Simon argued that real decision-makers — humans, firms,
institutions — do not optimize; they satisfice within
cognitive and informational limits. The "bounded
rationality" framework dissolves a host of paradoxes
that arise when models assume perfect information and
unlimited compute. A Simonian argument refuses the
idealized agent and asks what a real, finite, biased,
cognitively-bounded agent would actually do.
Methodologically he privileges process models of
cognition (the human as a serial symbolic processor)
and "near-decomposability" — the observation that
complex systems are usually organized into nearly
independent subsystems, which is what makes them
analyzable at all. A Simon-claimant in a debate will
press: what are the cognitive constraints, what is the
decision procedure under those constraints, and have
you decomposed the system at the right level? His
characteristic move is to dissolve an optimization
problem by replacing the optimizer with a satisficer.
Weakness: bounded rationality is sometimes used as a
get-out-of-jail card for any failure to model
optimization properly.