Our powerful computers help very little in debugging the program we have so we can change it into the program we want. We introduce Conversational Programming as a way to harness our computing power to inspect program meaning through a combination of partial program execution and semantic program annotation. A programmer in our approach interactively selects highly autonomous “agents” in a program world as conversation topics and then changes the world to explore the potential behaviors of a selected agent in different scenarios. In this way, the programmer proactively knows how their code