Governed autonomy is not a product category. It is an engineering discipline. These papers establish the argument, define the architecture, and demonstrate the practice.
Three documents. The first argues why behavioral governance alone cannot scale — and what architecture must replace it. The second maps empirical claims to documented research. The constitutional specification — what valid implementation requires — follows in the section below.
Autonomous agents require a control plane that is separate from the acting system, deterministic in its authority, and continuously observable by a human principal. This control-plane pattern is not new. Distributed software systems have already developed much of it. The discipline has not yet been applied to autonomous AI agents. That is the gap.
Updated in v2.3: threat model boundary, knowledge substrate integrity, and Constraint composition semantics — with a tenth implementation criterion.
Read the Paper →“The question of command is not a behavioral question. It is an architectural one.”
“The agent acts. The system governs. The principal commands.”
A traceability document, not a literature review. Maps each v2.3 amendment to documented research — alignment faking, memory poisoning, policy composition — with explicit boundaries on what the evidence does and does not establish.
Companion to v2.3. Claim-traceability table: Documented, Precedent, Open.
Read the Evidence Annex →“Evidence flows upward; claims do not flow downward.”
Where the argument becomes law. The AgentVector Codex is the constitutional specification that every jurisdiction implements — the criteria by which any valid implementation can be evaluated.
The constitutional framework underlying every jurisdiction: composable Laws, language-specific enforcement kernels, and domain governance for governed autonomy. The Codex does not argue for governance. It specifies what governance requires — in terms precise enough to be evaluated, audited, and replayed.
Eleven Laws in four groups. Three enforcement kernels. Four reference domain jurisdictions. Each implementation is evaluable against the ten criteria defined in the Agency Paradox.
Essays making a single argument from different angles, written for platform engineers who know Kubernetes and are beginning to grapple with AI agents in production.
Forthcoming Kubernetes Is a Governance System · GitOps Is Operational Law · Audit as Architecture
The papers make the case. The field operations show the same Laws enforced in aviation, desktop agents, and authorship.