He says he adds an output message, but I've tried this myself and I find that quite a lot of the time the agent prefers its own internal monologue over the output of a command.