The First Non-Human on Payroll

A new colleague is added to a company Slack channel. They never join meetings, never exchange greetings. They appear only as a single line on the weekly KPI dashboard. Conversion rate: +3.1 percent. The entity behind that line is not human.

Since 2025, experiments granting AI agents limited budgets and payment authority have surged. Running ad campaigns, optimizing cloud costs, handling all customer interactions—tasks that once required human approval are now completed by agents spending money autonomously. A non-human with a wallet has arrived.

What is being destabilized here is not jobs. It is nomenclature. We have long called entities that deliver results 'workers.' That word has always carried with it evaluation, compensation, accountability, and dignity. The entire package was premised on the human.

Machines are pulling at one end of that package. Performance is clearly measurable. Evaluation follows. Where there is evaluation, the form of compensation emerges; and where there is compensation, accountability must be located. When one piece moves, the rest begin to shift.

The counterargument is firm. AI is just an expensive Excel macro, the argument goes, and talk of a 'salary' is nothing more than metaphor. Attaching KPIs to a tool is a convenient managerial figure of speech—not evidence that a worker has taken up residence inside it.

That is half right. An agent does not go hungry and does not fear termination. The problem lies with us. Once we name something by KPIs, humans unconsciously begin treating it with the grammar reserved for colleagues. A box appears on the org chart, and that box displaces a human one.

This is where the question must shift from capability to role. The crux is not how intelligent the agent is. It is what humans choose to keep as their own province as performance-delivering non-humans multiply.

Consider accountability. If an agent causes damage while executing a budget, where does responsibility lie? Between the team that wrote the code, the manager who granted authority, and the executive who signed off, accountability evaporates. An actor exists, but the responsible party does not. That blank is the very center of labor ethics for the next decade.

When intelligence migrates outward, the center of gravity of labor shifts with it. Execution becomes cheap and abundant. What becomes valuable is deciding what to execute, signing one's name to accountability for outcomes, and the judgment to say that the machine's confident answer is wrong.

Education splits first. Training to produce answers quickly has already lost to machines. What remains is the capacity to formulate the right questions and the capacity to doubt the machine's confident wrong answers. Korea's classrooms, long organized around ranking students by correct answers, stand on the front line of obsolescence.

Organizations split too. The moment humans and agents are evaluated on the same dashboard, managers inherit a new task: how to read non-human performance alongside human performance. Whose credit, whose fault? This act of translation is itself the new managerial labor.

Dignity is the final stakes. An agent has no need for dignity. But the moment an agent is seated in a colleague's chair, the dignity of the human in the next seat is redefined. The habit of seeing people only through measurable performance begins to read humans themselves through the grammar of agents.

Consider a small-to-medium manufacturer in Busan. If an agent begins handling estimates and purchase orders, what will be written in the job description of the office administrator with twenty years of experience? Whoever names that role determines that person's dignity. This is not a problem that arrives first in the headquarters boardroom—it arrives first on the shop floor.

Korea must begin its preparation not with infrastructure but with language. Whether to call an agent a worker or a tool. Whose signature bears accountability when something goes wrong. What words to use to distinguish human work from machine work. Naming comes before law.

Whether machines will become like humans is no longer an interesting question. The real question lies with us. As we cede ground to performance-delivering non-humans, what will humans choose to keep as human work to the very end? What the first non-human on payroll throws at us is not a pay stub—it is precisely that question.