Questions this post answers
- Have I measured what my AI pilots are actually returning?
- If I can’t put the AI ROI case on one slide, have I earned the cut?
- When the market discounts even Alphabet for unproven AI, why would my board trust my AI-justified layoff?
One of the biggest questions facing businesses is what AI is doing to employment. Anthropic’s economists built a measure of what AI is actually doing at work, rather than what it could theoretically do. They combined task-level feasibility with real-world usage data, giving full automation greater weight than simple job assistance.
The gap they found is that AI is doing about a third of the work it could plausibly handle today — roughly three to one.
That gap is the whole story for anyone planning their headcount.
Capability isn’t the constraint. Deployment is — legal review, workflow friction, verification, and integration with software that was never built for any of this.
So it shouldn’t surprise you that the displacement hasn’t arrived. Using Current Population Survey data, the same researchers found no systematic rise in unemployment among the most exposed occupations since late 2022. They note that a real white-collar shock should have shown up by now.
Yale’s Budget Lab tracker reaches the same conclusion using different data. No clear relationship yet between AI exposure and employment.
If you are building next year’s plan around AI-driven headcount reduction, you are planning around something that has not yet happened at scale anywhere.
But there is one place the data moves, and it’s the place you’re least likely to be watching.
Hiring of workers aged 22 to 25 into the most exposed occupations has slowed by about 14% since ChatGPT launched. Separate research using payroll data found a 13% relative decline for the same age group — driven by slower hiring, not by separations.
Nobody is being walked out. Fewer people are being let in.
A stable headcount number can hide a closed front door.
Last week I wrote about cutting the management layer where your next leaders are built. This is the same problem one rung lower. Junior work is where people learn the judgment you’ll need from them in ten years. If AI absorbs the tasks that used to train them and you quietly stop backfilling those seats, you won’t feel it this year. You’ll feel it when you go looking for a successor and the bench is short.
So the strategic planning question should be to name which of your entry-level roles are actually apprenticeships — and to describe what your company looks like in five years if you stop filling them.
Let’s talk.