2026 Reports

AI Power at Tempo: Conversion, Harnesses, and Anticipatory Influence, v. 3

Tumin, Zachary; Edelmann, Rasmus

[NB: This is the third in a series of issues papers developed in connection with AI: The New Face of Power, a book in progress under contract with Columbia University Press (2027). The series circulates selected ideas, cases, and frameworks as they are being developed and tested, and invites discussion. These papers are not excerpts from the book nor substitutes for its fuller argument and evidence.]

As artificial intelligence moves from tool to infrastructure, the locus of organizational power shifts. Power no longer resides primarily in decisions made at the moment of choice. It resides as well in the systems that determine what is seen, what is prioritized, where work flows, and what happens by default before any formal decision is made.

This paper argues that AI constitutes a fourth face of power — pre-emptive power — exercised through anticipatory influence before any individual deliberates, alongside the three classical faces identified by Dahl, Bachrach and Baratz, and Lukes. Pre-emptive power operates through the harness: the socio-technical architecture surrounding and governing the conversion stack through which institutional purpose becomes machine-executable action. The paper identifies three interdependent harness domains — operational, institutional, and structural — each a site of managerial decision and potential failure, and introduces several new concepts to characterize their dynamics: replacement drift, building on Wu's mission drift; the AI cliff; encoded constraints; and constrained discretion. Together these give the framework analytical precision and practical application.

Harnesses are not passive architectures. Left ungoverned they default and degrade. A weak harness is one where constraints are underspecified, misaligned, or ungoverned. A strong harness is one where purpose is clearly encoded, authority clearly assigned, dependencies clearly understood, and learning loops active. Moore's strategic triangle, Christensen's value networks, Meadows' leverage points, Davenport's process transformation work, Simons' levers of control, and Mueller's ethics as architecture together give managers the frameworks to assess and strengthen harness readiness.

The framework is tested through eight cases spanning customer service, policing, social work, healthcare, transportation safety, software engineering, and geopolitical competition. The paper concludes with five managerial imperatives and a division of labor between leaders, who authorize and bound the system, and managers, who make it operate well.

Keywords: AI, power, pre-emptive power, anticipatory influence, encoded constraints, constrained discretion, harness, conversion stack, management, governance, replacement drift, AI cliff, intent engineering, mission drift, Moore's strategic triangle, Meadows' leverage points, Davenport, Simons's levers of control, Mueller, ethics as architecture