Same Method, Different Task: What the Bitter Lesson Does Not License in Social Simulation

The bitter lesson holds that general methods leveraging computation defeat hand-encoded domain knowledge. We argue it carries a condition its popular reading omits: in every case Sutton cites, the general method was scaled against the \emph{same task} the domain engineering was solving. Social simulation substituted across tasks instead, replacing hand-built agents with an artifact optimized for predicting text. The move is bitter-lesson-shaped, but the lesson does not license it. Prediction is a solitary act: the predictor bears no consequences from the interaction it models. Cooperation, defection, and norm enforcement are phenomena \emph{of interested parties}; they do not exist for a disinterested observer at any scale. The missing ingredient is not emergence, which is cheap. It is stakes. We show that several results reported as anomalies follow from this account and test it in a preregistered 540-agent resource-allocation pilot across three model families. In two of three families, persistent stakes reshape elicited strategies toward a solvency anchor (Claude Sonnet 4.6 is the exception, producing the cooperative script regardless of condition). No family reproduces human-like population heterogeneity, and no agent in any condition adopted a max-extraction schedule---even when high extraction was the only route to solvency, 22--32\% of persistent-stakes agents went insolvent by under-extracting. We argue the tractable programme is mixed-motive settings used to \emph{shape} agents rather than only to test them.