A well designed interface is instrumental in making software easy to learn and use. The design of statistical software is inherently subjective, and there are many difficult decisions involved in creating interfaces that work cohesively within the intended domain. A forecasting framework sets out the design language for the forecasting workflow, guiding users conceptually about how different steps relate to one another. Making each element of the workflow extensible enables developers to implement new models, accuracy metrics, or other forecasting techniques that can be immediately used in conjunction with other forecasting elements. In this talk, I will examine the design decisions made when creating fable, the tidy time series forecasting successor of the forecast package.