There was a certain comedy to it. I took a few passages from texts I had written — essays from this site, pages from a novel manuscript — and fed them into a tool designed to detect machine-written prose. The answer came back within seconds, and it was not ambiguous: entirely AI-generated.
My first reaction was not discomfort. It was recognition. The answer is correct in one precise and very limited sense. And that sense is worth spelling out, because it says something about the entire shift this site exists to describe.
What a detector actually measures
Tools of this kind do not look for ideas. They cannot. They look for traces: statistical patterns in how words follow words, how sentences are built and how a text breathes. A language model has a recognisable cadence, much as a machined part carries marks from the tool that shaped it. Those are the marks the detector sees.
In other words, it can say something about how a text was formed. It cannot say where it came from. That distinction matters. "100% AI" means roughly this: these sentences were assembled in collaboration with a language model. That is true. It does not mean the ideas, judgements, selection or responsibility belong to a machine.
How the texts are actually written
I carry roughly thirty years of manufacturing industry with me. That is where the perspective comes from: the factory floor, not the boardroom. An essay begins in an observation, an irritation, a contradiction I cannot make sense of. I decide what the text is about, which thesis it argues, which examples should carry it and where it should land. I decide what is true enough to print and what should be cut because an experienced colleague would otherwise roll their eyes.
Then I work with a language model. It helps me formulate, test, turn and polish. It proposes; I choose. It writes a sentence; I tear it down. That is roughly how I use AI in consulting work too: not as a replacement for judgement, but as a tool that lets judgement reach further. The result carries the machine's cadence because the machine was part of holding the pen. The thinking behind it is mine.
In short
The detector is right: the texts here are written in collaboration with a language model, and it shows.
The detector says nothing about: where the ideas come from, whether the argument holds or whether a conclusion is worth believing.
I am not hiding that. On a site about this very shift, silence would be the only truly dishonest choice.
Why I say this openly
It would be easy not to. Detectors are not infallible, and polished formal prose — exactly the kind I try to deliver — is often what they flag. I could lean on that uncertainty and say nothing.
But that would miss the point. AI-skiftet is about how this technology changes work, writing and society. Pretending that the site's own texts are produced in some other way would undermine the premise. I use the tool I write about. Admitting that is not a concession; it is the line.
What actually damages trust is not that AI was involved. It is the gap between what someone says about how a text came into being and how it actually came into being. That gap does not exist here, and I intend to keep it that way.
Scrutinise the right thing
If you run a text from this site through a detector, it will probably be flagged. Then you know the tool recognises the language model's fingerprints, and you know something you could already read on this page.
The interesting question has never been whether an individual sentence is machine-shaped. That question is technical and, frankly, rather dull. The interesting question is whether the argument holds. Whether the examples carry weight. Whether the conclusion is worth taking with you. A reader can judge that — and judge it far more sharply than any detector.
By all means, read critically. Just scrutinise the right thing.