The Substitution Narrative Just Failed Its Own Audit

Roughly 28,300 mass-layoff filings landed in New York during March 2026. Not one of them cited AI as the cause. That single number, surfaced by Kevin Indig in Growth Memo, is the quietest refutation of the loudest sales pitch of the decade: buy the model, remove the people, book the savings.

The substitution narrative was always a claim about the labor market. Claims about the labor market can be audited. When you run the audit, the story does not hold.

The Data That Refuses to Cooperate

Start with the filings. Mass layoffs in the United States carry a paper trail through WARN notices and state filings, and those filings name causes. If AI were displacing workers at the scale vendors imply, March 2026 would have produced a wave of AI-attributed cuts. It produced none worth counting.

The Yale evidence goes further. A team studying Current Population Survey data across 33 months found that measures of AI exposure, automation, and augmentation showed no sign of being related to changes in employment. Thirty-three months is long enough to catch a structural shift. The shift is not there. (These Yale figures are relayed secondhand through Growth Memo, so treat the exact framing as reported rather than read directly from the source paper.)

Then there is what people actually do with the tools once they have them. Microsoft’s 2026 Work Trend Index, drawn from 20,000 users, identified a 16% segment it calls Frontier Professionals. Eighty percent of that segment produce work that was impossible a year ago. Fifty-eight percent of all users in the study produce genuinely new output. That is the signature of augmentation, not replacement. People are not being subtracted from the equation. They are being multiplied across more work. (The Microsoft figures also reach us through Growth Memo’s summary.)

The executives confirm the direction. An NBER study of 750 leaders found the highest AI return on investment came from productivity growth, not labor cuts. The companies extracting real value are growing output per person, not shrinking the denominator.

What Substitution Costs Before It Fails

The substitution bet is not neutral while you wait for it to pay off. It carries a price the spreadsheet never models.

Klarna ran the experiment in public. It replaced human support with AI, announced the savings, and then reversed course, citing lower quality. The reversal is the tell. A company that loudly removed people had to quietly bring judgment back, because the work the AI absorbed turned out to need a human in the loop after all.

Trust is the other line item. Seventy-one percent of workers report fearing replacement. A team that believes the plan is to delete them does not lean into the tools that would make them faster. They hedge, they hoard knowledge, they wait. The substitution story is self-defeating: it produces the exact behavior that prevents the productivity it promised. You cannot frighten a workforce into augmentation.

So the audit returns two findings. The macro data shows no replacement at scale. The micro data shows that pursuing replacement degrades the very output you were trying to cheapen.

The Companies That Are Winning Measure Something Else

The firms capturing value are not counting heads removed. They are counting output produced by a human-and-agent team and watching that number climb. This is the measurement we have argued for repeatedly: judge the team, not the model, and judge it by what it ships per person.

We made this case before the March filings existed. In The Pinhole View of AI Value we showed that headcount reduction is one of four value levers, and the least durable one. Cost-cutting through substitution hits a ceiling the moment institutional knowledge starts to evaporate. The other three levers (deferred cost, more revenue, earlier revenue) only open when people stay and do more.

The same logic anchors our work on measuring the team, not the model. A model benchmark tells you what a system can do in isolation. A team benchmark tells you what your people plus that system actually deliver. Only the second number predicts whether the investment pays off.

The March 2026 data is the empirical floor under those arguments. We asserted the thesis. The labor filings now supply the proof.

Why the Narrative Persists Anyway

If the data is this clear, why does substitution still dominate boardrooms? Because it is the easiest number to model. Headcount reduction shows up cleanly in a pitch deck. Augmentation shows up as a messier story about capacity, throughput, and new revenue that did not exist before, none of which fits a single line on a slide.

This is the same distortion we documented in AI washing and unverified productivity: companies announce AI-driven cuts to signal modernity, then quietly discover the productivity was never measured and often never arrived. The announcement is a marketing act. The audit is an accounting act. They rarely agree.

There is also a trust dimension that compounds the failure. We covered it in agency, not agents: the value comes from giving people more agency through AI, not from replacing their agency with an agent. A workforce that trusts the tools will use them. A workforce that fears them will not. Substitution framing manufactures the second condition.

Do This Now

Replace the replacement question. Before your next AI investment review, swap the headline metric. Instead of asking “how many roles can this remove,” ask “how much more can each person ship with this in hand.” Define the human-plus-agent output baseline first, in concrete units (contracts reviewed, tickets resolved, features shipped), then measure the lift after deployment.

If the only number your business case produces is a headcount reduction, you are running the same bet that 28,300 March filings declined to make and that Klarna had to unwind in public. Build the augmentation baseline before you build the case. The data is now on the side of the teams that measure output per person, not heads removed.

---

This analysis synthesizes Stop trying to replace people with AI (Growth Memo / Kevin Indig, June 2026), which relays figures attributed to Yale (Gimbel et al.), Microsoft’s 2026 Work Trend Index, and an NBER survey of 750 executives.

Victorino Group helps teams build the augmentation baseline that the substitution spreadsheet cannot see. Let’s talk.