Federal and state employment programs for low-skilled workers typically emphasize rapid placement of participants into jobs and often place a large fraction of participants into temporary-help agency jobs. Using unique administrative data from Detroit's welfare-to-work program, we apply the Chernozhukov-Hansen instrumental variables quantile regression (IVQR) method to estimate the causal effects of welfare-to-work job placements on the distribution of participants' earnings. We find that neither direct-hire nor temporary-help job placements significantly affect the lower tail of the earnings distribution. Direct-hire placements, however, substantially raise the upper tail, yielding sizable earnings increases for more than fifty percent of participants over the medium-term (one to two years following placement). Conversely, temporary-help placements have zero or negative earnings impacts at all quantiles, and these effects are economically large and significant at higher quantiles. In net, we find that the widespread practice of placing disadvantaged workers into temporary-help jobs is an ineffective tool for improving earnings and, moreover, that programs focused solely on job placement fail to improve earnings among those who are hardest to serve. Methodologically, one surprising result is that a reduced-form quantile IV approach, akin to two-step instrumental variables, produces near-identical point estimates to the structural IVQR approach, which is based on much stronger assumptions.