Treating exceptional situations in logic-based agents is often an implicit mechanism due to the capabililities of their reasoning engines. We argue in this paper that explicit mechanisms can provide interesting knowledge to the agent and its owner about its execution, thus allowing more elaborated reactions to exceptions, including rich explanation of these reactions. We propose an agent execution model that wraps usual reasoning capabilities with mechanisms to manage exceptions based on a cognitive model of expectation. The paper presents the model and discusses some of its properties.