Learning is essential in allowing policies and programmes to become adaptive to uncertain and changing circumstances. In this article, we use the case of the Dutch National Collaboration Programme on Air Quality (in Dutch: Nationaal Samenwerkingsprogramma Luchtkwaliteit (NSL)) to argue that the organisation of learning processes influences the extent to which policies and programmes can adapt. Learning is a diverse process that focuses both on improving knowledge of the effects of possible policy strategies and on bargaining and negotiation in which we interpret and make sense of such knowledge. Learning strategies can be organised by either focussing on exploring new ways to understand and approach problems or on exploiting existing knowledge, measures and capabilities by optimising and refining them. To be adaptive, it is important to balance both strategies, as otherwise two main risks might undermine the programmes' adaptive capacity. First, there is the risk that the implementation of adaptive programmes is constrained by premature consensus, constraining the capacity to learn regarding the suitability of, and agreement on, existing problem definitions and programme goals during programme implementation. Second, there is also the risk of premature programming, constraining the capacity to learn regarding the suitability of the actions and approaches adopted. What makes the NSL such an interesting case study is that while the programme was designed to be adaptive on the basis of an on-going learning process, in practice, it largely failed to do so. On the basis of 67 interviews with stakeholders in the NSL and two focus group discussions, we show that the NSL failed to anticipate both risks. Processes of learning have become marginalised and focused predominantly on exploitation at the expense of exploration. As such, the NSL convincingly shows how a lack of organising learning is constraining the capacity to adapt to changing circumstances and also contributes to its possible failure.