Procedural problem solving is an important skill in most technical domains, like programming, but many students reach problem solving impasses and flounder. In most formal learning environments, instructors help students to overcome problem solving impasses by scaffolding initial problem solving. Relying on this type of personalized interaction, however, limits the scale of formal instruction in technical domains, or it limits the efficacy of learning environments without it, like many scalable online learning environments. The present experimental study explored whether learners' self-explanations of worked examples could be used to provide personalized but non-adaptive scaffolding during initial problem solving to improve later performance. Participants who received their own self-explanations as scaffolding for practice problems performed better on a later problem-solving test than participants who did not receive scaffolding or who received expert's explanations as scaffolding. These instructional materials were not adaptive, making them easy to distribute at scale, but the use of the learner's own explanations as scaffolding made them effective.