In AgentSpeak, agents react to events by selecting plans whose contexts are satisfied. Jason, a popular AgentSpeak interpreter, resolves conflicts between multiple applicable plans by choosing the first one defined in the source code. This approach, however, places an unnecessary burden on developers to manually order their plans. We propose a new context-aware selection function that automatically prefers the most specific plan by leveraging a semantic notion of specificity based on logical entailment. Experiments on Jason’s standard examples demonstrate that our function preserves intended behavior while enhancing maintainability. We also discuss performance trade-offs and limitations observed in multi-agent simulations.