For logical artificial intelligence to be truly useful,its methods must scale to problems of realistic size.An interruptible algorithm enables a logical agentto act in a timely manner to the best of its knowledge,given its reasoning so far. This seems necessaryto avoid analysis paralysis, trying to thinkof every potentiality, however unlikely, beforehand.These considerations prompt us to look for alternativereasoning mechanisms for filtered circumscription,a nonmonotonic reasoning formalism usede.g. by Temporal Action Logic and Event Calculus.We generalize Ginsberg’s circumscriptive theoremprover and describe an interruptible theoremprover based on abduction that has been used tounify planning and reasoning in a logical agent architecture.