The Clinton administration has made great efforts to promote a next-generation learning industry in the United States, but the goal has been elusive. The most visible results so far are 1) a plethora of expensive reports that restate subsets of the published literature on learning technology and reflect primarily the perspective of the various authors, 2) a variety of interesting media exercises (Net Day, Cyber Ed, etc.), and 3) a variety of large scale hardware-dissemination exercises (Tech Corps, TIIAP, Goals 2000). With the exception of a measurably improving internet infrastructure, all this activity has not brought us markedly closer to a commercially viable and self perpetuating next-generation learning industry. Why not? I believe we are already in possession of an important component for such an industry - a workable engineering discipline for the principled implementation of consistently effective courseware. I don’t argue that research on the optimization of courseware should stop. Further advances in this area are certain to occur as research proceeds. What I am claiming is that a documented approach already exists for engineering courseware that roughly doubles the effectiveness of today’s best mainstream courseware. Research on cognitively based tutoring systems, often called Intelligent Tutoring Systems, has generated the new approach. Development methods for current-generation ITSs are sufficiently mature to support routine application. These methods are derived directly from cognitive science, but the mainstream courseware industry has failed to adopt the methods.