Can Agentic AI Be a Cognitive Economic Agent?

This chapter asks whether agentic artificial intelligence can satisfy the features that cognitive economics attributes to economic agents. Agentic AI is the class of systems in which a large language model plans, acts through tools and revises its plans in the light of feedback. The benchmark is a definition of cognitive economics as the study of the adaptive interaction of embodied cognitive agents in markets and hierarchic organizations, summarised by Herbert Simon's scissors with the body as pivot. The chapter derives eight requirements from that definition and assesses agentic AI against each. The assessment is mixed. Agentic AI moves artificial intelligence from decision making towards problem solving. It works in open-ended environments through feedback loops, relies on external tools and memory, and already takes part in markets and organizations. In these respects it comes closer to the cognitive economic agent than any earlier form of AI. It fails on embodiment. It has no body whose regulation gives it something at stake, its competence is inherited from the recorded behaviour of embodied humans, and its goals are delegated. Agentic AI is therefore a pair of blades with a borrowed pivot. Three consequences follow. Embodied AI offers a partial, functional route to the missing pivot. Language-model agents used as models of human agents risk becoming a new as-if instrument. And because artificial agents now transact and coordinate alongside people, cognitive economics has to study them as a new kind of participant in economic life.