Worth reading · August 2026 · Differential leads

The canaries are twenty-two.

Payroll records for millions of workers through June 2026: no economy-wide displacement, and workers aged twenty-two to twenty-five in the most AI-exposed occupations sit 19 percent below where they would be. It runs through hiring, not firing. Experienced workers show no gap yet. That last word is the long clock.

The question

Is what I sell still worth selling? The Differential question is what the value of me doing this is when everyone can. The premium for being early is, by definition, temporary. Close behind it the Identity question: what is my value in this arrangement, and is what I bring still something I recognize as mine.

This paper is not about me. It is about the people who would have been hired into the seat I sat in at twenty-two. That is the Differential question read from the other end: not what I am worth now, but what the entry to this work is worth, which is where the premium was minted in the first place.

What they found

The sourceErik Brynjolfsson, Bharat Chandar, and Ruyu Chen, Stanford Digital Economy Lab, Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. First published August 2025; revised 12 August 2026 with payroll data through June 2026. Read the original →

Six facts, from a large payroll processor's administrative records covering millions of United States workers through June 2026. There is no evidence of widespread, economy-wide displacement. Employment of workers aged twenty-two to twenty-five in the occupations most exposed to AI stands 19 percent below where it would be had it kept pace with less-exposed peers, and experienced workers in the same occupations show no comparable gap. That divergence has widened steadily since the authors first documented it in August 2025. It operates primarily through reduced hiring of young workers rather than through separations. The declines concentrate in occupations where AI use substitutes for human tasks; where it complements workers, employment is flat or rising. And the adjustment is happening through employment rather than wages.

The result holds when technology firms and computer occupations are excluded, so it is not a story about one industry. The authors' own caveat, in their words: these are early, descriptive indicators, canaries in the coal mine, rather than causal estimates.

The three questions

Every research post runs the same three rows. Which of the nine it names, how good the evidence is, and whether the thing is being measured.

RowStatusEvidence
NamedWhich of the nine questions does this answer? Differential The paper reads the entry premium from the market's side. Identity second, because what the seat is worth changes what the people holding it call themselves.
EvidencedHow good is the evidence? inferred The authors say descriptive, not causal. The exposure measure is a model of which tasks AI can do, and the gap is read against a trend that did not happen. Strong, careful, and inferred by its own framing.
MeasuredIs this being read on the Data page? not readable No footprint of one practitioner reads a labour market. The nearest signals read what prospects ask me and how pricing moves, which is the premium from my side, not the entry from theirs.

What this cannot say:

  1. Cause. A hiring freeze, a rate environment, and substitution by AI look alike in payroll data, and the authors say so.
  2. Whether the gap closes when the twenty-two-year-olds are twenty-six, or follows them. The window is a year. A career is forty.
  3. Who is doing the work the missing hires would have done. Substitution is inferred from where the declines fall, not observed in anyone's output.
  4. What happens to experienced workers next. No gap yet is a reading, not a promise, and the authors do not make it one.

What I am doing about it

The status says what is missing. These are the moves, on my own work, that would supply it. Each is either read automatically off the footprint or asked of me on a cadence, and each costs something to keep doing.

  1. Auto · MonthlyRead what kind of questions prospects bring me. Capability asks, teach me the tool, against next-order questions, what should we build and who answers for it. The Differential signal reads the premium moving from one to the other.Costs: reading my own inbox as data, and finding the premium has already moved.
  2. Auto · MonthlyEngagement pricing over time, normalized to scope. A line I have a financial interest in reading generously, which is why it publishes with its definition.Costs: a number I cannot round up.
  3. Asked · AnnualWhat can I do today that I could not two years ago, and what fraction of it is still rare? The Differential question in the form only I can answer, once a year, in writing.Costs: an annual list that gets shorter.
  4. Auto · MonthlyHow am I describing my contribution, in bios and bylines? If the entry seat is closing, what I call the seat I hold matters, and my own public words are the record.Costs: reading my own bio as evidence, and changing it when the evidence says to.

What I keep

The seat, and what I owe from it. Nothing in the paper says the work is worth less. It says fewer people are being let in to learn it, and that changes what the people already in it owe to the ones who are not. The practice I keep is the one that develops the person, because the market has stopped doing it at the door.

What it costs to keep watching is a monthly read of my own inbox, and an annual list of rare things that gets shorter.

Their finding, revised August 2026; read by me against the nine dimensions, September 2026. Written with AI assistance; the reading is mine. — Clay

Want the nine readings taken on your own work?

The kit ships with the same instrument I am running on myself. Mentorship installs it with me in the room.

One list, no drip.

New Resources, Research, Data readings, and kit editions. This list is the only announcement I send.