Motor adaptation is a widely-used paradigm for understanding short-term learning. However, it is unknown how existing skillsets acquired through the long-term learning process that begins in utero can affect motor adaptation. Long-term learning likely causes changes in neural connectivity, which may shape the neural dynamics that can be produced. To understand the interaction between circuit connectivity constraints and a neural population’s ability to change its activity patterns, we modeled the neural dynamics of the motor cortex during skill learning and subsequent adaptation using a recurrent neural network. We trained the network on different skillsets with varying numbers of movements. We hypothesized that having a larger repertoire of movements would facilitate short-term adaptation since the activity is already primed to explore a larger range of possible activity states. Indeed, we found that larger skillset networks can adapt to perturbations more easily. In particular, multi-movement networks performed significantly better than single-movement networks. To understand how learning multiple movements impacts the underlying network dynamics, we examined the differences between networks initially trained on one or two movements. The dynamics of two-movement networks were more constrained, without leading to constraints in the output: two-movement networks had less variance in unit and population latent activity, but greater variance in motor output. They also had more predictable neural trajectories, suggesting that their dynamics have more organizational structure mapping motor output. When we reduced the structure with uninformative inputs, the differences in adaptation between multi-movement networks disappeared, showing that the structure facilitates adaptation. However, structure can also harm adaptation: networks with larger skillsets performed worse with larger perturbations and faster learning rates. Thus, learning multiple movements creates structure in neural space and highlights an inherent trade-off in skill acquisition: more structure facilitates adaptation requiring small changes in motor output, but can harm adaptation that requires large changes.