The day after the Mag-4 finished their April 30 earnings calls — the calls in which the word efficiency appeared 15 times — Valerie Capers Workman published a Fortune commentary with one of the cleaner contrarian thesis statements of the cycle.
The hyperscaler math driving cuts at Meta and Microsoft does not apply to the companies where most Americans actually work.
Workman has the credentials to make the claim land. She was VP of People at Tesla through the 2022 Autopilot cuts and the harder 2024 round. She is currently CHRO of Empower Pharmacy. She is the rare HR voice with first-hand experience of both tech-scale workforce reductions and mid-market healthcare staffing — exactly the two halves of the labor market she is now pulling apart.
Her math, in three lines:
- Meta: 8,000 layoffs starting May 20, plus 6,000 cancelled requisitions.
- Microsoft: voluntary buyouts to roughly 8,750 US workers, the first companywide separation program in Microsoft’s 51-year history.
- Combined Mag-4 2026 capex: approximately $650 billion — the largest single-year private-sector capex pool in the history of capitalism.
Then the punchline: per Bureau of Labor Statistics payroll data, more than 80% of US workers are employed at companies that are not hyperscalers. Regional healthcare systems. Mid-market manufacturers. School districts. Community banks. Regional law firms. State agencies. The small and mid-size businesses that form, in Workman’s phrase, the actual backbone of the American economy.
Why the math doesn’t generalize
Workman’s central observation is one LostJobs has been circling for two weeks but had not heard from anyone with her resume.
The hyperscaler unit economics are alien.
Meta serves 3.5 billion monthly active users. Automating a single workflow at Meta ripples across an audience equivalent to nearly half the planet — which means the marginal value of replacing one human-hour with one AI-hour is multiplied by an audience that no other employer in the world has. Microsoft, Amazon, and Google have similar leverage in different shapes.
A regional healthcare network in Pennsylvania has roughly 12,000 employees and serves roughly 1.5 million patients. The leverage of replacing a clinical scheduler with a calendar agent is the leverage on 1.5 million patient interactions, not 3.5 billion. The math does not arrive at the same answer.
Aaron Levie, the Box CEO, made the same argument from the venture-capital side last week. Workman is making it from the chief-people-officer side this week. The convergence is the news.
The half she does NOT dismiss
The piece is not a “stay calm, nothing is changing” piece. The honest part of it is that Workman concedes one thing the hyperscaler-skeptic crowd has been under-weighting:
Workplace surveillance and AI-usage tracking will generalize. The tools that capture keystrokes, monitor application usage, and track AI prompt activity are already commercially available — any mid-market company can license one today.
Meta is grading employees in performance reviews on their AI use. Meta has rolled out internal software that captures employee keystrokes, mouse movements, and click locations to train, in the company’s own language, the next generation of our AI models to use computers.
The disturbing version of this story is not “AI will fire you.” It is “AI vendors need training data of office workers doing office work, and your employer will sell them yours.” That distinction matters because the first story has a clear villain (the layoff memo) and the second one does not (the keystroke logger has been in the IT-spend line item for two budget cycles).
Where this puts LostJobs’s running ledger
Three thoughts.
One — the chorus is now four-deep. Aaron Levie, Falk and Tsoukalas, the Washington Post on Mag-4, and now Workman are all saying versions of the same sentence: the 100,000+ tech layoffs of 2026 are mostly hyperscaler-specific and will not generalize to the rest of the economy. When four credible voices from four different vantage points (VC, academic, journalism, HR) reach the same conclusion in 14 days, that is no longer one person’s take. That is a thesis.
Two — the panic itself has a constituency. Companies attributing layoffs to AI get a free pass on a 2021–2022 over-hire that was their own decision. Cable news gets a story that performs. Doomer-influencer Twitter gets engagement. The Mag-4 themselves get a productivity narrative that makes the capex line easier to explain to the buy side. The number of stakeholders profiting from the panic is large enough that the panic will continue regardless of whether the data supports it.
Three — the surveillance subplot is the actual sleeper risk. A regional healthcare network is not going to lay off a billing clerk because Claude can do it. A regional healthcare network is going to install Aware, Veriato, Teramind, or one of the half-dozen behavioral-analytics platforms now bundled into mid-market HRIS suites — because those platforms now also sell the “AI-usage scorecard” feature that Meta’s piloting at scale. The 80% of US workers Workman is reassuring will not be replaced. They will be measured. That is a different problem and arguably a worse one.
The dry coda
It is unusual for the most credible voice against an AI-jobs panic in a given week to be a former Tesla HR executive. It is more unusual still for that voice to be making the same argument as a Box CEO and two academic economists at the same time. The signal is that the panic narrative has run far enough ahead of the data that the people who actually count layoffs for a living are now publicly pulling on the brake.
The signal is not that nobody loses their job in 2026. The signal is that the people losing their jobs in 2026 work at four specific companies, and the rest of the economy gets the surveillance instead.
Workman’s last line is stay calm. It is good advice for the 80%. It is less applicable to the 20% she used to manage at Tesla.