The Sovereign Agency Matrix

Kendall Clark · Pentad Labs · 31 May 2026 · PLRN-000

Abstract

The value of an agent is not fixed by its speed. It is fixed by how far the human stands from the loop, and by the cost of verifying what the agent produces. We arrange agentic behavior on two axes, mode and persona, and read two laws off the arrangement. The first is a sovereignty measure that rises as neural flow control displaces the symbolic harness. The second is an inverse of Amdahl's Law: human oversight is the serial fraction, and it caps the whole. The two laws meet in a paradox at the corner of maximum autonomy.

1. Two axes

Agentic behavior resolves on two axes. Operational Mode fixes the human's proximity to the loop. Relational Persona fixes the agent's internalized expertise. The grid crosses them, and each cell names a working relationship together with the cost of verifying its output.

Mode	Intern	Contractor	Peer	Expert
HITL	Dictate (P)	Direct (P)	Jam (NP)	Shadow (IP)
HOTL	Audit (P)	Manage (NP)	Align (IP)	Govern (EXP)
HOOTL	Test (NP)	Utility (IP)	Proxy (EXP)	Oracle (EXP+)

The sovereignty level marks the point at which neural flow control $\Phi$ outpaces the symbolic harness $H$ . It is a calculated state, not a disposition.

$S = \frac{\Phi \cdot \Omega_{\mathrm{int}} \cdot \kappa}{\Gamma_H \cdot \Delta V}$

2. The law of agentic friction

Agent system efficiency is strictly bounded by the proportion of human oversight the work requires.

Let $h$ be the fraction of task time that requires human arbitration, and let $a$ be the speedup of the agent on the autonomous remainder. The system speedup is the reciprocal of the time the work still takes.

$S_{\max} = \frac{1}{\dfrac{1 - h}{a} + h} \;\xrightarrow{\;a \to \infty\;}\; \frac{1}{h}$

A tenth of the work held by a human caps the whole at tenfold, and the most powerful machine available changes nothing. Scaling past the ceiling means reducing $h$ itself.

2.1 Operational modes

The three modes differ only in where the human's time goes relative to the task:

HITL   human in the loop     Δt_human ∝ Δt_task
HOTL   human on the loop     Δt_human ≪ Δt_task
HOOTL  human out of the loop Δt_human → 0

The sovereignty level is computed, not vibed:

pub fn sovereignty(phi: Float, omega: Float, kappa: Float, gamma: Float, dv: Float) -> Float {
  // neural flow over the cost of arbitration and verification
  case gamma *. dv {
    0.0 -> infinity
    denom -> phi *. omega *. kappa /. denom
  }
}

Agent. Flow control determined by a neural network.
Harness. Flow control determined by symbolic logic.
Human. The biological anchor, supplying the objective function.

3. The observer's paradox

Drive $h$ toward zero and efficiency rises to its maximum in the Oracle corner. Verification cost rises with it, to EXP and past it, until the ability to confirm the outcome falls to nothing. The most value an agent can return is had only by surrendering the most control, and at the point of maximum value the correctness of the agent can no longer be proved. This is the terminal state of the matrix, and it is a transition from management to faith.