Who is accountable when the agent gets it wrong

A simple question disarms many projects: who is accountable for this?

In a Cognitive Enterprise, the answer is still a person with a first and last name. The DRI does not review the 4,000 outputs. They are accountable for the system that produces the 4,000. Accountability still exists; its object simply changes.

This reorganizes management work in a concrete way. The classic “how many direct reports per manager” becomes “how many agentic processes can one person supervise with quality,” and that number depends less on their talent than on the instrumentation in place.

With eval running, a cost dashboard, and defined gates, one person can oversee six processes in a few hours per week. Remove the instrumentation and that same number drops to one or two, because the person goes back to reviewing outputs case by case — in other words, back to operating instead of supervising.

That is why span of control measures operational maturity, not individual productivity.