Chief Agent Officer, agent manager, or Chief Workforce Officer?

When this series began, the question of who is responsible for a workforce of people and agents had no settled answer and few named candidates. By the end of 2026 it has two, each with institutional backing, and the difference between them and the role argued for here is now the clearest way to say what a Chief Workforce Officer is.

The first candidate is the agent manager. In February 2026 Suraj Srinivasan of Harvard Business School and Vivienne Wei of Salesforce argued in Harvard Business Review that companies need a new kind of leader to orchestrate how agents learn, perform and work safely alongside people, and compared the role to the product manager of the software era. Their example is a support-agent manager at Salesforce who watches the performance of generative-AI agents through dashboards and coaches them as a manager would coach staff.

The second is the Chief Agent Officer. In August 2026 a Microsoft security blog set out the case for a portfolio-level executive who treats agents as assets with a lifecycle: deciding which agents to build, reuse, scale or retire; ensuring every production agent has a defined owner, purpose and review schedule; governing their permissions; managing them from proposal to retirement; measuring whether they complete tasks correctly; and setting the balance of autonomy and oversight. The post is careful to say the role replaces neither the CIO, the CISO, the chief data officer nor the chief AI officer. It does not mention human resources, and it does not mention the people whose work the agents now do.

Both proposals are right about something, and both would be improvements on the vacuum most organisations have today. The agent manager is right that someone close to the work must watch the agents daily, and that this is a skill. The Chief Agent Officer is right that agents have a lifecycle, that every one needs an owner, and that the portfolio needs to be seen whole. Much of its six-point charter could be lifted into the governance half of the role described in this series without changing a word.

What both leave out is the other half. The agent manager watches the agents. The Chief Agent Officer governs the agents. Neither is responsible for what happens to the people: which human roles are redesigned around the agents, who carries the verification burden and whether it is budgeted, what happens to the entry-level work through which the organisation produces its own future experts, and how the composition of the workforce, the share of work done by people and by agents, is decided rather than discovered. In the Chief Agent Officer's world, those questions belong to HR, which has no authority over the agents. In the agent manager's world, they belong to no one above the team.

That split is the mistake this series has been describing from the first essay. An organisation that governs its agents in one function and its people in another has reproduced, with a new title, the arrangement in which technology decides and HR finds out afterwards. The agents will be well governed. The workforce will not be designed.

There is now evidence that the human half is where the damage lands. In May 2026 a team from Boston Consulting Group and Boston University reported in Harvard Business Review on a large experiment in which agents were framed as employees. The framing reduced individual accountability, increased unnecessary escalation, lowered the quality of review, and left employees less sure of their own roles, without improving adoption. Their conclusion was that the question is not whether to deploy agents but how to redesign workflows, roles and governance so that humans stay clearly accountable. That is a finding about people, and the role that answers it must have authority over people.

So the three roles are not rivals so much as layers, and the layering is the point. The agent manager is the line role: close to the work, watching performance, escalating. It reports, through the normal business structure, to leaders who own outcomes. The Chief Agent Officer's charter is the governance half of a larger mandate: lifecycle, ownership, permissions, portfolio. And the Chief Workforce Officer holds that mandate together with the half the others omit: the composition of the workforce, the allocation of work between people and agents by the cost of checking it, the measurement of the whole capacity including verification hours, and the redesign and protection of human roles. The CWO reports to the chief executive beside the CIO and CFO, because only from there can the role say that an agent which is cheaper, more capable and well governed should still not be deployed where no one can check its work, or where its deployment would hollow out the layer that produces next decade's checkers.

Could the Chief Agent Officer simply grow into that? Possibly, and if an organisation already has one, extending the charter is easier than creating a new office. But the direction of growth matters. A role born in the technology and security functions, named for the agents, will keep seeing the agents as the subject and the people as context. A role born from the workforce function, named for the whole of it, sees both as what it is for. The title is less important than which way the role faces, and this series has argued that it should face the work, not the machines that now do some of it.

The vocabulary question is not settled either. The same spring, Harvard Business Review published advice on creating onboarding plans for AI agents, the employment metaphor this series has spent several essays resisting, and the BCG experiment suggests the metaphor is not harmless. Selected, deployed, evaluated, updated, retired: the verbs are still available, and the evidence now favours them.

A year ago the question had no answer. Now it has three, and the choice among them will be made in the next two or three years, mostly by default, as titles are created to fill gaps that incidents reveal. The argument of this series is that the default will be a technology-facing role, that it will leave the workforce undesigned, and that the organisations which instead give one executive the whole of it will be the ones that still know how to do their own work in 2035.

What that whole mandate consists of is now written down. The Institute has published a competency framework and code of ethics for the role: six domains, eighteen competencies, three levels of scope, and the duties a holder owes to the people whose work they change. It is a draft, versioned and open to challenge, which is how a profession’s standards should begin.