Ideal Agents -- What's the Point?

Given that real human agents can’t possibly live up to the Bayesian model of belief and decision, what kind of norm is the Bayesian model supposed to be?

For example, one version of the usual Bayesian model begins by supposing that the ideal agent has only a preference ranking among alternative acts (or outcomes). Then probabilities and utilities are introduced as merely a way to model this preference ordering. So the ideal agent has no use at all for the utilities and probabilities that (according to a representation theorem) model the preference ordering. She already has her (ideally coherent) preference ordering, and so doesn't need the representing probabilities and utilities to help her make decisions. How, then, is this model supposed to be a normative guide for real agents?

Please let me hear what you think !!!