We study a dynamic moral hazard model, where a …rm can invest into future quality of its product which in turn is imperfectly observed by consumers. We analyse how investment incentives depend on the fi…rm's reputation and solve for long-term reputational dynamics.  Results depend intricately on the information structure of consumer learning. When consumers learn through good news, incentives are increasing in reputation and dynamics are convergent. When they learn through bad news, incentives are increasing in reputation and dynamics are divergent. Preliminary analysis on Brownian learning indicates that the results of the good news case carry over to a large class of learning dynamics.