From Agent Detection to Alignment Target

Once an agent is defined by variables rather than appearance, degrees and scales of agency become measurable — and detection becomes the first step toward naming an alignment target.

What decision changes?

When evaluating a system for alignment, first establish its degree and scale of agency using the variable-based definition, before asking whether it is aligned.

A colder, variable-based definition of agency does not produce a single yes-or-no answer. It produces degrees: how strongly a candidate boundary predicts and controls future interaction, how stable that boundary is under perturbation, and at what scale the boundary sits — a single process, a team, a firm, a market.

This matters because alignment work has to know not just that something is an agent, but which scale of agent is doing the work that matters. A firm and an AI service can both pass the same agency test while requiring very different alignment targets.

Agent detection is therefore the hinge between the colder definition and everything downstream: it is the step that turns “is this thing an agent” into “what, specifically, needs to be aligned.”