They Got Their Jobs Back. The Jobs Were Worse.

Klarna's CEO Sebastian Siemiatkowski stood in front of cameras in May 2025 and said something CEOs almost never say. He admitted a mistake.

"We focused too much on efficiency and cost. The result was lower quality."

Over the previous two years, Klarna had eliminated 3,104 positions — more than half the company — through a hiring freeze and natural attrition. No mass layoff. No WARN Act filing. Revenue per employee exploded from $175,000 to $1.2 million. Siemiatkowski became the poster child for AI-driven workforce optimization. He toured conferences. The stock surged before its IPO.

Then customer satisfaction cratered. The stock crashed 65% post-IPO. And Siemiatkowski announced Klarna would begin hiring humans again.

The headlines wrote themselves: AI Can't Replace Humans After All.

Except that's not what happened. Klarna didn't rehire 3,104 employees. It recruited gig workers.

Same Job. Everything Else Gone.

Klarna's new customer service model is explicitly "Uber-type" — their word, not mine. Remote. Fully flexible. On-demand. Targeting students, stay-at-home parents, and rural workers looking for supplemental income. The pay is described as "competitive."

What it isn't: permanent. Benefited. Career-tracked. Union-eligible. Protected by Swedish employment law, which had covered the original workforce.

Dimension	Before (FTE)	After (Gig)
Employment type	Permanent, full-time	On-demand, gig
Benefits	Full (Swedish labor law)	None
Career path	Agent → team lead → manager	None
Job security	High	Zero
Union representation	Yes	No
Training investment	Weeks of onboarding	Minimal (AI handles most)

The function is identical. A human talks to a customer who needs something the chatbot can't handle. The title might even be the same. But the job — the actual economic relationship between worker and company — is unrecognizable.

And here's why this matters beyond one Swedish fintech company: Forrester's Predictions 2026 report estimates that half of all AI-attributed layoffs will be quietly rehired — offshore or at significantly lower salaries. Fifty-five percent of employers already regret AI-related cuts. But admitting the mistake means admitting the AI narrative was wrong. So they hire back cheaper, call it a different program, and move on.

Gartner independently predicts the same thing: by 2027, 50% of companies that reduced customer service headcount due to AI will rehire staff to perform similar functions, "even if the roles return under different titles."

The Degradation Pipeline

What Klarna built is a template. Not for AI displacement — we already had that story. For something worse: a four-stage pipeline that turns full-time employment into gig labor and eventually into nothing, while technically counting as "jobs" the entire way down.

Stage 1: Full-time employee. Salary, benefits, career ladder, protections, union eligibility.

Stage 2: Contractor or outsourced worker. Lower pay, fewer benefits, no career path, no union. This is David Weil's "fissured workplace" — a pattern that took decades to play out in the pre-AI economy.

Stage 3: AI plus gig fallback. On-demand only, zero benefits, no job security, no training investment. The Klarna model. AI handles 80% of volume; humans get the overflow. The human is a safety net, not a colleague.

Stage 4: Fully automated. No human needed. The Duolingo endgame — CEO Luis von Ahn announced in April 2025 that contract translators and content creators would be replaced by AI entirely. A hundred and forty-eight AI-created courses launched the same week.

At every stage, the worker is technically employed. They don't show up in unemployment statistics. They don't trigger WARN Act notifications. They don't file claims. They are invisible to every instrument the government uses to track labor market health.

The pipeline is invisible because each step looks like a job.

It's Already Everywhere

Klarna got the headlines because the CEO talked. Most companies don't.

In August 2025, Google terminated 200-plus "super raters" — PhD-holding contractors who trained Gemini through a Hitachi subsidiary called GlobalLogic. Internal documents suggested the workers were, as TS2 reported, "training the bots to take their jobs." When the workers tried to unionize through the Alphabet Workers Union, Google said they weren't Google employees. GlobalLogic said the decision wasn't theirs.

That same month, Elon Musk's xAI laid off roughly 500 data annotators — a third of the team — shifting toward "more specialized AI trainers." The broad workforce became a narrow expert core. The rest became available for gig annotation at lower rates.

Scale AI cut 200 employees and 500 contractors after a major Meta investment. The contractors — already the degraded tier — got cut first.

Chegg laid off 22% of its workforce in May 2025 as AI tutoring replaced human tutors. Many were already freelancers. The platform model itself was the degraded employment form. AI replaced the replacement.

And then there's the $200-billion BPO industry — Teleperformance, Concentrix, Wipro, TCS, Infosys — which was the original degradation pipeline. Western companies outsourced customer service to cheaper offshore labor decades ago. Now AI is displacing that already-degraded workforce. When the Indian IT index crashed 5% after Claude Cowork launched — wiping ₹1.3 lakh crore ($15.6 billion) in a day — the market was pricing in the next stage of the same pipeline.

The Earnings Penalty Nobody Talks About

Harvard and the NBER quantified this in 2025. Ni et al. studied WIOA training program outcomes and found that workers targeting high-AI-exposed occupations face a 29% earnings penalty compared to those targeting low-AI-exposed jobs. Workers coming from high-AI-exposed jobs had 25% lower earnings returns after retraining.

Average pre-training income: about $40,000 a year. The "successful" retraining outcomes — the ones the program counts as wins — still involve significant downward mobility.

That 29% isn't displacement. Those workers found jobs. They're employed. They're earning less, in less stable positions, with fewer benefits, doing work that an algorithm is learning to do cheaper. They're in the pipeline.

The freelancer data tells the same story from the demand side. An INFORMS/Organization Science study found that after AI tools launched, freelancers saw a 2% decline in jobs secured and a 5.2% drop in monthly earnings. Writing "About Us" pages — a staple gig — dropped 50%. Translation fell 30%.

The most counterintuitive finding: top-performing freelancers got hit hardest. For every 1% increase in past earnings, an additional 1.7% decrease in monthly income post-AI. The "great equalizer" floods the market with adequate-quality work and destroys the premium for excellence.

Why Nobody Can See It

The Bureau of Labor Statistics counts you as employed if you worked one hour in the survey week. Gig. Contract. Part-time. One hour. Employed.

The WARN Act triggers on mass layoff events — 500 or more in 30 days. Klarna's 3,104 positions dissolved over 24 months through attrition. Zero triggers.

Unemployment insurance requires you to be fired, not to have your terms degraded. If you were a Klarna employee making $50,000 with benefits and you're now a Klarna gig worker making $45,000 without benefits, you didn't lose your job. You lost your career, your security, and your union. Nobody counts that.

Warner-Hawley — the bipartisan AI disclosure bill currently stuck in committee — would require quarterly reporting of AI-related headcount changes. But it tracks headcount. If Klarna cuts 3,104 FTEs and hires 2,000 gig workers for the same function, the headcount change is -1,104. The quality change — permanent to precarious, benefited to bare — is invisible.

Every measurement instrument we have, and every one being proposed, is designed to count jobs. None of them measure the quality of those jobs. The pipeline exploits that gap.

The Bottom Line

The AI displacement story everyone expects is mass unemployment. Millions out of work. Bread lines. Political crisis. And that may come. But the likelier near-term outcome is something more insidious: the same work, done by the same people, in the same buildings — but stripped of everything that made it a career instead of a task. Benefits gone. Career ladder gone. Union gone. Security gone. The job title survives. The job doesn't.

If Forrester is right that half of AI-attributed cuts result in degraded rehiring, then for every person AI puts out of work, there's another person AI put back to work — worse. The displacement numbers need to roughly double. The pipeline is the story.

And the pipeline is invisible to everything we're building to see it.