Agent based systems and architectures provide a firm foundation for design and development of an autonomic system. The key challenge is the selection and efficient use of effective agent architecture. A model-driven approach accommodates the underlying architecture to automate, as much as possible, the development process. The Cognitive Agent Architecture (COUGAAR) is a distributed agent architecture that provides the primary components and an implementation platform for this research. COUGAAR has been developed primarily for very large-scale, distributed applications that are characterized by hierarchical task decompositions and as such is well suited for autonomic systems. We propose a framework for the agent-based, model-driven architecture for autonomic applications development. The framework consists of two main parts, General COUGAAR Application Model (GCAM) and General Domain Application Model (GDAM). Some COUGAAR related performance issues are also discussed.