Carrying out previously developed plans in the real world, such as picking a friend up at the airport or cooking dinner, requires that we possess some basic abilities for behaving rationally. As we all know, things just never go to plan 100 percent of the time, so possessing the facilities to cope with this uncertainty makes us good execution agents. However, the term "rational" is easy to state, and difficult to define. Just what abilities do we possess that allow us to reason in a rational manner? Researchers in the field of Artificial Intelligence (AI) have been attempting to make a precise definition of rationality since the beginning. Though we still do not have a universally accepted definition, we do have systems which to varying degrees appear to demonstrate qualities which can be said to be rational. If we characterise those systems, we can define a minimum set of capabilities which allow us to design a basic rational agent. This thesis proposes a design of a reactive execution agent (REA) which accepts and executes task directives in a dynamic environment. This design draws on an amalgam of ideas emerging from situated rational agency research. It yields an agent which is able to accept task directives (or reject them), reason about directives to determine how to achieve them (maintaining commitments to achieving other directives), respond to execution failures, and communicate with a superior agent when further deliberation is required beyond the abilities of the REA.