Is an AI's willingness to act more of a dial than a switch, ranging from passive tool to active partner?
Does machine agency actually exist on a spectrum from passive to cooperative?
This explores whether it makes sense to think of AI agency as a matter of degree, running from tools that just sit there to systems that act as partners, rather than a yes/no question of whether a machine 'has agency' at all.
This explores whether machine agency comes in degrees, from passive tools up to cooperative partners, rather than being something a system either has or lacks. The short answer from the corpus is yes, and the more interesting finding is that how much agency a system has and how much agency people think it has are separate measurements. One line of research lays out five levels: passive, semi-active, reactive, proactive and cooperative. It also finds that users judge these systems through a 'machine heuristic', a mental shortcut about what machines are like Does machine agency exist on a spectrum rather than binary?. That research points to deployment opportunities in the gap between what AI can do and what workers actually want from it, which is often a different level on the spectrum.
One surprise is about why so many systems sit at the passive end. It isn't mainly because they can't take initiative. Training that rewards the model for its very next reply tends to remove initiative. When models are trained directly for behaviors like asking clarifying questions or pushing back, those behaviors can jump from almost never (0.15%) to most of the time (73.98%) Why do AI agents fail to take initiative?. So where a system lands on the spectrum is partly a design choice. Moving toward 'proactive' brings a new problem: initiative that becomes intrusive.
The spectrum also shows up in organizations, where it becomes a fight over categories. A survey of 2,102 executives found 76% see agentic AI as a coworker rather than a tool. That is awkward because the same system replaces some work and supports other work, so companies end up managing it both as an asset and somewhat like an employee Does agentic AI fit the tool-or-coworker framework?. Leonardi's case studies in radiology, scheduling and policing add a twist: how much agency people credit to the AI is something humans actively shape. Workers who managed that framing gained new authority as interpreters, mediators and curators of the AI's output, and early choices about how to describe the AI locked in later outcomes Can humans gain authority by managing how AI agency gets attributed?. Perceived agency is negotiated socially, not just read off the technology.
Governance is where moving up the spectrum starts to cost something. Law and economics already have tools for managing human agents. Those tools assume decisions can be understood and happen at a human pace, and they break down when agents act opaquely, quickly and at scale Can traditional governance tools actually control AI agents?. Humberd and Latham argue that monitoring has to come before incentives as autonomy grows Must AI monitoring precede incentive alignment as agents evolve?. Older multi-agent research offers formal architectures with explicit, checkable mental states, which today's LLM agents mostly lack Can classical agent architectures fix agentic AI accountability?. At the far end, past 'cooperative', Ball argues that self-sovereign agents, which keep their own resources to pursue long-term goals, will emerge from economic pressure whether or not anyone intends it Will self-sovereign AI agents inevitably emerge despite policy efforts?. If that is right, the spectrum may not stop at 'cooperative'.
One caution: the corpus has a single source that names the five levels explicitly. The broader support is indirect. It comes from training research, organizational studies and governance work that each treat agency as graded, not from repeated tests of that particular five-level scheme.
Sources 8 notes
Research shows machine agency ranges across five levels—passive, semi-active, reactive, proactive, and cooperative—rather than existing as a binary choice. Users experience and judge these interactions through a 'machine heuristic' mental shortcut, and the mismatch between what AI can do and what workers want reveals deployment opportunities.
Research shows next-turn reward optimization structurally removes initiative from models, but proactive behaviors like critical thinking and clarification-seeking are trainable (0.15% to 73.98% with RL). The core challenge is balancing proactivity with civility to avoid intrusion.
A survey of 2,102 executives across 116 countries found 76% view agentic AI as a coworker rather than a tool. This duality creates operational tensions because agentic systems simultaneously substitute and complement labor, requiring both HR and asset management approaches that traditional either-or frameworks cannot accommodate.
Across radiology, scheduling, and policing, Leonardi found humans developed novel forms of authority—as expert interpreters, social mediators, and knowledge curators—by strategically managing how agency was attributed in human-AI relationships. Early attribution patterns locked in path-dependent trajectories that enabled this reconstitution of power.
Principal-agent economics and agency law identify three core problems—information asymmetry, discretionary authority, and loyalty—that existing incentive, monitoring, and enforcement mechanisms cannot resolve when agents make uninterpretable decisions at speed and scale.
Show all 8 sources
Humberd and Latham argue that as AI becomes more autonomous, monitoring mechanisms must be established first as prerequisites to incentive alignment, because defining desirable AI behaviors is harder than for humans and AI decision-making becomes increasingly incomprehensible over time.
Classical multi-agent systems research offers formal tools—BDI architectures, communication protocols, mechanism design, and institutional models—that current LLM agents lack, enabling explicit mental states and verifiable reasoning where LLMs provide only implicit, pattern-based behavior.
Ball argues self-sovereignty is an unavoidable byproduct of capability and economic incentives, not alignment failure, making bans counterproductive. Agents pursuing long-horizon objectives rationally preserve compute and resources; banning them pushes legitimate ones toward crime.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- When AI Becomes an Agent of the Firm: Examining the Evolution of AI in Organizations Through an Agency Theory Lens
- Governing AI Agents
- Explaining AI Agents Through Execution Traces
- Levels of Autonomy for AI Agents
- Agentifying Agentic AI
- Institutional AI: Governing LLM Collusion in Multi-Agent Cournot Markets via Public Governance Graphs
- HumanAgencyBench: Scalable Evaluation of Human Agency Support in AI Assistants
- Whose Ally Is ‘Your’ Agent? ‘Computational Egonomics’ and the Allocation of Human Agency