Somebody has to run it.

If you run a growing company, you have lived some version of this story. An auditor, a customer, or an insurer asks a hard question. You buy the tool that answers it. The demo is excellent, the dashboard is beautiful, and the first month feels like progress. Ninety days later the dashboard reports forty-one failing controls, the alerts route to an inbox nobody owns, and the person who configured it has a different job.

The tool did nothing wrong. A service desk triages nothing on its own. Compliance automation collects evidence that a person still has to read. Awareness training assigns courses that a person still has to chase. Access reviews produce lists that a person still has to act on. Each of these products is a workbench, and a workbench without a worker is furniture.

For a decade, growing companies were sold the bench and told it was the carpenter.

The honest fix was always people. Enterprises do exactly this. They put a service desk team, a security function, a compliance manager, and an access administrator on payroll, and in trained hands the tools become what the brochure promised. Below a few thousand employees the math never closed; a bench like that costs multiples of the tool stack every year, and it has to exist before the first ticket is answered. So growing companies made the only rational move available to them. They bought the dashboards, skipped the people, and carried the quiet cost of software that was never fully awake.

A whole market settled into that gap. Vendors could not sell you the people, so they sold you the promise that the product needed none. Every renewal repeated the promise, and every renewal, somebody at your company quietly worked an unpaid second job pretending it was true.

Then the arithmetic flipped. The repetitive center of operations work (reading the alert, assembling the evidence, preparing the access review, drafting the questionnaire answer) can now be carried by software at a marginal cost near zero. AI is the reason, and this is the one place in this essay where we will name it, because we do not sell it. It is how the economics work. It is what makes a staffed operations bench affordable at fifty employees when it used to require five thousand.

What matters is what a company does with an inversion like that. Every vendor in our categories faces the same choice: keep it as margin, or spend it on something customers can actually hold.

We spent it on people, on both sides of one platform. Operators work inside the same product you log into; there is no second, private system where the real work hides. Open any queue and a named person is behind it, with a schedule, a photograph, and the authority to finish things. When evidence gets filed, a person filed it. When an access request closes, a person closed it, in the same room you can walk into at two in the morning.

Others are arriving at the same reading of the numbers. Researchers at MIT, studying how enterprise AI deployments actually fare, found that most pilots produced no measurable profit impact, while deployments bought from specialized external partners reached production roughly twice as often as tools built internally.¹ And the most public experiment in removing the people outright has already run in reverse: Klarna, having automated its customer service and said so loudly, began hiring people back into it.²

The bill behaves the way the rest of this behaves. The structure is published, no usage meter runs inside it, and people are what you buy. The economics underneath are our problem to run, and we run them the same way we run everything else: where you can see it.