We study learning in organizations comprised of short-lived agents wit
h heterogeneous beliefs. It is shown that the sequence of actions conv
erges in probability to the set of ex-post optimal actions. Under some
identifiability conditions, beliefs also converge in probability to p
oint mass on the truth. Moreover, while actions and posterior beliefs
converge in probability, they can simultaneously almost surely fail to
converge. Our simulations of Kiefer's 1989 monopoly learning model su
ggest that if the firm is operated by a sequence of short-lived manage
rs with heterogeneous beliefs then complete learning of the demand cur
ve obtains.