On Wednesday April 29, the question that was sitting in every HR slack channel in the country was the one Box CEO Aaron Levie had just answered the night before in Fortune: if Silicon Valley has cut 100,000 jobs in 117 days and called 47.9% of them “AI-driven,” when does that wave land in the rest of corporate America?
Levie’s answer is, roughly, it won’t. Or at least it will land much smaller, much slower, and look almost nothing like a layoff press release. Same Wednesday, Fortune ran a second op-ed from Apollo chief economist Torsten Slok arguing the same conclusion via a 160-year-old British paper on coal demand. Two professionals who have to be right for a living, on the same morning, both pushing back against the dominant LostJobs-style narrative.
Worth taking seriously. Worth taking apart.
What Levie actually said
Levie’s framing on the a16z podcast cycle and in his own LinkedIn posts over the last fortnight was the same line, repeated four ways: “My job these days is just bring reality to the valley, and then bring the valley to reality.” The Wednesday Fortune piece was the long-form version of the second half of that sentence.
His three claims, paraphrased:
- Tech workers are uniquely automatable right now. They are engineers. Their output is code. Their tools are flexible enough to invoke an AI agent inside the IDE. The feedback loop between “AI wrote a thing” and “the thing compiled and the test passed” is seconds. No accountant, claims adjuster, regional-bank loan officer, or hospital case manager has that loop. Theirs runs in days, gated by data living in a 1998 mainframe, gated by regulators, gated by the consequences of being wrong.
- The “last mile” eats most enterprise AI rollouts. Per the Q1 2026 enterprise AI survey Levie cited, 72% of enterprises have at least one AI workload in production — but only 28% describe their AI adoption as “mature.” The other 44% have a thing running in a corner that nobody trusts to run a thing that costs money to be wrong about. That gap is what’s going to take three to five years to close, and during those three to five years the headcount line on those companies’ P&L stays roughly flat.
- Jobs shift, they do not vanish. Per the April 16 Benzinga writeup: “We’re going to use AI to accelerate output in one area, and then eventually you run into a new bottleneck somewhere else in the process that still requires humans.” Cap the partner with one AI agent and the bottleneck moves to associates; cap associates with the next agent and the bottleneck moves to client onboarding; cap onboarding and it moves to compliance. Each move keeps the headline org-chart roughly the same shape.
What Slok actually said, on the same day
Apollo’s Torsten Slok came in from the economics side. His April 28 argument is built on Jevons paradox — William Stanley Jevons’s 1865 observation that when steam engines got more efficient at burning coal, total UK coal demand went up, not down, because cheaper coal-power unlocked new uses faster than efficiency reduced old ones.
Apply to legal services: if AI cuts the cost of producing a contract by 80%, the universe of disputes worth litigating triples, the universe of small businesses that can now afford a lawyer quintuples, and gross billable hours go up. Apply to accounting: a 90% cheaper audit means small private companies and family offices that never bought one now buy one. Apply to medical imaging: faster reads create demand for more imaging.
Slok’s headline: “AI will create more lawyers and accountants, not fewer.”
The combined Levie + Slok read is: stop modeling the next decade as the tech-layoff wave at corporate-America scale. The wave will look nothing like that.
Where the argument has holes
Both arguments are stronger than the Twitter version of the LostJobs thesis they’re pushing back against. Both also have specific weak points worth naming.
On Levie’s “tech is uniquely automatable” claim: The class of work being automated in tech right now is not “all engineering work.” It is junior and mid-level individual-contributor coding plus L1/L2 customer support plus first-pass legal review plus first-pass financial reconciliation. That class of work exists in every Fortune 500 company under different titles. Salesforce already cut 4,000 support roles by routing tier-1 to Agentforce. Citi already cut staff who were literally in its “AI Champions and Accelerators” program. The six big banks shed ~15,000 in Q1 while booking record $47B in profits. Banks are not tech companies. The wave already crossed the line Levie says it won’t cross.
On Slok’s Jevons argument: Jevons paradox is real, but it requires the demand curve for the underlying service to be elastic enough to absorb the productivity gain. Coal in 1865 had functionally infinite latent demand — every factory in Britain wanted more steam. Demand for white-collar back-office labor in 2026 is not infinite. A bank only needs so many monthly statements generated; a hospital only needs so many claim packets coded; a Fortune 500 only needs so many Q3 earnings decks built. When AI cuts the time-per-task by 80%, in industries with inelastic demand you get fewer people doing the same task, not more people doing five times the work. Slok’s argument works for creating new categories of legal/accounting service; it doesn’t work for the existing back-office headcount on the existing P&L.
On the “Wall Street loves it” data point both pieces dance around: Of 28 tech companies that announced AI-related layoffs this year, 17 saw their stock prices rise on the day. That number is the actual operating reality for any Fortune 500 board sitting in a Q2 2026 strategy offsite. If announcing AI-driven headcount reduction is the move that gets the stock a green day, the move gets made — independent of whether Levie’s “last mile” thesis or Slok’s Jevons argument is correct in the long run. Boards optimize for the next earnings call, not the 2030 elasticity-of-demand curve.
What we think the honest synthesis is
The Levie + Slok read is roughly correct on the time horizon for true displacement. It is roughly wrong on the time horizon for the announcement of displacement. Those two clocks are running at different speeds, and the gap between them is exactly where the next 12–24 months of headlines live.
Real displacement — the kind that actually shrinks the headcount on a public-company 10-K and stays shrunken — requires the “last mile” to be solved. Levie is right that for a regional bank or a Cleveland Clinic that takes years.
Announced displacement — the kind that shows up in the trackers, in the buyout offers, in the “we’re streamlining around AI” memos — requires only that the CFO believes Wall Street will reward the framing. Wall Street has already told 17 of 28 CFOs the answer is yes. The next round of Fortune-500 CFOs sitting in a Q2 strategy review will see that data.
So the LostJobs prediction for the next two quarters: the count of corporate-America-ex-tech AI-attributed layoff announcements rises sharply through Q3 2026, even though the actual completed integration of AI into those workflows lags by two to three years. The headlines arrive first; the productivity gain arrives later; the headcount that was supposed to “shift, not disappear” disappears first and the new bottleneck Levie predicts gets re-created by a smaller team working harder.
What LostJobs is watching
- Whether the Q2 Challenger report shows AI-attribution share rising in non-tech sectors. March 2026 was 25% across all industries, dragged up by tech. If the May or June print shows financial services, healthcare, or insurance crossing 15% AI-attribution, Levie’s “tech is special” frame breaks publicly.
- Whether the next non-tech Fortune 100 company copies Microsoft’s first-ever-buyout structure. The buyout-vs-RIF choice Microsoft introduced last week is now on every CHRO’s desk. The first non-tech adopter — call it a P&G, a Travelers, a UnitedHealth — ratifies it as the cross-industry default for 2026–2027 reductions.
- Whether Slok’s Jevons-paradox prediction shows up in the BLS legal-services and accounting-services employment series. If the May and June BLS prints show growth in those categories, his argument has empirical legs. If they show flat or down, the elasticity-of-demand assumption underneath the paradox doesn’t hold for U.S. white-collar work in 2026.
The dry coda: the most-shared line on Wednesday April 29 from the Levie piece was not Levie’s. It was a Sam Altman quote the article cited about “AI washing” — the practice of relabeling normal cost-cutting as AI-driven because it gets a better stock-price reaction. Altman is the CEO of the company most responsible for making “AI” the word that gets the better reaction. Both halves of that sentence are the actual story.