According to common views of skill acquisition cognitive control diminishes in the course of skill learning, with perceptual-motor processes operating largely independently of cognitive control in advanced stages of expertise. We outline an alternative theory called Mesh which proposes that cognitive control continues to play a major role in advanced skill. Fundamentally, skills do not automate strongly because real-world skilled action is too complex: it is commonly subject to the requirement of achieving precise outcomes in the context of high degrees of situational variability, whereas the development of strong automaticity requires a high degree of constancy of perception-action relations. Conversely, skill learning ameliorates the speed and capacity limitations faced by cognitive control of novice actions. Overall there is increased integration between cognitive and motor processes in advanced skill rather than greater independence. Mesh incorporates a model of action control that explains how the combination of flexibility and precision of skilled action is achieved. Conceptual representations of the situation serve as the basis for high level action specifications that configure lower order action production processes for the particularities of the situation. This model identifies an important role in action control for forms of action awareness such as sense of agency and sense of control.