This working paper examines the institutional and infrastructural conditions required for safe AI-mediated transaction-initiation in high-value, regulated markets. As AI systems increasingly participate in discovery, evaluation, and negotiation on behalf of human principals, a new bottleneck emerges: authorized action. The paper introduces Agent Action Infrastructure, a four-layer architecture extending from verified representation through computational eligibility, governance, and safe bounded action initiation. The main contribution is the Action Boundary Object (ABO): a machine-readable economic record that encodes not only what an asset is, but which actions AI agents may perform on it, under which mandates, with which verification requirements, and with which audit obligations. The paper formalizes actionability as an infrastructure condition beyond visibility, eligibility, and agent-readiness. It introduces the Agent Actionability Index (AAI), the Action Signal Quality (ASQ) metric, and the concept of Action-Derived Demand Signals. It also discusses Action Gatekeeping, Action Sovereignty, Transactional Sovereignty, and the relationship between Agent Action Infrastructure and emerging agentic commerce protocols. HomeSelf is used as an implementation case for early-stage actionability in real estate markets. It is presented as a bounded, non-binding coordination and representation infrastructure, not as evidence that autonomous real estate transactions are presently feasible or desirable.
Marco Patrone (Fri,) studied this question.