Comment by Rory Blundell

Every AI agent should have a clear identity, role, and level of authority, along with oversight and recourse mechanisms. It’s imperative that every AI agent has a named human responsible for its actions. For enterprises to confidently trust AI agents with high-stakes, non-routine work, a specific human must ultimately own the outcome. When a human is ultimately responsible for an agent’s outputs, oversight becomes an active part of how the AI agent operates. But naming an accountable human is only half of the answer to achieving human-agent harmony. If a human sponsor is going to make good on that responsibility, they need the tools to actually steer the agent in real time. The existing human-in-the-loop models fall short, often giving people oversight at a single decision point. Namely, the AI agent proposes an action and the human approves or rejects it. Meaningful human oversight must come earlier in the process, also known as human-in-the-lead. The human worker sets the strategic intent, defines policy boundaries, and actively provides real-time steering during the AI agent’s execution.