Does the use of co-execution agentic AI reduce deployers’ ability to exercise effective control? Regulations may struggle to keep pace or may apply to an increasingly narrow range of situations. AI deployers are called to find realistic solutions to balance the potential of general-purpose executing agentic AI with the immense number of tasks that AI can execute much faster than any ordinary human could process them. Deployers may therefore need to move from continuous observation towards bounded autonomy, sealing a sort of virtual moral pact with the AI agent aimed at what might be called ‘virtuous machine autonomy’ - the expectation that the agent will remain within its intended mandate. As AI has already shown itself capable of implementing autonomous harmful actions, deployers’ containment and isolation tools can become essential when human monitoring and virtuous autonomy are insufficient.