Issue Autonomy is framed as a graduated spectrum with designed human entry points

Summary

Deloitte advises organisations to set clear limits on agent decision-making through graduated levels of autonomy, each with triggers for human oversight. It describes three phases. Augmentation, in which agents enhance human capabilities, is today's reality, automation of tasks within processes defined by people is an emerging capability, and true autonomy with minimal oversight is a future state tied to artificial general intelligence. Success, the report argues, depends on agent supervisors who step into workflows at deliberately designed points to handle exceptions that need human judgment. That role is less about checking every output than about strategic handoffs at critical decisions. Leaders are urged to keep reassessing AI capabilities so they delegate only responsibilities agents are suited to, supported by zero trust authentication that continuously verifies agent actions.