Who bought the agent?

Ask any chief executive how many people the company employs and you will get a number within a minute. Ask how many AI agents it employs and you will get a pause, then a referral to IT, then a guess. That gap is the whole argument for the Chief Workforce Officer in one question.

Here is how an agent usually arrives. A business unit has a backlog. A vendor demonstrates a system that clears it. The cost is a software subscription, which sits comfortably inside the unit's own budget and needs no headcount approval. Procurement checks the contract, security checks the data flows, and the agent goes live. Nobody in that chain was asked the question that matters: what share of this unit's work will now be done by something that is not a person, and who answers for it?

This is not negligence. It is the natural result of treating an agent as software, because software has always been bought this way. The trouble is that an agent that handles customer complaints, screens applicants or drafts contracts is not doing what software used to do. It is doing work that, last year, was a job. The decision to give it that work is a workforce decision, and workforce decisions used to pass through a function whose whole purpose was to think about them.

Human resources is not in the room because, formally, nothing has happened to any human. No one was hired or dismissed. The headcount report is unchanged. HR discovers the deployment months later, when it shows up as a hiring freeze in that unit, a reorganisation request, or a quiet rise in attrition among the people whose work was partly taken. By then the decision has been made and the budget spent.

Finance is in the room, but finance sees the one number that favours the agent: its cost against the cost of the people it displaces. The number finance does not see is the cost of checking the agent's work, because that cost lands on existing staff, unbudgeted, as a few extra hours a week of reviewing outputs that used to come from a colleague. It never appears as a line item, so it never enters the comparison.

IT is in the room, and IT is accountable for the agent working. That is a real and demanding responsibility. It is not the same as being accountable for what the agent does. When the system is up, responsive and secure but giving a thousand customers the wrong answer, IT has met its obligations. Somebody else was supposed to be watching the content, and in most companies that somebody does not exist.

So the composition of the workforce, the question of what fraction of the organisation's work is done by people and what by agents, is currently decided as the sum of many small software purchases, each individually sensible, none of them seen as a whole by anyone. That is how an organisation wakes up one day to find that a quarter of its customer interactions and half of its first-draft analysis are produced by systems nobody has evaluated against the standard they would apply to a person.

The Chief Workforce Officer exists to put that decision back where it belongs. Not to slow it down, and not to oppose agents, but to make it a decision rather than an accident. A CWO asks, for each task family, what the mix of human and artificial capacity should be and how fast it should change. A CWO insists that every agent has a named human who signs for its output before it is switched on. And a CWO keeps the one register nobody else keeps: the list of agents the organisation actually employs.

In most companies this will be the chief human resources officer with a larger mandate, and the title can come later. What cannot wait is the mandate. Until someone holds it, the answer to "who bought the agent?" will go on being "a department," and the answer to "who is responsible for it?" will go on being a meeting.