Defining AI agents: From definitions to stopping conditions