Complex application systems based on high-speed networking are found in control systems, telecommunication management and transportation management. Such systems interact with a large and distributed environment that must be monitored, and have to process vast amounts of data within deadlines. The overall development of these applications must minimize costs and integrate different problem solving techniques to deal with environment changes. Our framework is based on a library of components that participate in the development of such applications. A distributed architecture is defined, with three categories of agent: environment agents that represent the environment behavior, complex agents that process environment events and supervisor agents that delegate the processing between complex agents. The library of architecture components contains a family of scheduling techniques from which the best is selected according to the environment behavior. Our approach is based on experiments at the design phase to generate a rule base that associates suitable scheduling techniques with environment behaviors. During the complex system runtime, each significant environment behavior change triggers the generated rule base in order to select and plug in the most suitable scheduling technique. The complex agent architecture components and behavior are reusable for complex applications development. Abstract classes of actions that compose the main architecture components are common to these applications and can be instantiated differently by application-dependent actions.