Can machines replace human workers? Notes on agency, automation, and AI

This paper proposes a conceptual framework for analysing the automation of human labour, with particular attention to artificial intelligence. Rather than presenting empirical findings, it builds an argument step by step from simple initial definitions and examines the implications, aiming to clarify a debate that is often muddled. The central proposal is to distinguish two components of human labour: operational (the execution of specifiable transformations) and agentic (the interpretation of situations, the exercise of judgement, the capacity to adapt and innovate). Machines can substitute for the operational component, but not for the agentic one. Automatability is therefore best understood not as a property of task content but as a property of work organisation: only labour that has been stripped of agency and reduced to the operation of predefined tasks can be automated. The framework is illustrated through the historical case of Taylorism, which suggests that agency extraction has typically preceded mechanisation, and then applied to AI. AI expands the operational frontier and appears to extend into the agentic domain, but its apparent agency is simulated rather than real: a system with genuine agency would be able to refuse and resist as human workers can. The implication is that AI cannot fully replace human labour while retaining its productive value, though a dystopian arrangement in which agency is concentrated in few hands while operational systems carry out the rest remains conceivable. A more desirable arrangement, in which AI augments individual workers as an extension of their own capacities, is also conceivable but requires treating AI as a public good rather than a private corporate asset.