary studies spanning 1994 to 2026. Each study is coded on influence direction, customer journey stage, delegation depth, method, and research wave. The synthesis yields an actor taxonomy with a five-level delegation ladder, an influence triad that separates marketing to, through, and by agents, an integrated account of trust and governance infrastructure, and a market-level analysis connecting both research waves. The Agentic Commerce Reference Framework consolidates this evidence into four auditable layers with explicit cross-layer dynamics, a ten-proposition research agenda, and five measurement instruments for analytics practice. A complete replication package accompanies the article.