The Day a Payroll Chart Stole the AI Layoff Narrative
By Sunday afternoon, the loudest voice in the AI-and-jobs debate wasn’t a founder on stage or a viral screenshot of a souped-up chatbot. It was a macroeconomist pointing to a humble payroll series and saying, in effect, look for yourself. Torsten Sløk, Apollo’s chief economist, declared there is “zero evidence” in the latest weekly ADP data that AI is causing net job losses. That single line, amplified by mainstream coverage, shifted the day’s conversation from anecdotes about cut-and-paste redundancies to the stubborn arithmetic of aggregate employment.
The claim behind the headline
Sløk’s note was short but pointed. If AI were already chewing up the labor market, you’d expect the weekly payroll data to flinch. It hasn’t. Instead, the hiring that accompanies an investment supercycle appears to be offsetting, and perhaps outrunning, any substitution effects. Companies are recruiting implementation specialists to wire new models into messy production systems. The data center build-out is demanding electricians who can tolerate 400-volt realities, civil contractors who pour foundations that don’t sag, network engineers threading fiber, and a small army of vendors shipping switches, transformers, and cooling gear. For Sløk, this is textbook rebound: as the cost of intelligence falls, usage scales, and so does the web of work around it—a modern echo of Jevons’ paradox.
That mechanism led him to a bolder near-term read: if AI is simultaneously pulling in labor and bidding up specialized inputs, it should show up as both stronger hiring and some stickiness in prices. He floated that May’s nonfarm payrolls could top the roughly 95,000 consensus—an unfashionably high number in a season of think pieces eulogizing “AI-killed” roles.
Why a payroll series became the main character
AI discourse has been dominated by micro stories: a support team trimmed here, a content mill slowed there, or a startup that can now do in hours what previously took a week. Sløk dragged the conversation into macroeconomics, where totals matter more than testimonials. His daily charts are breakfast reading on Wall Street and in C-suites; when he says the net jobs needle hasn’t budged, investors and operators recalibrate. By Sunday evening, the question hanging over the week wasn’t whether AI automates tasks—it was whether the Bureau of Labor Statistics would vindicate the idea that the build-out phase is labor-accretive.
Aggregate truths, local pain
“Zero evidence” is a phrase built for headlines and easily misunderstood. It refers to net effects in broad payroll measures, not to the lived experience inside specific industries or roles. Distributional shifts are real and can be sharp: a back-office unit that freezes vacancies rather than announcing layoffs, a media shop that relies on fewer production assistants, a customer service org that quietly caps headcount because deflection rates just improved. None of that disappears because the sum of payrolls is flat or rising. Aggregates can conceal churn and scar tissue; they also capture something else: the offsetting swell of complementary work that proliferates when a new general-purpose technology demands facilities, standards, integration, compliance, and continual tuning.
Attribution is messy—and that’s the point
Even if the totals look resilient, pinning a causal badge on AI is slippery. Independent research from the New York Fed two weeks prior noted weakness in job postings for AI-exposed occupations that predates ChatGPT’s late-2022 splash. In other words, some of what we’re watching now may be trend continuation, not a clean shock. Sløk’s choice of a high-frequency aggregate is both a strength and a limitation: it’s less prone to the storytelling vortex that swirls around layoffs, but it can’t disaggregate cohorts or prove motive. If AI is changing hiring behavior at the margin—especially through “hiring avoidance,” where roles quietly evaporate before they’re posted—those signals may bleed into the data slowly and unevenly.
The build-out economy and its inflation shadow
There’s a deeper macro angle in Sløk’s framing: AI as a capital deepening cycle that looks, for now, more like industrial construction than disembodied software. Power constraints, grid interconnect queues, transformer backlogs, and land near fiber backbones aren’t abstractions; they are constraints that summon labor and pricing pressure. If that’s the phase we’re in, the early macro signature is not sweeping job destruction but rising nonresidential investment, tight niches for skilled trades, and upward pressure where supply is slow to scale. Productivity gains may well arrive, but the first act in technologies like this is often infrastructure and integration—expensive, people-heavy, and inflation-prone until capacity catches up.
This week’s test
All of which puts an unusual amount of weight on normally staid releases. JOLTS on Tuesday will hint at whether openings in AI-sensitive categories are still firm. Friday’s Employment Situation will tell us if May payrolls deliver the resilience Sløk expects. A clear upside surprise would harden the thesis that, in the near term, AI is net additive to labor demand because the economy is still building, wiring, and adapting to it. A downside miss wouldn’t settle the argument either; measurement lags and sectoral crosswinds can blur a single month. But the frame has shifted: the question isn’t only “who got automated?” It’s “how big is the implementation economy, and how long does it dominate before substitution shows up in the totals?”
The uncomfortable equilibrium
For readers of this newsletter, none of this absolves the strategic imperative to redesign work. The aggregate can rise while individual teams are reconstituted, junior rungs are thinned, and entry paths narrow. That’s a productivity story with human consequences, even if the sum of jobs grows. But Sunday’s headline did something important: it reminded us that macro labor markets are not spreadsheets of tasks. They are dynamic systems where a plunge in the price of intelligence can, paradoxically, create a surge of demand for the people and materials needed to distribute it. If the ADP series is right, we’re still in that surge. The build-out is the job.