This research addresses a system of flexible worker assignments in a s
etting where there are more workers than machines. When organized usin
g this system, a production line balances itself by shifting the workl
oads continuously and automatically in response to changes in the stat
e of the system. The system is, in effect, buffering itself against va
riation by altering the work assignments on the fly. This allows the s
ystem to operate with very low levels of work-in-process inventory (WI
P). In this paper, the workers (rather than machines) are the factor t
hat limits the rate of output. We also assume that the line has a ''U'
' shape, but many of the results do not depend on this topology. An in
dustrial example is described. The system has some interesting and cou
nter-intuitive properties which we demonstrate under a variety of circ
umstances through an exploratory approach that uses both Markovian and
simulation models. Several different policies are compared under cond
itions of processing time uncertainty. We demonstrate that a flexible
assignment system can outperform fixed assignments in a variety of cir
cumstances. Of particular interest is the near absence of balance dela
y, even when the tasks cannot be divided equally among the workers.