Position paper · v2.2 · March 2026

The Agency Paradox:
Governed Autonomy as Infrastructure

The central question, restated

The original framing asked: who is in command when autonomous systems act? The answer is not a more disciplined human. The answer is a better-designed system.

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 pattern is not new. It has not yet been applied to autonomous AI agents. That is the gap.

What cloud-native systems already solved

Four components: declarative desired state, continuous reconciliation, policy enforcement at the boundary, and an observable audit trail. These four together produce a governed system — one that acts autonomously within declared constraints, enforces its own boundaries, and produces continuous evidence that governance is occurring.

The DevOps parallel

If DevOps unified software creation and operation under one discipline, governed autonomy seeks to unify AI reasoning and operational authority under one control-plane model. The connection — a substrate-independent, constitutional control plane for autonomous agents, composition-aware by design — is the discipline waiting to be named.

The composition problem

Action A is permitted. Action B is permitted. Action C likewise. But the sequence of A, B, and C together constitutes scope creep that no single evaluation would have caught. The architectural response is composition-aware governance: session-level state, composition tracing, and pattern detection for boundary-probe signals.

Nine properties of a valid implementation

01 Separation of governance from execution — the agent cannot modify, bypass, or disable the governance layer
02 Determinism — same action, same policy, same session state → same verdict, always
03 Fail-closed default — when governance state is uncertain, the default is denial
04 Compositional evidence — full evaluation trace, not just the verdict
05 Composition-aware evaluation — session-level state as input to deterministic evaluation
06 Substrate independence — same authority model regardless of where the agent runs
07 Principal observability — full operational picture reconstructable from governance evidence alone
08 Knowledge substrate currency — staleness is a governance failure, not a minor degradation
09 Bounded execution surface — an explicit, finite action surface governance can evaluate
The agent acts. The system governs. The principal commands.
ClawLaw — the implementation ↗ ← All Intelligence