The canary that won't mean-revert

A June newsletter passed along an alarming stat: New York Fed data showing 22-to-25-year-old graduates facing the steepest job-finding difficulty on record, with software roles diverging as AI eats entry-level work. Two things are wrong with that, and the way they're wrong is the actual story.

The measurement that best supports the AI story is ADP's payroll ledger — the same data that cuts the paychecks — not the survey chart that went viral.

The strongest evidence for the thesis isn't the Fed report at all. It's a separate Stanford study by Erik Brynjolfsson's group that tracks ADP payroll records — the same data that cuts the paychecks — for young workers in jobs most exposed to AI. Since late 2022 their employment has fallen about 13 percent relative to everyone else; young software developers specifically sit roughly 20 percent below their peak. The June 27 update extends the numbers to April 2026, and the decline hasn't bounced back. It has kept compounding at about half a percentage point a month for nearly four years running.

The Fed chart the note actually cited, meanwhile, is the shaky one. A published critique shows its viral major-by-major unemployment rates rest on tiny survey sub-samples: the computer-engineering figure has a 95-percent confidence interval running from about 4 to 11 percent — wide enough to be Great-Recession-bad or pre-pandemic-fine, you genuinely can't tell. Broader data shows roughly 90 percent of recent grads employed in 2023, above the entire pre-Covid decade. Not a record low.

So the same claim is simultaneously better-supported than advertised and partly debunked, depending on which dataset you reach for — and economists still argue whether AI, the retreat of remote work, or a glut of CS graduates is driving it. What survives the sorting is narrow but sturdy: for the youngest workers in the most automatable jobs, the hole opened after ChatGPT and, four years on, has not filled back in.