Self-efficacy is defined as a person's subjective confidence in their capability of executing an action and has been shown to be one of the most powerful motivators of human action predicting performance across a variety of domains. Self-efficacy has been associated with brain level neural processes and efficacy-like confidence mechanisms are incorporated into decision making in many cognitive informatics and cognitive computing models. Current computational implementations, however, do not directly model self-efficacy at either the theoretical or neural level. This paper reports on the computational modeling of self-efficacy based on principles derived from the Unified Learning Model (ULM) as instantiated in the multi-agent Computational ULM (C-ULM). Description of the modeling of self-efficacy within the C-ULM is provided. Results from simulations of self-efficacy evolution due to teaching and learning, task feedback, and knowledge decay are presented. The C-ULM simulation is unique in tying self-efficacy directly to the evolution of knowledge itself, consistent with recent neurological findings, and in dynamically updating self-efficacy at each step during learning and task attempts. Implications for research into human motivation and learning and for cognitive computing are discussed.