When an AI tool sits between you and your organization, who actually gets the credit or blame for what it does?
What types of professional authority emerge when humans mediate between organizational systems and AI?
This explores what new kinds of workplace power and expertise people gain when they sit between an organization and the AI systems it uses: interpreting, supervising, translating, or vouching for what the AI does.
This explores what new kinds of workplace power and expertise people gain when they stand between an organization and its AI systems. The surprising answer from the corpus is that this authority comes mostly from controlling the story of who did what, and less from technical skill. In studies of radiology, scheduling and policing, Leonardi found people taking on three new roles: expert interpreters who explain what the AI's output means, social mediators who manage relationships around the system, and knowledge curators who decide what the AI gets fed and what gets trusted. The common thread is that they gained standing by managing how much credit or blame the AI got for decisions (Can humans gain authority by managing how AI agency gets attributed?). The patterns of credit set early on tended to lock in, so whoever shaped them first gained a lasting advantage.
This authority isn't permanent. Leonardi's related 'agency loops' model shows that when people credit AI with independent judgment, they feel less in control and less responsible, even while still holding formal authority. That loss then moves through a cycle in which humans can step back in and reorganize their role (Does attributing agency to AI actually reduce human sense of control?). So mediating authority is something people keep winning back, not something they hold once and keep. A useful idea from formal models of organizations explains why the go-between position matters: in MOISE+, having authority over a role assumes you have a channel to communicate with it, and having that channel assumes the parties know each other (How do authority, communication, and acquaintance relate in organizations?). In organizations using AI, the human mediator often *is* that channel, which is a structural reason the role carries power.
Why can't the AI simply hold this authority itself? One answer is that expertise is granted by a community of peers, based on membership and a track record. Individual accuracy isn't enough, and AI can't take part in that kind of validation (Can AI ever gain expert community trust through participation?). Someone still has to vouch for the output, and the person who does becomes the credentialed face of it. Agency law and economics point to the same gap from the other direction. Information asymmetry, unchecked discretion and loyalty are problems that existing oversight tools can't solve for fast, opaque agents (Can traditional governance tools actually control AI agents?). Human mediators fill that gap.
Consulting frameworks are now trying to formalize these roles. Deloitte describes 'handoff supervisors' placed at deliberately designed decision points, supported by audit logs and digital identity systems (How should organizations supervise autonomous AI agents?). McKinsey puts humans 'above the loop', setting strategy for squads of agents, while admitting that governance practices haven't caught up with deployment (Can AI agents truly operate above the loop at enterprise scale?). Puranam treats the whole arrangement as a question of organization design: who decides what, and how each side learns (How should humans and AI systems divide decision-making tasks?). One complication: agentic systems can now hand tasks *back* to people, so delegation runs in both directions (Can AI systems delegate tasks back to humans?). Sometimes the mediator is being directed by the system rather than directing it.
The part you might not expect to care about is a hidden risk in these roles. The skill that gives mediators authority, which is shaping who gets credit, can also mislead the mediators themselves. The 'LLM Fallacy' describes people mistaking AI output for their own ability. This is a self-perception error, separate from hallucination and from over-trusting the machine (How does AI-assisted work reshape how people see their own abilities?). The interpreter whose authority depends on blurring the line between their own contribution and the machine's is the person most at risk of losing track of that line.
Sources 10 notes
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.
Leonardi shows that when humans credit AI with independent judgment, they experience reduced sense of control and responsibility—even while retaining formal authority. This displacement circulates through five phases: delegation, attribution, contingency, reassertion, and reconfiguration, making it strategically manageable rather than permanent.
MOISE+ formalizes a one-way hierarchy where authority over a role presupposes a communication channel to that role, which in turn presupposes agents can represent each other. The model enforces this at specification time, not runtime.
Expertise is validated through social participation and track record within expert communities, not individual accuracy alone. AI cannot enter this validation circle because it lacks social embeddedness, testable judgment history, and ability to participate in the consensus-building processes that define expert paradigms.
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 10 sources
Deloitte proposes that managing autonomous AI agents requires humans stationed at intentionally designed decision points, supported by digital identity systems, cryptographic receipts, and immutable action logs—translating user control preferences into organizational governance infrastructure rather than interface features.
McKinsey describes AI-first workflows organized around outcome-aligned agent squads with humans providing strategic oversight, citing bank case studies showing efficiency gains. However, industry governance lags deployment ambition, and agent interactions create risks beyond individual component reliability.
Puranam frames human-AI pairs as multi-agent organizations requiring design choices about task division and learning configuration. He distinguishes three task types and shows learning mechanisms produce different performance outcomes based on feedback structure and decision architecture.
Holldack, Banh and Strobel argue that agentic information systems capable of independent decision-making can transfer rights and responsibilities to humans, reversing the traditional one-directional delegation model. This requires moving beyond agency-primacy frameworks to delegation theory that accounts for bidirectional task assignment.
Research shows the LLM Fallacy operates through misattribution of AI outputs to personal capability, independent of output accuracy or reliance behavior. It requires interventions that clarify human-machine contribution boundaries, not just better system accuracy or forced verification.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Assistant or Actor? Student Trust, Control, and Delegation Regret When Using a General-Purpose AI Agent
- Homo agenticus in the age of agentic AI: Agency loops, power displacement, and the circulation of responsibility
- Explaining AI Agents Through Execution Traces
- When AI Becomes an Agent of the Firm: Examining the Evolution of AI in Organizations Through an Agency Theory Lens
- Governing AI Agents
- Institutional AI: Governing LLM Collusion in Multi-Agent Cournot Markets via Public Governance Graphs
- The Oversight Fallacy: Why AI Agents Require More than Humans-in-the-Loop
- Agentic information systems