This article investigates the design of incentives in a dynamic adverse selection Framework where agents' production technologies display learning effects and agents' learning rates are private knowledge. In a simple two-period model with full commitment available to the principal we show that whether learning effects are over- or underexploited crucially depends on whether more effcient agents also learn faster (so costs diverge through learning effects) or whether it is the less effcient agents who learn faster (so costs converge). We further show that an overexploitation of learning effects can occur also if the full-commitment assumption is relaxed.