BLOCKCHAIN IN THE AGENT ECONOMY: IDENTITY, AUTHORIZATION, AND MACHINE-TO-MACHINE SETTLEMENT

Abstract

Autonomous AI agents increasingly conduct economic transactions on behalf of individuals and organizations. This paper examines which payment architecture is most suitable for such transactions. Using a conceptual research design grounded in transaction cost economics, it separates agent based payments into three functions: identity, authorization, and settlement. It then compares delegated fiat, managed on chain, and native on chain architectures. The analysis shows that identity and authorization do not inherently require blockchain technology. Its distinctive contribution lies in enabling settlement through a shared ledger that is not controlled by a single institution. This feature is most valuable when multiple parties lack mutual trust and access to a common intermediary, when value can remain on chain, and when limited recourse is acceptable.A full cost comparison also challenges simple claims that blockchain settlement is inexpensive. Although public network fees may be low, total costs can be dominated by access, compliance, currency conversion, custody, liquidity and expected fraud losses. Evidence available through July 2026 remains promising but limited. Amazon Bedrock AgentCore Payments is still in preview, while x402 activity remains concentrated among a small number of facilitators and stablecoins. The most plausible institutional outcome is therefore the coexistence of conventional payment systems, managed blockchain services, native on chain settlement, and central bank digital currencies. The paper contributes a decision framework and research agenda rather than an empirical estimate of adoption.