Episode Details

Source: Soren Kaplan, “AI Can Do the Work. It Still Can’t Replace the One Thing Companies Need Most,” Inc., July 27, 2026. Underlying study: Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” Stanford Digital Economy Lab, revised November 13, 2025, with the authors’ follow-up note of February 9, 2026.

The first supervisor I hired at Lansing Grand River was Dennis Boutwell. He did not arrive formed. He arrived willing.

What formed him was three years on a floor where the work could go wrong in front of him. A bad call at nine in the morning showed up at nine-forty as a stopped line and forty people standing there looking at him. Not a footnote in a quarterly review. Forty people, looking at him, while the cost ran. He had somebody over his shoulder for most of the first year. Less the second. By the third he was the one somebody else stood next to.

Dennis is now Director of Manufacturing Excellence at Advanced Drainage Systems.

Nobody hired that. Nobody could have. It was deposited into him a piece at a time by a system that guaranteed him a runway long enough to be wrong on, and a man standing there who could tell him which wrong mattered. Somebody paid for those three years, and it was not Dennis.

This week a technology recruiter told the Wall Street Journal that one experienced engineer can now deliver the output of an entire team. In Inc., Soren Kaplan asked the right next question. What happens to the inexperienced ones?

Then he answered it with three bullets aimed at a manager. Create stretch experiences. Build coaching into the pipeline. Measure judgment growth. Do it this week — pick one or two people who look at risk and hand them a hard decision.

Every one of those bullets asks a manager to spend real money building an asset no accounting system will record him building, in the same quarter that the technology which closed the gate is booked as a productivity gain on the line above his. He is being told to be generous with something the ledger says does not exist. That is not a coaching problem. That is a bookkeeping problem wearing a coaching problem’s clothes.

The entry-level job was never mainly a job. It was the formation mechanism. The senior engineer everyone wants to hire right now was made on that runway — production failures, bad architecture calls, teams that didn’t perform as promised. Firms are harvesting that capital and defunding the mechanism that produced it, in the same spreadsheet, and nothing requires anyone to notice.

That’s the argument.

The canary

The numbers are real and more careful than the headlines about them.

Stanford’s Digital Economy Lab, working from the payroll records of the largest payroll provider in the country, found roughly a 16 percent relative employment decline for workers aged 22 to 25 in the most AI-exposed occupations — and it holds after controlling for firm-level shocks. Older workers in the same occupations stayed flat or kept growing. The adjustment came out of headcount, not pay. The declines cluster where AI automates rather than augments.

Kaplan quotes 20 percent. That’s the raw peak-to-trough figure for young software developers, not the study’s estimate of the effect. Use 16.

The authors have also been straight about their own limits: interest rates don’t explain the pattern, but under the strictest controls the divergence only turns significant in 2024, and they say plainly they don’t think AI is the sole cause. The gap is still widening. No reversal.

So the causal question is live. Fine. It is also beside the point.

Whatever closed the entry gate — the model, the rate, the layoff, the reorg — the gate is closed, and there is no institution on earth whose job it is to open it.

It isn’t a tech story