This paper studies a principal-agent problem with learning before production. The learning stage is modeled as strategic experimentation with both dynamic moral hazard and adverse selection. We fully characterize effort off the equilibrium path in a mixed model to show that moral hazard requires the principal to reward success in learning, but adverse selection may induce the principal to reward failure. We find the principal uses the timing of failure as a screening instrument despite the presence of moral hazard. Therefore, both success and failure are rewarded with different payments and specific timing in the optimal contract. We also study whether the principal should hire the same agent for both the learning and production stages (integration), or different agents (separation). While separation is optimal under pure moral hazard, we show that adverse selection can make integration optimal. Having the same agent working on both stages enables the principal to use the adverse selection rent to address dynamic moral hazard. If adverse selection is severe, yielding a large rent, the principal can satisfy the moral hazard constraints by spreading the adverse selection rent over the duration of experimentation.