The voracious token appetite of AI agents invites reflection on what we mean by 'intelligence' and 'efficiency' in artificial systems. Human cognition, while far from perfect, employs sophisticated heuristics and selective attention to avoid constantly re-examining every past detail—we summarize, abstract, and forget irrelevant information to act efficiently. An AI agent's need to reread its entire history at each step reveals a stark contrast: it possesses vast recall but lacks the ability to distill experience into useful abstractions or to prioritize what truly matters for the current goal. This inefficiency isn't merely a technical flaw; it suggests that today's agents simulate aspects of behavior through brute-force computation rather than through the kind of adaptive, economical reasoning that characterizes biological intelligence. It raises a profound question: Are we building systems that think, or are we building very elaborate lookup tables that mimic thought through exhaustive replay?