Electronic commerce has evolved through four paradigms over three decades—B2B, B2C, C2C, and O2O—each designed around humans clicking buttons as the primary mode of interaction. The emergence of large language model (LLM) agents capable of autonomous planning, negotiation, and execution creates a qualitative shift that the existing paradigms do not accommodate. When the buyer, the seller, or both are represented by autonomous agents acting on behalf of human principals, the design assumptions of human-centric interfaces, trust heuristics, and economic mechanisms no longer hold. This paper formalizes A2A Commerce (Agent-to-Agent Commerce) as a distinct fifth paradigm of electronic commerce. We make four contributions: (1) a formal definition grounded in three essential properties—Agent-Native, Bilateral Autonomy, and Delegated Authority—that jointly distinguish A2A Commerce from AI-assisted variants of existing paradigms; (2) the AI Commerce Maturity Ladder (ACML), a six-level framework (L0–L5) for classifying the agent-mediation depth of any commerce system along five autonomy dimensions; (3) seven core design principles and a reference transaction lifecycle for implementers, with identity and trust discussed at the level of abstractions; and (4) an explicit interoperability map between A2A Commerce primitives and existing agent communication protocols, including Google's A2A Protocol, the Model Context Protocol (MCP), and conventional web APIs. We argue that the paradigm shift is inevitable in markets where agent-mediated interactions become dominant, and that establishing shared vocabulary and normative structure now reduces fragmentation and accelerates ecosystem formation. Companion specification repository: https://github.com/agent-show/a2a-commerce-spec
Xiang Gao (Thu,) studied this question.