In monopsony models, wage markdowns induce deadweight loss and are therefore inefficient. Yet markdowns also arise in models with backloaded efficiency pay, where they are designed to induce effort among early-career workers and are thus efficient. To reconcile—and empirically distinguish—these two mechanisms, we build a dynamic model incorporating labor market power and endogenous effort. Estimating a team production model on novel data on U.S. public accounting firms, we find evidence of both: markdowns for junior workers and markups for senior ones reflect incentive-providing backloading, while monopsony power induces a ‘lifetime’ wage markdown of 15%.