The curse of dimensionality limits jointed structures. Robots are famously hard to program because they operate in such high-dimensional spaces. Dozens of joints (at three degrees of freedom each) form a numerical landscape in which optimization is difficult, because getting lost is so easy. The problem is that joints are independent and thus their actions are combinatorial. If ball-joints were replaced by thousands of muscle fibers spanning a skeleton, the problem would be paradoxically simpler to solve.
Tensegrity is the midpoint between joints and the continuum. An arbitrary arrangement of joints and muscle fibers would form a tensegrity structure, equivalent to those public sculptures made entirely from compressive struts which don't touch, held in place at their ends by tensile cables strung between the ends. Practically, tensigrity structures are even harder to control than robots, but their functional division between tension and compression allows a thought-experiment. If one imagines coalescing the struts into fictive "bones," and locally collimating the cables (i.e. muscle fibers) into parallel bundles, then in the continuum approximation we have a simple mechanical structure: rigid links surrounded by a 3-D tensile mat.
Force-pixels form virtual muscles and force-geodesics. If the "bones" are co-linear they form a spine, and if the muscles are well-organized they form a cylindrical shell of co-axial stripes and two counter-rotating helical shells of muscle. Those component fibers effectively form a fully symmetric vector field of potentially activated tensile micro-forces, each vector in the field separately addressable, like a pixel. While a brain could in principle activate those force-pixels in arbitrary combinations to form all kinds of virtual muscles, the most mechanically efficient combinations (both for exerting force and for detecting vibrations) will be long, minimally-curved arcs optimizing some variational principle (such as "minimum mean-squared force gradient producing this particular shape in equilibrium"). The optimum geometric arrangement of mechanical forces for any particular task will belong to the class of curves known as "geodesics", so these activation stripes would be "force-geodesics."
In this view there are no isolated muscles like "biceps," "triceps" and so on, only millions of well-coordinated fibers. Likewise, no there is no "strength," "flexibility," or "tone" of muscle tissue, only the relative degree of high-speed coordination a given fiber has with its close neighbors. The entire problem of motion and comfort comes down to the data used by local algorithmic strategies of proprioception and control.