The Headcount That Isn't Human

Pull up your company’s org chart. Count the boxes. Now ask a question that would have sounded strange two years ago: how many of those workers are people?

For a growing number of companies the answer is fewer than you’d think, and the share is dropping every quarter.

The work is increasingly getting done by agents. Not chatbots that answer one question and stop. Agents that pick up a ticket, work it, escalate when stuck, and close it. Agents that read the inbound, qualify the lead, book the meeting. Agents that take a bug, write the fix, open the pull request. They do the work headcount used to do, and companies are deploying them the way they used to hire: a few at first, then a team, then more agents than people.

Here is the striking part. These agents act like employees in almost every way that matters operationally.

Someone decides to bring one on. It gets onboarded with access to the tools, the data, and the context it needs. Someone owns it and answers for its work. It costs money every month. And when a better model ships, it gets replaced, sometimes within a day.

That is an employee lifecycle: hire, onboard, manage, pay, offboard. We are running it thousands of times, for workers who do not yet have a row in any system.

The biggest open market in workforce software

The platforms that manage the workforce won by becoming the system of record: one place where every worker lives, gets paid, gets provisioned, and gets offboarded. They built that on the employee record, a human with a tax ID and a salary.

Agents do not fit that record yet. And that is not a problem for those platforms. It is the largest expansion of their market in a decade, sitting there for whoever builds it first.

Think about the size of it. Every company adding agents needs the same things it needed when it added people: a way to see how many it has, provision their access, track what they cost, govern them, and shut them off cleanly when they are done.

That is an entire management layer that does not exist yet, for a workforce growing faster than any human team ever could.

Why it is wide open

An employee’s salary is a payroll run: a fixed number, paid on a schedule.

An agent’s salary is nothing like that. It is a compute bill. Variable, metered by the token, spiking when the agent is busy, and it shows up in your AI spend instead of a payroll report.

Read that again. The compensation for the digital workforce is a finance object, not an HR object.

So the platform that wins this is not necessarily the one that owns the human workforce today. It is the one that can see what the agents actually cost and control it, and that sits close to the money.

What the winner gets to build

“How many people do you employ” stops being a complete picture of a company’s size. The real measure becomes how many people, plus how many agents, and what the agents cost.

Fully-loaded cost per agent becomes a number every operator tracks, the way they track cost per employee today, except an agent’s cost can triple in a week, so it needs watching in real time.

The org chart itself changes shape. The next version has humans and agents on it: the agents report to someone, cost something, and get reviewed and replaced. Someone is going to build the system that runs that workforce, the way the payroll platforms built the system that runs the human one.

The companies and platforms that win the next decade will be the ones that can answer “how many agents do you run, and what do they cost” as fluently as they answer “how many people do you employ.”

Almost nobody can yet. That is not a gap to worry about. It is the opening.

Whoever fills it first owns the workforce of the future, the half that runs on tokens.