AI Job Displacement - May 2026
Record growth + record cuts is no longer a contradiction in the model. It is the model.
Why do I write about the macro jobs picture?
A big part of my work is reading signals for what’s coming. The shape of jobs is a critical element in a bigger picture, and I want to tell the story of that bigger picture. That story runs right through the monthly jobs numbers, and if you can read what they say (and don’t say), you can begin to see this next reality emerging.
That gives you options. Choices.
It can also calm your nervous system, drive your strategy, and help you get ahead of events. Maybe even a little clarity and calm for good measure. If you’re awake at all, you could probably use more of that right now.
What I’m sharing today isn’t completely intuitive, and it might seem distant from what your work looks like right now. That’s okay. I’m going to do my best here to explain this moment in a way that helps you see beyond the headlines and around the corner.
We can start with the April jobs numbers: what they were, how they align with my projections, and what’s actually happening in this moment.
April beat.
The economy added 115,000 jobs in April, unemployment held at 4.3 percent, and wages were still up 3.6 percent from a year earlier, according to the latest Employment Situation report from the Bureau of Labor Statistics (BLS Employment Situation). Solid.
That was better than the roughly 70,000 consensus number moving around before the release. If all we expected was a recession alarm, the alarm didn’t sound.
I believe the April number, +/- 10% (previous months often get revised). The number’s real. The beat’s real. The surface reading is real.
The problem is the instrument.
On the same morning the report landed, Thomas J. Thompson described the economy as E-shaped: aggregate data still stable, confidence underneath more fragmented across industries, income levels, and skill sets (Thompson on LinkedIn). Consensus has begun moving toward what this series has been tracking for nine months.
The dashboard still says the labor market is fine. Under the dashboard, the labor market keeps splitting.
Thompson nailed it. The April print shows how it works.
The number hid the split
The headline jobs number rose. The participation rate fell. What does that mean?
Labor-force participation dropped to 61.8 percent in April, while the employment-population ratio slipped to 59.1 percent (BLS Employment Situation). For those who like simple math, you can increase a percentage by increasing the numerator or decreasing the denominator. The headline jobs number includes both. The percentage looks consistent, but the pie is shrinking.
It’s 4.3% of a lot less people.
Part-time workers for economic reasons increased by 445,000 to 4.9 million, and people unemployed less than five weeks rose by 358,000 to 2.5 million (BLS Employment Situation). The underemployment rate, U-6, rose to 8.2 percent from 8.0 percent in March (BLS Table A-15). Theresa Sheehan flagged that 8.2 percent as the highest since October 2021 if we exclude the late-2025 government-shutdown period (LinkedIn News).
Then there’s the concentration.
Healthcare added 37,000 jobs. Social assistance added 17,000. Transportation and warehousing added 30,000. Retail added 22,000 (BLS Employment Situation). Add those together and they account for 106,000 jobs, or 92.1 percent of the entire 115,000 gain.
If you’re looking for a broad labor market, that isn’t it. It’s a very narrow bridge.
The Information sector lost another 13,000 jobs in April and continues to get hammered. The losses showed up in telecommunications, motion picture and sound recording, and computing infrastructure, data processing, and web hosting. Since its November 2022 peak, Information is down 342,000 jobs, or 11.0 percent (BLS Employment Situation).
That’s where we find the the fault line.
The old dashboard counts job gains and job losses in the same room and tells us the room is still standing. But the new labor market isn’t one room. It’s a split house. Healthcare, logistics, retail, and social assistance are absorbing labor. Information is shedding it. The headline adds the two together and calls the sum a labor market.
That made sense when the same business cycle touched most workers through the same door. It makes less sense when one part of the economy is hiring humans to meet human demand while another is rebuilding the work process around software, agents, and fewer layers of management.
One report can be noise, but four reports in one week start to show a pattern.
Four prints, one wound
Last week told the same story four times:
On Monday, May 5, the March JOLTS report showed the Information layoff rate rising from 1.3 percent to 2.4 percent over the prior year, the largest increase of any sector and almost five times the national average (Indeed Hiring Lab). Information job openings were down 33 percent from a year earlier, while Professional and Business Services openings were down 20 percent (Indeed Hiring Lab).
On Wednesday, ADP reported that private employers added 109,000 jobs in April. Education and health services added 61,000, trade, transportation, and utilities added 25,000, and Professional and Business Services lost 8,000 (ADP Media Center).
On Thursday, Challenger reported 83,387 announced job cuts in April, up 38 percent from March. Technology companies announced 33,361 cuts in April, and AI was cited for 21,490 cuts, making it the top stated reason for the month (Challenger Gray & Christmas).
On Friday, BLS confirmed that Information lost another 13,000 jobs and remained down 342,000 from its November 2022 peak (BLS Employment Situation).
JOLTS, ADP, Challenger, BLS. Four different reports. One gaping wound.
The signal’s no longer a single-source anomaly. It’s not a weird line item. It’s not one bad tech company, one awkward earnings call, one CEO with a new deck and an expensive consultant.
We have a clear sector pattern. Openings are down. Layoffs are up. Cuts are being tied to AI budgets. Payroll is still sliding.
That creates the puzzle that’s kept the old dashboard alive: if the restructuring is real, where are the claims? Why aren’t people coming onto the unemployment rolls?
This one surprised me so much I laughed out loud when I realized it.
The severance hid the body
Cloudflare answered the question about claims last Thursday.
Note: they also host josephlogan.com, and I’m glad they’re out there.
On May 7, the company said more than 1,100 employees would be let go and that departing employees would receive full base pay through the end of 2026, with U.S. healthcare support through year-end and equity vesting through August 15 (Cloudflare blog). Generous. In the same post, Cloudflare said internal AI usage had increased more than 600 percent over the prior three months and framed the move as preparation for the agentic AI era (Cloudflare blog).
LayoffHedge estimates the cut at about 21.3 percent of Cloudflare’s workforce and reports cash restructuring costs of $105 million to $110 million, plus $35 million to $40 million in non-cash stock-based compensation charges (LayoffHedge Cloudflare tracker). Cloudflare’s official Q1 release showed revenue of $639.8 million, up 34 percent year over year (Cloudflare Q1 2026 financial results).
That’s the new pattern: growth plus cuts, AI adoption plus severance, public confidence plus private rupture.
The severance is the misdirect because claims data is built around a simpler story: worker loses a job, income stops, a claim begins. But a worker paid through year-end doesn’t show up in the numbers. A worker with severance, healthcare, equity vesting, and time to search sits outside the urgent claims channel. Someone laid off today might not show up in BLS numbers until next year. This is why continuing claims can stay steady while the labor market is being rewired.
Simpler: layoffs with long severance don’t hit the jobs numbers for months.
Layoffs in tech in particular usually come with a severance package. The buffer means the job disappears today but doesn’t show up in the numbers for another six months. It also turns time into camouflage. The executive announcement arrives in May. The budget savings begin now. The human claims may not surface until November, December, or January.
(I know that’s right after the midterm elections, by the way. Not our focus right now.)
Cloudflare isn’t alone. PayPal plans to cut around 20 percent of its workforce over two to three years (the long play) while creating an AI transformation and simplification team reporting to CEO Enrique Lores (TechCrunch). Coinbase cut 14 percent of employees and described the change as a shift toward small, high-context teams using AI (Fortune). Meta announced a planned reduction of roughly 10 percent of staff while pouring billions into AI infrastructure (CNN). Block cut 40 percent of its workforce, nearly half, with Jack Dorsey pointing to intelligence tools and smaller teams (CNN).
That’s “the twenty”.
It’s not a precise number, but it’s roughly right (that’s what we do here). 20% is a CFO-defensible cut large enough to change the cost base, small enough to be sold as focus, and easy enough to explain with the words AI, speed, fewer layers, and better execution.
It surprises me how normal it already sounds. When I say “[company] did the twenty”, you likely already know exactly what I mean.
The countercase earns a chair
There are other explanations for what’s happening with jobs, and I should mention the main countercase here because lazy certainty is how projections rot. Some of these have some merit, and I take them seriously.
Guy Berger called the April jobs report “good, not great” and argued that 2026 had started better than he expected, while warning that higher energy prices could bite later through weaker consumer spending and higher company costs (Berger on LinkedIn). That’s a serious point. Energy shocks can slow hiring. Iran risk is high. Input costs matter.
But the Information sector isn’t bleeding because of oil.
Energy can explain a cyclical headwind. It doesn’t explain a 2.4 percent Information layoff rate in JOLTS, a 33 percent year-over-year decline in Information openings, Challenger’s AI-linked cuts, and BLS showing Information down 342,000 jobs from peak (Indeed Hiring Lab, Challenger Gray & Christmas, BLS Employment Situation). Energy tends not to show up that fast in employment data, if at all.
The AI-washing critique also has merit. Caroline Hyde raised the question directly on May 7, asking whether tech was a leading indicator or an outlier and whether some firms were using AI as cover for older overhiring problems (Hyde on LinkedIn). Marc Andreessen made a similar argument in late March, saying AI had become a “silver bullet excuse” for layoffs tied to post-COVID overstaffing (Fortune).
They’re not wrong.
The 2022 to 2024 tech cycle was bloated. Some of what we’re seeing is delayed discipline cosplaying as futurism. Every board right now has permission to say “AI” and sound strategic instead of embarrassed.
But that’s Wave One, and it has an expiration date.
Wave Two is capital reallocation. Andy Challenger cut through the chatter: “Regardless of whether individual jobs are being replaced by AI, the money for those roles is” (Challenger Gray & Christmas).
Chatbots don’t walk into a cubicle and put on a badge. But payroll budgets are being moved from people into systems. On that point, the evidence isn’t that cute anymore.
The ECB study adds another useful guardrail. Its March blog found that euro-area firms making significant use of AI were about 4 percent more likely to hire additional staff, and firms investing in AI were nearly 2 percent more likely to hire (ECB blog). That sounds like optimistic until we read the next line of the mechanism. Firms using AI to reduce labor costs showed negative effects on hiring and positive effects on layoffs (ECB blog).
Both are true.
AI-intensive firms can hire. AI-using firms can cut. Companies can add talent while the median worker loses bargaining power. The market can expand at the frontier and hollow out in the middle.
Falk and Tsoukalas give us the game theory under the hood. In “The AI Layoff Trap,” they argue that rational firms can be pushed into an automation arms race, displacing more workers than is collectively optimal because each firm acts on its own incentives (arXiv). In their model, each firm’s profit-maximizing automation rate can become a strictly dominant strategy, even when the collective result weakens demand.
Put simply: companies kill each other’s demand by firing their market.
That’s why the countercase strengthens the thesis. It doesn’t matter whether every layoff is pure AI replacement. What matters is whether AI has changed the dominant strategy.
It has.
The split has two ends
There are two stories inside the same report.
At the top, knowledge work is being restructured really fast. Management layers are getting thinner. Pure coordination jobs make less and less sense. Entry-level work gets squeezed because the tasks that once trained people are the easiest tasks to hand to software. The apprenticeship ladder isn’t so much breaking as losing rungs.
At the bottom, workers on the margins lose footing. Sheehan didn’t treat the report as a tech story alone. She looked at U-6, concentration, participation, and workers being pushed into the weaker edges of the market (LinkedIn News).
Healthcare, social assistance, transportation, and retail are absorbing the pressure. That doesn’t make the system healthy. It means the absorbing sectors are carrying both ends of the split: displaced knowledge workers looking for a new bridge, and lower-wage workers trying not to fall through the floor.
That’s how a labor market can look stable from altitude and brittle at ground level.
The old dependence contract said the job would hold a bundle of things: income, training, insurance, status, schedule, and belonging. The new work system is breaking the bundle before a replacement shows up. We need to get busy making one.
That’s the deeper reason the BLS dashboard fails. It was built to count jobs (bundles). It wasn’t built to measure the bundle’s contents.
The bill comes due later
Here’s the falsifiable prediction.
If the severance-buffer mechanism is real, continuing claims should rise sharply in Q4 2026 or Q1 2027 as severance windows expire and displaced workers move from corporate transition packages into the public labor-market statistics.
If that doesn’t happen, the thesis here weakens.
If Information layoffs normalize, Information openings rebound, part-time for economic reasons rolls over, labor-force participation rises, and continuing claims stay low after the major severance windows close, then Berger’s noise hypothesis wins.
That’s the right test.
The next checkpoint comes sooner. The April JOLTS release in early June will tell us whether the March spike in Information layoffs was a one-month break or the next leg in the same structure. If Information layoff rates fall back and openings recover, the wound starts to close. If they don’t, the April payroll beat will look less like reassurance and more like a mask.
The point isn’t to predict doom. Not my thing. The intention is to make the forecast visible enough to be wrong. Good projections need an escape hatch. Otherwise they’re sermons with charts.
So the marker is simple: watch Information. Watch continuing claims after the buffers. Watch the concentration of job growth. Watch whether new hiring creates new ladders or just absorbs bodies. That’s where the “AI creates new jobs” thesis gets tested. That test should begin in about 6-9 months.
The next workforce hasn’t arrived
The bottleneck isn’t skill alone.
Skill matters. It always has. But the deeper bottleneck is the dependence contract. We built a society where the job became the container for almost everything adults need: income, healthcare, training, identity, dignity, rhythm, and proof that we still belong somewhere.
Then we started rebuilding work without rebuilding the container.
That’s why the emergent workforce hasn’t yet emerged. It’s late because the institutions around work still assume the old bundle will hold.
It won’t.
I can see scenario where a new market opens up for buying the workflows without the bundle, sort of like unbundling cable for streaming. Maybe human skills get streamed by subscription.
This is where the change happens subtly at first while the picture still reads strong. Severance kept the public numbers calm. The BLS dashboard reported stability because it was built for a market where the job was still the unit that mattered most.
But the thing underneath is changing, and it appears to be gaining momentum.
That can be frightening. It can also be useful. A broken dashboard is more than a warning. It’s an invitation to build better instruments.
We need measures that distinguish absorption from strength. We need claims data that can see severance delays. We need career systems that rebuild the apprenticeship ladder instead of pretending entry-level work will return unchanged. We need income and benefit models that don’t make human dignity depend on whether a company still needs a person’s tasks bundled into a job.
The old contract isn’t coming back whole, but the next one can still be built.
Buckminster Fuller on repeat: “You never change things by fighting the existing reality. To change something, build a new model that makes the existing model obsolete.”
The dashboard’s broken.
Good.
Now we can stop worshipping the gauge and start building the instrument panel for the world we’re actually entering.
What’s next?
I’m going to close for now with a good news/bad news scenario I believe is more good than bad.
When I read what others are anticipating, I see two distinct camps:
Don’t worry about AI—it’ll create far more jobs than it eliminates.
AI will end employment forever and people will descend into poverty and anarchy.
I see a different path.
The assumption that AI will create jobs makes it easy to ignore the problem altogether. The assumption that AI will end employment forever also limits our capacity to address the problem. Why care about the job market at all if you believe it’s going away forever?
The first step is to be curious about whether a problem exists, and the second one is to remain open to actually solving it.
I believe the truth is in the middle: first a contraction, then a period of uncertainty and financial strife, then a massive expansion in work for people. Much of what I’m doing here is in service of making that middle period shorter and shallower.
The new AI jobs don’t yet exist, at least not broadly. Most of them haven’t yet been invented. And once they exist, it’s going to take a minute for the market to price them and for people to achieve the cognitive and emotional skills to inhabit them.
I also believe those new skills obviously will tend toward places where people naturally add value: judgment, discernment, relational skills, emotional awareness, perception, intuition, nonobvious connections. Companies are taking on a new shape, and the people who work with, for, and through them will need to get really good at offering their services beyond staring at a screen and clicking keys.
I hope that helps make the picture clearer. The April numbers were good and will likely remain so even after the inevitable revision. If they continue to be good, things might be shaping up differently than I suspect they are. That would be good news in the short term.
But the structural signals have been showing us an irreversible trend for a year or more. The bad news is that what we’ve been doing in employment is changing irrevocably in tech and knowledge work. The worse news is that there’s no reason to believe change in Information stays there. We’re going to see an appetite for cost restructuring, and I imagine we’ll see a growing appetite for new capacity.
The good news is that new capacity creates both new demand and new opportunities to meet demand, which creates new customers who can afford that demand. I think we’ll hit a difficult trough, and I believe we’ll come out the other side into something new.
One new thing I’d like to see is BLS numbers that match what work looks like now. The ones we’re using have curb appeal but don’t hold up to closer scrutiny. If what we’re measuring is the health of the economy, we have better ways to report that.
These reports are a narrative based in projections using the Hari projection tool. Hari shows what’s most likely to happen, tipping points beginning to consolidate, and when and where decision points will happen. This same curated approach has been helping venture capital and private equity assess prospective investments, monitor their portfolios, and support their company founders. If you’re curious about what this looks like at the company level, see more here: josephlogan.com/hari