The Agent-Native Enterprise Operating System (AE-OS): An Open Framework for Next-Generation Enterprise Resource Planning

Enterprise information systems were designed for people entering transactions. Autonomous software agents can perform much of that work, but the systems they operate in cannot establish what authority an agent holds, why it acted or whether its record can be trusted. This paper asks what architecture would let an enterprise information system have most of its digital operational work performed by agents while retaining the control and auditability that organizations, auditors and regulators require. Its thesis is that this is possible if identity, business semantics, policy enforcement, transaction integrity and evidence are provided once, as a common foundation beneath every agent. Following design science research, the paper defines agent-native systems and proposes the Agent-Native Enterprise Operating System (AE-OS), a reference architecture with enterprise resource planning (ERP) as its first application. It contributes a delegation model that carries authority from a person's or organization's credential to an agent's action, a causal evidence ledger issuing a receipt for each agent action, and accounting invariants mapped to SOC 1 control objectives. The ledger, receipt and capability-attenuation primitives are demonstrated in Python and Rust reference implementations that agree with each other, with proof sketches for tamper evidence and append-only consistency. The architecture has not been evaluated in a working organization. The paper sets out the evaluation that would confirm or refute the thesis there, including pilot measures and an illustrative cost model, and lists open research questions.