Who is actually managing the work your AI does?
Delegating work to AI still requires domain, managerial and technical supervision focused on the organisational outcome.
Imagine employing somebody to generate sales opportunities. They identify prospects, research them, decide whom to approach, prepare messages, follow up and arrange meetings. At some point you would probably ask how they were performing. Are they approaching appropriate customers? Are their messages accurate? Are they representing the business properly? Are they learning from unsuccessful approaches? Now imagine exactly the same work is performed by an AI agent. Who asks those questions? It is tempting to make this an IT or AI-governance responsibility. Technical governance certainly matters. Agents need appropriate permissions, security, monitoring and auditability.
But an agent can remain completely within its technical permissions and still do poor work. That is a management problem. As AI moves from something employees use towards something to which organisations delegate work, responsibility for managing that work cannot simply move to whoever operates the technology. The people responsible for the organisational outcome need to remain involved. That may require several forms of expertise. A domain specialist may judge whether the work itself makes sense. A manager may hold authority to make commitments on behalf of the organisation. A technical specialist may provide assurance that the AI system is operating as intended.
These people are not interchangeable. Consequential artificial agency may therefore require a supervisory structure, rather than merely “a human in the loop”. This will become more important as managers acquire access to tens or potentially hundreds of specialised agents. Agentic AI may increase managerial leverage much faster than it increases managerial capacity. Being able to delegate fifty streams of work does not mean one person suddenly acquires the ability to understand fifty streams of work. So before asking what else an agent could do, there is another question worth asking:
Who is actually managing the work our AI is already doing?