AI as an x-ray

A few years ago I put together a report every quarter. Thirty slides, about two weeks of work, tidy charts. At one point I tried to find out who actually read it and never got a clear answer. I kept making it anyway, because that was how it was done, and because two weeks of work looked like proof that the thing was serious.

Today the same report takes four minutes with AI. And that is where the real story starts. When work gets close to free, everyone suddenly sees that nobody needed it. What bothered me was not the machine. It was those two weeks per quarter.

What the x-ray is

AI does not break the bone. It shows the fracture that was there yesterday, and the year before that.

The Danish anthropologist Dennis Nørmark and the philosopher Anders Fogh Jensen called this pseudowork. It is activity that looks like work, fills the calendar, produces files, and changes nothing in the outcome. It used to be easy to hide, because it cost time, and time looked like proof.

Three layers show up in the picture. First: documents nobody reads, the reports and summaries and roll-ups that travel from one folder to the next. Second: meetings that exist so somebody feels informed. A machine now writes their content in a minute, and then you can see how much of it there ever was. Third, and the least comfortable: intermediate steps. Rewrites, reformatting, moving things from one template into another. They exist not because they were needed, but because the alternative was hard for somebody.

The test is simple. Take any recurring piece of work and ask what would happen if this once it simply did not get done. If the answer is "somebody would call", the work is real. If the answer is "nobody would notice", you have your diagnosis.

Why the picture is unpleasant to look at

An x-ray does not heal. It only shows, and it shows without any tact.

Pseudowork is usually not one person's fault. It is what is left of an old system, and sometimes it is somebody's entire job description, handed to them on the day they were hired. So the answer is almost never "let that person go". The answer is to hand them a real problem, and real problems are not in short supply.

There is one more easy trap. AI is excellent at producing pseudowork faster. If you used to write one unnecessary document a week, you can now write ten, and they will look better. The x-ray helps only if somebody actually looks at the picture.

Where I use this framework

It opens my talk at HR Week Lithuania in September. Not the tools, but the question of what the tools light up.

When people ask me where to start with AI, I do not look for a new tool. I ask them to open last week's calendar and mark everything that nobody would have noticed if it simply had not been done. The list is usually an uncomfortable length. Mine was longer than I expected.

In my own work I keep one rule: if AI does some task very well and very fast, that is not a win, it is a question about why that task was in my week at all.

One thing I did not understand right away, and it cost me one ruined conversation. You cannot show somebody the picture before you have something for them to do next. Otherwise it is not a diagnosis, it is an accusation, and the conversation ends right there. It took exactly one time for me to learn that, and I still regret it.

AI did not steal that job. It just showed there was no job there.