Two weeks ago, companies across Australia opened a new financial year with freshly signed budgets. If yours runs on a calendar year, you are halfway through a plan built on the same foundation. Underneath nearly every line of it sits an assumption so old nobody says it aloud: to do more, you add people, and people cost money.
That assumption has held for as long as there have been companies. This issue is about the fact that it has started to give, and about the instrument that breaks when it does. I have just published a research paper on the full pattern, and I will point you to it at the end. Here is the argument in one sitting.
Because the unit of work has always been an employed person, three things are chained to one another: the output, the people who produce it, and the cost of employing them. Want more output, hire more people, carry more cost. Need to cut cost, shed people, lose the output they were producing. There has never been a lever that moved one without moving the rest. I call this the headcount chain.
The chain is also the only reason finance can plan at all. Capacity is abstract and arguable, so finance counts the thing that stands in for it: headcount. The budget you signed is, underneath, a headcount plan. That substitution was fair for a century, because one person was one bounded lump of capacity. Count the people, know the capacity.
Copilots did not change this, and it is worth being precise about why, because the copilot is the example everyone reaches for. A copilot makes a person sharply more productive, but the work still runs through the person, so capacity is still capped by their hours. Five engineers with copilots are five engineers. You plan them as five. The chain holds.
An agent is different in the one way that matters: it is a unit of capacity that is not a person. You commission it by the task or the volume rather than employing it by the block of time, and it scales without a hiring request. There is a clean test. To double a team's output, must you roughly double the people? With a copilot, yes. With an agent, no. That no is the chain breaking.
Here is the part I want you to carry into your next planning conversation.
When a team rebuilds its work around agents and doubles its output with the same thirty people, the headcount line still reads thirty. Nothing on it moved. And it has stopped predicting anything. Headcount and capacity have come apart, and the master instrument finance steers by now returns a number that no longer means what everyone in the room assumes it means.
The hiring freeze tells the same story. Freezing hiring used to freeze capacity and cost together, because they moved as one. When the capacity comes from agents on a separate variable line, the freeze grips nothing. Capacity climbs, the agent bill climbs with it, and not one name is added to the payroll. The lever is still bolted to the wall. It no longer connects to anything.
And before the relief arrives, the cost comes back. Agents are probabilistic, so you wrap them in review, retries, human oversight, rework. Every layer attaches cost back to the output, and none of it shows up as headcount. Klarna lived both chapters of this in public: the assistant doing the work of seven hundred service agents while headcount fell, then, a year later, the admission that it had cut too deep and the rehiring of people for the work that needed judgement. The chain breaks, then re-forms where you are not looking. Whether it re-forms above or below the cost of the person it replaced is the whole game, and almost nobody is measuring it.
Find the one team in your company whose output has clearly run ahead of its headcount. Put a single honest measure of what it produces next to the headcount you are still planning it with, and watch the gap. Then watch what your existing controls do with that gap: the freeze that no longer touches the agent spend, the budget line straining to hold a cost it was never shaped for.
That exercise, the full argument, the Klarna and Accenture evidence, and the five instruments a board builds when the single dial stops working are in the paper. It is free to read and cite, no gate:
The Headcount Chain: How agentic AI breaks the instrument every company plans with https://aideliverydiscipline.com/research/the-headcount-chain.html
One thing before you close this issue.
Is there a team inside your company whose output has clearly run ahead of its headcount over the past year? Reply ([email protected]) with one word. Yes or no.
I read every reply myself. I am building a picture of how far the break has actually travelled, and everyone who replies gets the aggregate when it is complete.
The chain held for a century because the unit of work was a person. The unit is changing. The instruments have not.
New issues arrive by email first. Subscribe here.
