The Timing Gap

AI job displacement for June 2026 and the reality of jobs and productivity.

I’ve been tracking the evolution of the workforce for the past couple of years, and I’ve been getting the feeling that we’re all describing the parts of it we see and understand—not the full picture. The last few weeks have brought some of those parts into sharp relief. “AI is going to take all the jobs” has had a longer run, but “AI will create more jobs than ever before” has been growing louder by the day. And in the background there’s a refrain that “AI has failed to deliver productivity gains”.

All valid positions at this moment in time. But I see them differently.

What I follow are structural signals. Broader tendencies on longer time frames. Those tell a very different story, one that looks different than the headlines and social media opinions. That’s what I want to show you.

Today is the monthly update on projections for AI-driven job displacement. It comes with three solid months of impressive job growth and some unusual movement beneath the numbers. Tomorrow I’ll share new projections on when and how AI will create new work, likely in a different form than classic W2 employment. The “when” is one of the most important questions right now. And Thursday I’ll take on the question of productivity gains. It’s an old story with some novel twists.

Understanding this moment requires broadening the aperture and moving beyond simple positions and opinions. Whether you’re investing, running a company, or finding your footing in your career, my intention is to give you a more complete picture of how the next few quarters are likely to go.

No controversy, just timing

Let me start with three counterintuitive propositions.

First, I do not believe there has been any meaningful job displacement by AI.

Second, I do not believe there has been any meaningful job creation by AI.

Third, I do not believe there have been any meaningful productivity gains from AI.

For each of these propositions, there’s an implied yet. I believe all three will happen. I haven’t seen compelling arguments to the contrary. The question guiding me has been when, and the answers for each of these three propositions reveal different timelines that explain why the current data looks so contradictory.

The debate between the people who warn of immediate mass unemployment and the people who promise a glorious era of new work is a false debate. It hides the gap between the two positions. Both positions are true. They’re simply measuring different things at different points on the same timeline.

Displacement is a function of executive strategy and capital allocation. It happens at the speed of a board meeting. Job creation is a function of structural integration, new business formation, and labor market reallocation. It happens at the speed of human institutions. Those aren’t the same speed.

If there’s a gap between the moment the old roles dissolve and the moment the new roles form, that gap is a valley. That’s where the pain is. How deep and wide that valley is matters more than who’s right.

A house divided

The Bureau of Labor Statistics reported on June 5 that the U.S. economy added 172,000 jobs in May. The unemployment rate held at 4.3 percent. Revisions to prior months were positive, adding 93,000 jobs to our previous record.

By any honest reading, that’s a solid headline. We have to acknowledge that directly. The April print we analyzed last month was revised upward from 115,000 to 179,000. That’s a meaningful correction to the directional read, and it belongs on the record.

But the headline aggregates a labor market that’s no longer one room. It’s a split house, and the split is widening.

The May gains were concentrated in three sectors. Leisure and hospitality added 70,000 jobs, five times its 12-month average. Local government added 55,000. Healthcare added 35,000. Those three sectors account for 160,000 of the 172,000 total gain.

The sectors most exposed to AI-driven displacement told a different story. Information was flat. Financial activities lost 22,000 jobs. Professional and business services added a meager 6,000.

Core white-collar employment peaked in April 2023 and has fallen 2 percent since then. All other private employment is up 3.7 percent in that span. These white-collar sectors added an average of 49,000 jobs a month in the decade leading to 2023. Since then, they have lost an average of 19,000 jobs a month.

This is the same structural pattern that manufacturing followed after 2001. It’s a sector-specific contraction that runs alongside a healthy aggregate labor market for years before the full damage becomes visible.

The unemployment rate is 4.3 percent of a shrinking denominator. Labor force participation held at 61.8 percent, the lowest since 2021. Over the past year, roughly two million people did not enter the labor force who would have under prior trends. They are not unemployed. They are simply not counted.

Long-term unemployment rose to nearly two million in May, representing 27.5 percent of all unemployed. That’s not a cyclical number. Cyclical unemployment resolves when the economy adds jobs. Structural unemployment compounds regardless of the headline.

Severance camouflage

There’s a simple reason why the headline jobs number can hold while the structural picture deteriorates. Tech layoffs operate differently from traditional industrial layoffs.

When a factory closes, workers file for unemployment insurance within days. When a technology company executes a twenty percent workforce reduction, the typical severance structure delays the appearance of those workers in public data by months. It’s not evasion. It’s just how the math works.

The base severance floor for most major technology companies is sixteen weeks. A worker laid off in May with sixteen weeks of severance doesn’t exhaust that runway until September. They don’t file for unemployment insurance until their severance ends. They don’t appear in continuing claims data until October or November.

Cloudflare cut eleven percent of its workforce in May but is providing full base pay through December 31. Those workers will not show up in the claims data until January 2027. Meta’s April cuts carry sixteen weeks of base pay plus additional tenure weeks.

Beyond the severance buffer, there’s a second layer of delay. The demographic profile of displaced white-collar workers is affluent. Many hold unvested equity that continues to vest through their severance window. Many have savings sufficient to extend their personal runway. Some research suggests that up to seventy-five percent of workers displaced by AI-related actions don’t file for unemployment insurance at all.

The practical implication is that the 97,006 announced cuts in May won’t appear in public payroll data until late autumn. The 397,755 cuts announced year-to-date are, in large part, still inside their severance windows. The headline jobs number is measuring a labor market that existed three to six months ago, not the one being restructured today.

The software paradox

It’s tempting to look at the current layoffs and ask why we started with software engineering.

Coding seems to be among the hardest cognitive tasks we have. It requires logic, syntax, and complex architecture. If AI is coming for jobs, surely it should have started somewhere easier. It seems backward that we’re cutting the builders first while the roles that look more automatable haven’t moved much.

The reason companies cut software engineers first isn’t because coding is easy to automate. It’s because software engineering has the highest cash velocity and the lowest institutional friction.

Consider the roles that look like obvious automation candidates: accounts payable clerks, HR administrators processing onboarding paperwork, procurement analysts running vendor comparisons, corporate compliance officers tracking regulatory checklists, IT administrators managing access provisioning, paralegal staff doing contract review, operations coordinators scheduling and routing. These roles are genuinely more exposed to AI than software engineering. The tasks are more rule-bound, the outputs more standardized, the judgment requirements lower.

But automating any of them requires touching systems that don’t belong to the department doing the cutting. Accounts payable means renegotiating how vendors invoice and how the ERP talks to the bank. HR administration means integrating with payroll, benefits, and legal compliance systems that were built by three different vendors over fifteen years. Procurement automation requires connecting sourcing tools to contract management to spend analytics to supplier databases. Corporate compliance means convincing legal and risk to sign off on a machine making decisions they currently own.

Every one of these automations is technically feasible today. None of them can be approved in a single meeting. They require cross-functional alignment, vendor negotiations, data governance reviews, and someone willing to own the liability when the system makes a mistake. The institutional friction is the bottleneck, not the technology.

Software engineers, by contrast, are expensive, concentrated in high-margin technology firms, and their output is already measured in code commits and deployment cycles. An executive can look at a team of thirty engineers, observe that AI coding assistants have increased individual output by forty percent, and make a headcount decision in an afternoon. No vendor negotiations. No ERP integrations. No legal sign-off. Just a spreadsheet and a conversation with HR.

The technology sector is the laboratory. The Challenger data, which showed AI as the primary reason for forty percent of all May cuts, is the proof of concept.

The spillover into the rest of the white-collar world won’t look gradual. It’ll look sudden. The automation doesn’t happen all at once, but the enterprise software providers are doing the institutional work right now. When SAP or Workday or ServiceNow bakes the automation of accounts payable directly into the tools every mid-sized company already uses, those companies won’t run pilots. They’ll just click a button. The friction disappears overnight, and so does the job.

The quiet period

There’s a final piece of nuance that explains the quietness of the past month. It’s not in the labor market data. You’ll find it on the financial calendar.

Within a thirty-day window, Anthropic confidentially submitted a draft S-1 registration statement to the SEC. OpenAI followed with its own confidential S-1 submission, targeting a valuation of nearly one trillion dollars.

During that same window, Sam Altman walked back his long-held support for Universal Basic Income, pivoting to “universal basic compute” as the preferred policy frame. Jensen Huang of NVIDIA and Demis Hassabis of Google DeepMind publicly called out “lazy AI layoff logic,” arguing that companies citing AI as the reason for cuts were using it as a convenient excuse for post-pandemic overhiring.

These events aren’t a conspiracy. They’re just a structural alignment of incentives.

Pre-IPO quiet periods are legally structured. S-1 risk-factor sections are written by lawyers whose job is to minimize liability. A chief executive who spent three years warning of AI-driven mass unemployment can’t include that warning in a prospectus without creating a material risk disclosure that complicates the offering.

The people with the largest financial and reputational exposure to a bad displacement narrative all softened their public framing inside the same thirty-day window that their IPO filings landed. The displacement thesis has simply become inconvenient for a specific set of actors at a specific moment, and those actors have the platform to say so loudly. That’s not evidence the thesis is wrong. It’s evidence the thesis is expensive.

The causal chain

The signal-based projection framework we use tracks a six-link causal chain:

Capital Investment → Infrastructure Build → Executive Strategy → Hiring Demand Shift → Payroll Effects → Unemployment

All six links are now active. The final link, visible unemployment, remains masked by the severance buffer and the measurement limitations of the household survey. It’s not hidden, just delayed.

The capital investment is active, with hyperscaler capex projected to reach nearly 700 billion dollars this year. The infrastructure build is active in the data center expansion and GPU deployment. The executive strategy is active, with AI cited in forty percent of May cuts. The hiring demand shift is active, with entry-level tech job postings down sixty-seven percent since 2023. The payroll effects are active, with white-collar sectors contracting. Each of these links was a prediction. Each has confirmed.

The framework’s prediction is that the final link activates visibly in late autumn and early winter, when the severance windows from the spring layoff cycle begin to close. That’s not a long wait.

The May print was solid. The structural picture isn’t. Those two things will stop being true at the same time sometime around Q4. That’s your window. It’s not closing dramatically. It’s just closing.

---

I track these structural signals monthly; you can subscribe to receive the updates directly as they develop. If you want the underlying analyses to these narratives, send me a message and I can share the details and sourcing. Intellectual honesty and showing the work are core to these projections.