Critical Thinking Is the First Skill AI Ate

The IBM Institute for Business Value surveyed 1,500 chief human resources officers and 8,800 employees, and the finding that crawled out of the data is this: 60 per cent of employees worry that AI is eroding their skills. Asked which skills, they named critical thinking first, with human judgment close behind. Fifty-seven per cent of the CHROs agreed that critical thinking is now one of the workforce's most important capabilities — a figure that would be reassuring if it weren't an admission of what's being lost. The release then went out over PR Newswire under a headline about how AI "puts critical thinking at the center of workforce priorities". Which is a bit like announcing that the fire has put your house at the centre of your accommodation plans.

Sixty per cent is the number. Not a fringe, not the usual doom-posting minority, but a clear majority of the people doing the work, reporting that the tool is making them worse at it. And the skill they name first is the one the sales decks promise the machine cannot touch. For three years the pitch has been identical: the model eats the routine, the human keeps the judgment. The humans have now written back to say that it's the judgment that's going.

Two weeks ago I wrote about the Pew Research poll in which 34 of 37 countries named *fewer jobs* as AI's most likely contribution to work. The industry's answer, then, was a reskilling pamphlet. Now we have the follow-up study, and the reskilling turns out to be the accelerant. Fox News has been touring a Meta-backed training site in Houston, part of something called America's Workforce Academy, where an ABC vice-president calls the effort "the largest workforce development investment in history". Watch it closely and you notice what the workers are actually being trained for: construction tied to the datacentre build-out. The reskilling is real. It is simply pointed at the thing doing the replacing.

Meanwhile, the Insurance Journal has the receipt. Claims adjusters are 98 per cent critical of AI in their Glassdoor reviews — the most hostile occupation measured — while sector employment is down 21 per cent year on year and entry-level adjuster roles are down 50 per cent. This is the controlled experiment the HR departments keep asking for. The promise was that AI would absorb the drudgery and leave the judgment; what happened is that it absorbed the bottom rungs where judgment was built, kept a smaller senior group to supervise the machine, and left those survivors rating the arrangement at 98 out of 100 on resentment. Glassdoor is not a scientific instrument, admittedly. Neither is a survey a vendor publishes about itself, and only one of the two shows 98 per cent agreement.

Note, too, what the CHROs did not say. Fifty-seven per cent of them put critical thinking at the top of the capability list, and not one of them, in the summary at least, proposed hiring more people to do thinking. Their budgets are measured in cost per head, and skill erosion is, from that ledger's point of view, not a liability but a saving: a workforce that needs the model is a workforce that cannot leave it. A study like this is not a warning shot aimed at the C-suite. It is a procurement document, and the finding sits in paragraph four because paragraph four is not where the buyer lives.

Here is the mechanism, and it is dull. Judgment is not a module installed at thirty-five. It is what remains after you have been wrong a few thousand times cheaply: the junior claim, the rough first draft, the filing nobody senior wanted. Delete the cheap wrongness and you have deleted the apprenticeship, then promoted somebody into a role requiring taste without noticing the taste isn't there. The 60 per cent is not nostalgia; it is arithmetic, done by the employees, while the executives commission studies about it.

The vocabulary deserves a moment's contempt. "Human in the loop" is not a job description; it is a liability shield. The loop requires a person to catch the model's error with less information, less time and less experience than the person who used to do that job properly. "AI-augmented workforce" means the tool does the interesting half and you sign for it. And a study about skill erosion, funded by a company selling the erosion, distributed by PR Newswire, headlined for the C-suite — the artefact is the finding. Put the damage in paragraph four and the word "priorities" in the headline, and you have written the entire genre in one document.

So: the fix, and it is unfashionable. Keep paying juniors to be slow. Preserve the bottom rung even when the model could clear it, refuse to let AI take every first draft, and measure erosion inside your own organisation instead of polling strangers about theirs. The firms that won't bother will discover around 2031 that they have a layer of senior staff with no successors and a workforce that trusts the model more than it trusts itself. When the bill arrives it will arrive as an incident, and the person holding the loop will be the one who never got to practise. Sixty per cent is a warning with an invoice attached; the CHROs already know, which is why the headline says priorities and not damage.