The Replacement Story Has a Hole: Companies Are Cutting Workers Faster Than AI Can Replace Them

Meta announced 8,000 layoffs on May 20, 2026, the same week the company moved 7,000 more employees into AI teams, according to NPR. The arithmetic is the story: companies are betting that artificial intelligence will do the work they just cut people from doing, and most of them cannot prove the bet is paying off.

Tech sector job losses in 2026 have already passed 102,000 across 130 companies, according to TechSpot's layoff tracker. In March alone, AI was cited as the reason for 25 percent of all job cuts, or about 15,341 of 60,620 total, per Challenger Gray. Goldman Sachs estimates AI is already reducing U.S. employment by roughly 16,000 jobs per month, according to Yale Insights. Against that, Google, Amazon, Meta, and Microsoft are expected to spend $725 billion combined on AI infrastructure this year, per Financial Times reporting. The numbers do not add up — not because the job cuts are fake, but because the productivity gains AI was supposed to deliver have mostly not materialized.

The National Bureau of Economic Research found AI had little to no impact on employment or productivity in almost 90 percent of firms over the past three years, according to Yale Insights. That finding, from a research institution with no incentive to minimize AI's effects, suggests companies are cutting headcount ahead of the productivity payoff rather than after it arrives. They are betting on a future where AI handles the work, and using current labor reductions to fund the infrastructure that future requires.

Salesforce cut roughly 4,000 customer-service positions after AI agents began handling about half of all cases, according to Yale Insights. What got less attention was what followed: the company had data showing AI was handling the volume. It had less data — most companies have less data — showing AI was handling it as well as the workers it replaced.

The contradiction is not just a measurement problem. Companies cutting experienced workers are simultaneously cutting the institutional knowledge their AI systems were trained on and the expertise they'd need to govern those systems once deployed. A customer-service AI trained on how senior agents handled escalations is a different tool than one trained on junior agent responses. Cut the senior agents and retrain the model, and you have changed what the system knows. Keep the senior agents and you have a cost problem. The choice itself is a symptom of how little most companies understand what their AI is actually doing.

Gartner projects that half of companies that cut staff citing AI will rehire those roles by 2027, often under different titles, according to Forbes. That forecast — which would mean re-hiring the workers the companies are currently celebrating for replacing — suggests the current wave of cuts is partly cost-taking rather than permanent productivity transformation. When combined with the finding that 55 percent of companies that made AI-attributed layoffs report regret, primarily over quality degradation and loss of institutional knowledge, according to DigitalApplied, the picture shifts: the replacement story has a gap at its center that nobody is auditing.

The companies that cut and rehired offer the clearest evidence of the pattern. Klarna, the Swedish fintech company that famously replaced 700 customer-service workers with AI, later began rehiring as quality metrics declined, according to DigitalApplied. The company did not announce this as a failure. It announced it quietly, after the initial coverage had moved on.

Meta's situation is more complex. The 8,000 people leaving represent real job losses. The 7,000 moving into AI teams represent a genuine shift in where the company is investing. What is harder to determine — and what Meta has not published — is whether the AI work those 7,000 will do is the same work the 8,000 were doing, a expansion of AI capability atop existing work, or a replacement of one workforce with a different, smaller one. The company has data that would answer that question. It has not published it.

The gap between what companies are spending on AI infrastructure and what they can show for it is the structural vulnerability in the current wave of layoffs. A company that spends $1 billion on AI and cuts 2,000 workers producing measurable output has made a bet whose resolution is not visible in its current financial statements. The workers it cut are visible. The productivity gain is not. Until it is, the replacement story has a hole in it — and the companies that cut ahead of the data are funding an experiment whose results they have not yet collected.