DS Cohen & Associates
My grandson asked me a question I couldn’t answer the way a grandfather wants to. He and his housemates are a year from graduating and wanted to know whether anything would be left for them when they got there. Their worry was not that AI would take a job they were after. It was that nobody would open the job to them at all.
I told them I would look at what the evidence shows rather than what the headlines shout.
Erik Brynjolfsson, Bharat Chandar and Ruyu Chen have updated their study of AI and American employment using ADP payroll records for millions of workers through June 2026. Real payroll, not job postings and not survey sentiment. There are three findings that really matter.
Nobody is being fired. Separations for young workers in the most exposed occupations fell as fast as everywhere else. The entire divergence comes from hiring that never happened, which is exactly what my grandson and his housemates were afraid of.
The damage lands on one group. Employment for workers aged 22 to 25 in the most exposed occupations now sits 19% below where it was against their less exposed peers. Older workers show no comparable gap. A year ago the gap was 15%. And it continues to widen.
The outcome tracks how AI is used, not whether it is used. Where organisations use it to automate human tasks, employment for the younger demographic drops. Where its primary use is to complement people, employment is flat or rising. The technology hasn’t changed, but the outcome certainly has. The difference is a decision somebody made in a meeting.
Which occupations are most exposed? Accountants, customer service representatives, human resources specialists, software developers, market research analysts, loan officers, paralegals, bookkeeping clerks, tellers, receptionists and executive assistants. That is the professional office. It is very nearly a description of what a university degree is sold as preparation for. And human resources specialists sit in the top quintile, so those of us in this profession are not commenting from the sidelines. We have spent two years telling recruiters they are safe if they use AI. The data invites a harder question: who is exposed?
Now the part that is not for the graduates.
If you are a chief people officer or a hiring manager, you are not cutting people. You are quietly declining to add them. Nobody appears on a severance line. No announcement is required. The saving lands in this year’s numbers, and every report on your desk tells you that you are running a disciplined organisation. Nothing in the report is wrong yet something is still happening to your organisation, and the report cannot show you what it is.
So, look at what you actually removed.
Tacit knowledge is not issued at hire. It is earned by doing the codified work first badly, then adequately, then well, in front of someone who corrects you. The junior analyst who builds the deck learns to read the room from the senior colleague who marks it up. The junior developer who writes the tedious function learns architecture from the review. The tedious work was never the point. The correction was the point. Automate the task and you have not saved a salary. You have closed the only classroom your organisation runs.
Take out the bottom rung and the ladder still looks intact for a while. Your senior people are still there. Your middle is still there. Nothing breaks this quarter, which is exactly why nothing gets questioned. It breaks in 2032, when the people you would have promoted were never hired in 2026, and every experienced person who walks out takes the corporate knowledge with them because there was nobody standing beside them to receive it.
So, answer this in your next workforce planning meeting. If you stop hiring at the entry level in 2026, who exactly do you promote in 2032?
If the answer is that you will buy experienced talent from the market when you need it, look hard at the market you are counting on. Every organisation making this decision is planning to draw from the same shrinking pool, and not one of them is refilling it. That is a bidding war. Watch the NBA in free agency. When a handful of teams chase the same scarce player, the player sets the price, and the guarantees climb far past anything the work is even capable of producing. You have not cut your labour cost. You have deferred it, compounded it, and handed the leverage to the people you will need most.
Strip that strategy of its language and here is what remains. Automate the foundation, keep the roof, and hope the building holds.
I have worked with organisations on five continents, and the ones still standing decades later were never the ones with the cleverest cost line. They were the ones that kept building the bench when it was inconvenient. Efficiency is not longevity. You can be brilliantly efficient right up to the point where nobody left in the building knows how the place works.
None of this is a technology problem. The tool decided nothing. The firms in this same study that use AI to complement their people show flat or rising employment, and they run the same software you do. They made a different choice.
So put people first. Do not make your people servants to the technology. Use the technology to make your people more capable and keep hiring the ones who are not capable yet, because that is the only place capable people have ever come from.
Choose well. A young person at a kitchen table, and a great many like him, are waiting to find out what you decided.
David S. Cohen is the author of “Selecting the Best: Fostering a Workplace Driven by Values for Lasting Success,” amplifies each of the points of this article using a combination of research and anecdotal stories. The appendix contains sample behavioural interview questions. Selecting the Best is available on Amazon and other online book sellers.
