Sheet metal spinning is an incremental forming technique practiced for a long time only by experienced, skilled craftsmen. Over the last sixty years, attempts have been made to automate the process, but today the industrial practice still relies heavily on the skill of experienced operators. Complete analytical solutions to predict workpiece failure based on the path of the tool are not available, and finite element simulations of the process are too time-intensive to be of practical use for online toolpath correction. Therefore, today the design of toolpaths to avoid failure in spinning remains an art acquired by practice. In this paper, we approach the issue by designing an enhanced teach-in/playback system. We connect a haptic device to a CNC spinning machine; this device allows a human operator to control the working roller manually while feeling the force applied to the workpiece. Position, force and workpiece shape sensors allow collecting information on the toolpath followed by the operator and on its influence over the mechanics of the process. The opportunities offered by this system to derive rules for toolpath design are explored in two case studies on force control and wrinkling recovery, with new insights on the relationship between toolpaths and failure. A research agenda is outlined to exploit the full potential of haptic metal spinning.