This paper presents AgLOTOS as an algebraic language dedicated to the specification of agent plans in ambient systems taking into account timing constraints. It offers a rich and modular approach to express and compose elementary plans in order to execute them concurrently. We show how a plan is built automatically as a system of concurrent processes from the belief-desire-intention attitudes of the agent. The AgLOTOS semantics allows to deal with possible plan execution failures, caused by the passing of time for performance of actions. In this situation, this information is captured at the mental process level allowing to update the agent mental state. Moreover, the associated semantics accords with the possibility of revising the agent plan, as the set of intentions changes.