From Copilots to Coworkers: How Autonomous AI Agents Reshape Organizational Roles and Decision Authority: A Structured Integrative Review and Research Agenda

Abstract

Autonomous artificial intelligence agents extend generative AI from producing content on demand to pursuing goals through planning, memory, tool use, environmental feedback, and iterative action. This shift matters organizationally because agency is no longer confined to recommendation: systems can increasingly initiate and coordinate consequential work. This article integrates research on AI-enabled work, algorithmic management, organizational design, human–AI collaboration, automation, and responsible AI to explain the transition from copilots to digital coworkers. The review distinguishes five configurations—tool, copilot, teammate, delegate, and coworker—and argues that the central organizational variable is not technical capability alone but the allocation of decision authority. We synthesize evidence showing that AI assistance can improve speed and quality in bounded tasks, often benefiting less experienced workers, while also producing uneven performance at the frontier of model capability. Extending this evidence to agentic systems, we develop an authority allocation matrix based on task risk, reversibility, verifiability, and auditability. We identify three mechanisms through which agents reshape organizations: role reconfiguration, authority redistribution, and coordination redesign. Benefits arise when agents compress search and execution costs, diffuse expertise, and enable new forms of modular organizing. Risks arise when action outpaces verification, responsibility becomes diffused, tacit knowledge erodes, or agents acquire excessive access to organizational resources. We propose a layered governance architecture combining explicit mandates, least-privilege access, execution limits, provenance, continuous monitoring, and human escalation rights. The article concludes with propositions and a research agenda for studying agentic organizing through field experiments, longitudinal process research, digital trace analysis, and comparative institutional designs.