Agents are not employees, and the people will know

The strongest objection to managing agents is not technical or economic. It is the feeling, shared by most people who do real work, that something is being got wrong when a piece of software is spoken of as a colleague. That feeling deserves to be taken seriously rather than managed, because it is tracking a real distinction.

Hannah Arendt, in The Human Condition, separated three things that the English word "work" runs together. Labor is the repeated effort that sustains life and leaves nothing behind. Work is the making of durable things. Action is what people do among other people: speaking, deciding, beginning something whose consequences they cannot fully control and must answer for. Her point was that modern societies had collapsed the three into one and lost the sense of what was distinctly human about the third.

An autonomous agent can labor and it can work. It can repeat, and it can make durable things, including texts and code and decisions that persist. What it cannot do, in Arendt's sense, is act, because action requires someone who can be held to account for a beginning, who can be asked why, who can be blamed or forgiven. An agent can be switched off. It cannot be held responsible, because nothing is at stake for it in being wrong.

This is why the people who objected when Lattice proposed putting agents on the org chart were not being sentimental. They were defending a distinction that their own standing depends on. An employee is a person who acts, and whose action is recognised as theirs. If an agent can be an employee, the recognition becomes a formality that attaches to anything that produces output, and the person's claim to be treated differently from the output weakens. The offence was a correct perception of a category being quietly dissolved.

Karl Polanyi made a related point from the other side. He called labor a "fictitious commodity": something traded on markets as if it were produced for sale, when in fact it is human life, which was not. Societies that forget the fiction, he argued, break themselves. The agent inverts his warning. Here at last is a productive input that really is a commodity, produced for sale and copied at will. The danger is no longer that we treat people as things. It is that, having things that work like people, we stop being able to say why the difference should matter in how we organise.

What follows for the Chief Workforce Officer is not a philosophy seminar. It is a design constraint. Everything in the agent lifecycle must be built so that it cannot be mistaken for the employee lifecycle, not because agents deserve less, but because people deserve the distinction to remain legible. Agents are selected, not hired; evaluated, not reviewed; retired, not dismissed. The accountable human who signs for each agent is a person precisely because only a person can be answerable. And the redesign of human roles around agents should move people toward action in Arendt's sense, toward judgment, exception, relationship and the decision that someone must own, rather than leaving them as supervisors of labor they no longer do.

There is a version of the hybrid workforce in which human beings become the residue, the part of the process that remains because it is not yet automated. There is another in which the agents take the labor and the work and people are freed, and required, to act. Which version an organisation gets is not determined by the technology. It is determined by whether anyone is responsible for the design, and that is the case for the role in its oldest form: someone has to decide what work is for people, and why.