The rapid advancement of Artificial Intelligence (AI), particularly Large Language Models (LLMs), has intensified interest in automating knowledge work—cognitive tasks involving information processing, analysis, and decision-making. Unlike earlier automation technologies that primarily displaced physical labor, AI targets the cognitive domain, raising novel questions about accuracy, skill preservation, and appropriate human oversight. This paper describes a systematic framework for evaluating when and how to deploy AI systems for knowledge tasks. Drawing on insights from robotics, outsourcing, and offshoring literature, the authors identify key distinctions in cognitive automation, including concerns about information veracity, scalability, and the risk of expertise erosion. The proposed framework evaluates tasks across four critical dimensions: criticality (consequences of failure), accuracy (AI system performance), novelty (task variability and environmental complexity), and observability (ability to monitor system performance). Based on these assessments, the framework recommends deployment strategies ranging from human-in-the-loop (HITL), where AI assists human decision-makers, to human-on-the-loop (HOTL), where AI operates autonomously with human supervision, to human-out-of-the-loop systems requiring minimal oversight. The paper provides practical guidance for organizational leaders on benchmarking AI systems, managing risks, and maintaining critical competencies while leveraging automation benefits. As AI capabilities continue to evolve, the framework emphasizes the importance of ongoing reassessment and adaptive management to balance efficiency gains with the preservation of essential human expertise and organizational resilience.