We first consider a benchmarkmodel without certification and learning only through the agent’s work history. Even in this simpler setting, we find surprising results. The key driver of these results is that high-quality agents can be forced out of the market after a string of bad signals, after which learning stops and the agent’s reputation has no chance to improve. Since learning may be halted prematurely, having access to the agent’s work history alone does not guarantee a socially efficient outcome. In fact, our result shows that for many types of information processes,work history information does not provide any social benefit at all. This result is reminiscent of work on “bad reputation” (Ely and Välimäki 2003; Ely et al. 2008) where the ineptness of reputation is due to the myopia of the short-lived principals. But in our model, the failure of reputational forces alone results not because of moral hazard, but because of an informational friction between principals and agents. Principals are not willing to hire once the agent’s expected quality falls to the price level, but there is still positive social value from hiring at this because an agent with a low reputation might still be of high quality—hiring has an informational value, but since the principal only interacts once, it does not appropriate this value. Thus, principals would force agents, even agents who face no moral hazard issues, out of the market inefficiently due to their short-sightedness.