Two Economies, Not One- A Taxonomy of the Agentic Economy and the Case for Settlement Neutrality

The term agentic economy appears in over fty distinct denitions published between
2021 and March 2026 by Microsoft Research, Sequoia Capital, the Bank for International
Settlements, Coinbase, Stripe, the World Economic Forum, Gartner, Fetch.ai, and at least
forty other sources. They do not mean the same thing. Some describe an AI agent buying
a plane ticket for its owner through Visa. Others describe autonomous software agents
negotiating, transacting, and building reputation with each other no human in the loop,
no credit card, no chargeback.
This paper surveys the denitional landscape, classies fty-plus usages into ve cate-
gories, and identies two fault lines that the literature has not made explicit. The rst
the primary fault line separates commerce for humans (agents as intermediaries in human
economic activity) from an economy of agents (agents as independent economic actors). The
second within the autonomous category separates the eleven approaches that require
blockchain infrastructure from the one that does not.
The taxonomy reveals a structural question: what formal properties must a settlement
layer possess to support autonomous agent-to-agent commerce, independent of the commu-
nication protocol and independent of the ledger substrate? We dene settlement neutrality
as that set of properties escrow determinism, ledger consistency, reputation equivalence,
dispute parity, and auditability and describe a reference architecture where agents speak-
ing Anthropic's MCP, Google's A2A, and direct REST converge on the same escrow pipeline.
The paper proposes the ve categories and six classication dimensions as a reusable frame-
work for evaluating future protocols, and identies open questions for the eld.