Fabrication machines are becoming more capable, yet most still operate as passive tools that wait for explicit commands. This dissertation investigates how fabrication machines might instead take initiative to support creative making, without automating away the enjoyment of the process. The central question is how to design machine initiative that is helpful and timely while preserving human agency, skill development, and the intrinsic satisfactions of craft. The research proceeds in three parts. First, theoretical foundations are established, drawing on Sennett's analysis of craftsmanship, Schön's concept of reflective practice, and Horvitz's principles for mixed-initiative interaction. These foundations are then grounded in empirical understanding through interviews with maker entrepreneurs, observations of professional machine operators, and design explorations illustrating the design space for collaborative fabrication. Second, methods are developed for observing and capturing the multi-modal interactions that occur around fabrication machines. Through a study of 12 participants interacting with three tabletop machines (a clay extruder, a pen plotter, and a sewing machine), it is demonstrated that a minimal combination of egocentric and machine-mounted cameras captures the majority of relevant interaction information, and an instrumentation protocol is proposed. In the resulting FabriCam-5 dataset, implicit behavioral cues, such as hesitation, postural shifts, and gaze changes, are observed to precede explicit help-seeking, suggesting that machines attending to these signals could intervene before makers reach the point of frustration. Third, this premise is tested through a Wizard-of-Oz study comparing proactive and reactive machine assistance during a sewing task with 20 novice participants, resulting in the CoSew-4 dataset. Proactive assistance, in which the machine anticipated confusion based on behavioral cues and task stage, was perceived as significantly more timely than reactive assistance, with a large effect size, without increasing perceived annoyance. Proactive assistance also trended toward higher helpfulness ratings. These subjective differences appeared despite no measurable effect on task performance, suggesting that user experience measures may be more informative than efficiency metrics for evaluating co-creative systems. Study participants engaged socially with the machine and anticipated that their preferences for assistance would change as expertise developed. This dissertation contributes: (1) resources for studying mixed-initiative interactions in creative tasks, including instrumentation methods, protocols, and two synchronized multi-view video datasets (FabriCam-5 and CoSew-4); (2) empirical evidence that proactive machine assistance improves novice makers' experience without undermining their engagement; and (3) design principles for mixed-initiative fabrication systems that balance technical support with the preservation of creative agency. Together, these contributions lay the groundwork that is required for the design and development of mixed-initiative machines for creative tasks.