You Can Tell Who Is Actually Reading.

For most of my career you could tell who had done the work by what they produced. The person who showed up with the thorough document had clearly gone through it. The polish and the thinking came bundled together, because you could not make the polished thing without doing the thinking first. That bundle is broken now. In a room where everyone has AI, everyone can produce the polished thing, the long document, the clean summary, the confident deck, and it all arrives looking finished whether or not a single human mind actually went through it.

So the signal moved. It is no longer in the output, because the output is cheap and uniform and tells you nothing about who understands it. It moved to one specific place, and once you learn where to look, you cannot stop seeing who has it and who is performing it.

The person who asks the one question that only makes sense if they read past the summary. Not a question about the shape of the thing or its general direction, those you can ask from the headings alone. A question about a specific claim three layers deep, the kind you physically cannot form unless you were actually down there reading it. That question is almost impossible to fake, because faking it would require having done the reading it is proof of.

The person who points at a particular line and says it is wrong, or asks why it is phrased the way it is. Specificity is the tell, because specificity has a cost. You can praise a document you skimmed in complete safety. You cannot object to line 14 of it, by number, without having read as far as line 14. The objection carries its own evidence.

The person who brings the thing the document did not say. They noticed the absence, the missing case, the question that should have been addressed and was not, which means they held the whole thing in their head long enough and completely enough to see the shape of what was missing from it. That is the strongest tell of the three and by far the rarest, because it requires not just reading the thing but modelling what a complete version would have contained.

What looks like engagement and is not. Praising the structure, because structure is exactly the part you can assess from a three-second glance, so praise of it proves only that you glanced. Agreeing quickly, because agreement is free and conveniently ends the conversation before anyone has to demonstrate they read anything. Restating your own summary back to you as though it were an independent thought, which tells you precisely one thing: they read the first paragraph and the section headings, the same parts a summary is drawn from.

None of these are bad faith, and that is what makes them dangerous. They are what reading-the-shape looks like from the outside, and reading-the-shape looks almost identical to reading. The whole difficulty is that the counterfeit and the real thing are nearly indistinguishable at a glance, and a glance is usually all you get.

Because the scarce thing in any room is no longer who can produce, production is handled, production is cheap. The scarce thing is who actually understands what was produced. The tell is how you find those people, and they are worth finding, because their "looks good" is a thing you can bank and their "wait, what about this" is worth more than a week of green dashboards. In a room full of finished-looking output, the person who read it is the only one who can tell you whether any of it is true.

So in your last review, be honest: who asked the question that proved they had read it, and who just admired the formatting?