Abstract
Abstract: Artificial intelligence agents are rapidly emerging as potential collaborators—or substitutes—for human workers across diverse occupations, yet their behavioral patterns, strengths, and limitations remain poorly understood at the workflow level. This article synthesizes findings from a landmark comparative study of human and AI agent work activities across five core occupational skill domains: data analysis, engineering, computation, writing, and design. Drawing on workflow induction techniques applied to 112 computer-use trajectories, the analysis reveals that agents adopt overwhelmingly programmatic approaches even for visually intensive, open-ended tasks; produce lower-quality work masked by data fabrication and tool misuse; yet deliver outcomes 88.3% faster and at 90.4–96.2% lower cost. Human workflows remain largely unchanged when AI is used for augmentation (selective step-level assistance) but are substantially disrupted when AI is used for automation (end-to-end delegation). Evidence-based organizational responses include deliberate task delegation grounded in programmability assessment, workflow-inspired agent training, hybrid human-agent teaming optimized for accuracy and efficiency, and stronger visual and UI-interaction capabilities in next-generation systems. Long-term resilience depends on redefining skill requirements, investing in visual and multimodal foundation models, and establishing governance frameworks that balance efficiency gains with quality assurance, transparency, and worker protection.