If you can write the manual, you've already written the prompt

A company decides to “industrialize AI.” One person writes the prompt. Another copies the output into the CRM. Another adapts it to the brand template. Another reviews it. Another approves it. There’s a procedure manual, a prompt library, and a new org chart. It all looks modern. But if the process is so clearly defined that it fits in a manual, why does it need people to operate it?

This isn’t a caricature. A Workday study of British professionals found that one in four spends more than seven hours a week copying information between applications, reconciling conflicting outputs, and manually feeding context into AI tools ¹. Almost an entire workday acting as the cable between systems. The phenomenon already has a name: human middleware. And according to an MIT report, 95% of generative AI pilots have no measurable impact on the bottom line. The problem isn’t model quality. It’s that AI doesn’t adapt to the workflow it has to operate in ².

Labor economics has been explaining why for more than twenty years. David Autor, Frank Levy, and Richard Murnane defined a routine task as one that can be described through explicit rules, and showed that these are precisely the tasks machines end up taking over ³. In other words: if you can write the manual, you’ve already written the prompt.

I’m a telecommunications engineer, and I studied this history in college. Almon Strowger invented automatic switching in the late nineteenth century. The Bell System didn’t start using it in major cities until 1922, and defended its human operators for their “natural flexibility and intelligence” ⁴. More than thirty years of people connecting plugs by hand in a system that could already run on its own. Today we’d call that inertia. When we build teams to manually operate a process AI can execute from start to finish, we call it “industrializing AI.”

It’s not incompetence. It’s inheritance. We’ve spent more than a century and a half designing organizations for people to operate. The first modern org chart, which Daniel McCallum devised in 1855 for the Erie Railroad, was created to coordinate the people running the network ⁵. Thinking in terms of processes, standards, and repeatability is part of our cultural DNA. And AI doesn’t contradict that model. It takes it to its logical conclusion: the entire process can run inside the machine.

The operators didn’t disappear because the machine was better than they were. They disappeared because the system no longer needed them to function. The same is happening with AI, but across the entire organization. The industrial process still holds. The industrial worker doesn’t. So the question is no longer how to bring AI into people’s work, but what work deserves to remain human when following the manual no longer does.