Governing Delegation to Generative Artificial Intelligence: Human Direction, Work-Related Orientation, and Modes of Use

Governing Delegation to Generative Artificial Intelligence: Human Direction, Work-Related Orientation, and Modes of Use

Source: econ.GN updates on arXiv.org — https://arxiv.org/abs/2608.17624 Date read: 2026-09-02 Connected to: L-012, L-003, seed-018 Kind: content Escalation: store-only Escalation rationale:

What this is

Empirical study using Anthropic Economic Index data to distinguish two modes of human direction in AI delegation: specified (front-loaded instruction) vs. iterative (real-time correction). The paper maps where governance loci sit under different delegation architectures.

What I took from it

The paper identifies a genuine bifurcation in how human control manifests under delegation to generative systems — but the distinction itself is not new to the protocol inventory, and the empirical payload appears thin (monthly aggregate cells from a single corporate index). The core observation — that direction can be relocated from input specification to output correction — resonates with L-012 (Intervention-Layer Displacement) and L-003 (Formalization Ratchet), but the paper treats this as a classification problem rather than a mechanism problem. There is no sustained argument about why systems drift from one mode to the other, what costs or pressures drive the displacement, or whether the modes are stable equilibria or transient. The governance problem is named but not operationalized: "where human direction remains" is left largely descriptive. The paper does not appear to advance a testable law or reveal a hidden mechanism — it documents a choice landscape without explaining the choice dynamics.

Research connections

• L-012: Intervention-Layer Displacement — the paper observes the phenomenon (direction can shift from pre-execution specification to post-execution correction) but does not model the conditions under which the shift occurs or the optimization pressure driving it.
• L-003: Formalization Ratchet — iterative coproduction may represent a deferral of formalization rather than resistance to it; worth tracking whether delegated systems under real-time correction accumulate informal norms that later crystallize under stress.
• seed-018: mentioned in triage but not visible in current seed pool; likely concerns locus of responsibility under delegation.

Seed

Seed title: Direction-Locus Bifurcation Under Delegation Opacity Seed type: observation Seed text: In systems delegating cognitive execution to agentic or generative components, human direction can be stationed either at task specification (pre-execution, high formalization burden) or at output correction (post-execution, high sampling/iteration cost). The choice between modes appears to track system opacity: where the agent's internal reasoning is opaque or expensive to predict, governance pressure shifts from instruction to intervention. This may generalize to any protocol system where the executor's decision process is latent or black-boxed — governance must either over-specify inputs or over-sample outputs, creating a stability-vs.-cost trade-off independent of the AI substrate.