In March 2023, Goldman Sachs published a report that launched a thousand panicked headlines. Generative AI, its economists wrote, could expose the equivalent of 300 million full-time jobs to automation. Within weeks, the World Economic Forum, McKinsey, OpenAI researchers, and IBM had all published their own forecasts, and the message blurring across news chyrons was simple: the robots are coming for your job.
Three years later, it is time for an honest accounting. We pulled every major 2023 forecast, checked each claim against the actual employment data, layoff records, and the forecasters’ own updates, and graded them. The result is not the story most people expected.
Here is the short version: the predictions were not wrong. They were misread. Nobody serious predicted 300 million people fired. They predicted 300 million jobs *touched* by AI, and on that measure, the forecasts are holding up uncomfortably well. What nobody quite predicted was *how* the pain would arrive: not through mass layoffs, but through hiring that quietly never happens, concentrated on the youngest workers trying to get their first rung on the ladder.
The 5 Forecasts That Scared Everyone in 2023
Before the verdicts, here is what was actually claimed. This matters, because most of the 2023 panic was based on headlines about reports few people read.
1. Goldman Sachs (March 2023). Generative AI could expose 300 million full-time jobs to automation across major economies, with roughly two-thirds of US and European jobs exposed to some degree. Administrative workers and lawyers were flagged as most exposed. The report also projected AI could lift global GDP by 7% over ten years, and explicitly noted that historically, automation-driven job losses have been offset by new occupations. (Source)
2. World Economic Forum, Future of Jobs Report 2023 (May 2023). Based on a survey of 803 companies employing 11.3 million workers: 83 million jobs eliminated and 69 million created between 2023 and 2027, a net decrease of 14 million jobs, with 23% of all jobs expected to change. (Source)
3. McKinsey Global Institute (June 2023). Generative AI could add $2.6 to $4.4 trillion in annual economic value, and combined with other technologies could automate work equivalent to 60 to 70% of a typical employee’s day. Office support work had 87% technical automation potential. (Source)
4. OpenAI, OpenResearch and the University of Pennsylvania (March 2023). The “GPTs are GPTs” paper estimated 80% of the US workforce could have at least 10% of their tasks affected by large language models, and 19% could see half or more of their tasks affected. Crucially, the paper measured *exposure*, not replacement, and did not distinguish augmenting workers from replacing them. (Source)
5. IBM Institute for Business Value (2023). Executives estimated 40% of their workforce would need to reskill within three years because of AI and automation. But 87% believed employees were more likely to be *augmented* than replaced. (Source)
Goldman Sachs Said 300 Million Jobs Were Exposed. What Happened?
Verdict: on track, as a measure of exposure.
Here is the distinction the headlines destroyed: “exposed to automation” never meant “eliminated.” Goldman Sachs’ own researchers stressed that most affected jobs would be *complemented*, with only 25 to 50% of the workload in exposed occupations actually replaceable.
In April 2026, Goldman Sachs updated its own work. The 300 million exposure figure still stands. But the bank now estimates that only about 2.5% of US employment faces direct elimination risk even if today’s AI capabilities were deployed across the whole economy. The net drag on hiring from AI substitution runs around 16,000 jobs a month, mostly through attrition rather than firings. (Source)
The scarier-sounding half of the forecast, the 7% GDP boost, was always a ten-year projection. Three years in, with only about 7% of companies having fully scaled AI beyond pilot projects, the macroeconomic productivity effect remains small. That one is simply too early to judge. (Source)
The WEF Predicted a Net Loss of 14 Million Jobs. Where Are We?
Verdict: superseded by the forecaster itself.
This is the most interesting reversal in the whole set. The WEF’s 2023 report, built on employer expectations, projected a net *loss* of 14 million jobs by 2027. Then came the WEF’s Future of Jobs Report 2025, with a bigger survey base of over 1,000 companies, and the forecast flipped: 92 million jobs displaced but 170 million created by 2030, a net gain of 78 million.
The 2023 employer survey turned out to be too pessimistic about net job creation. The churn it predicted is real, around 22% of jobs disrupted, but the direction of the net number reversed. If you cited the “14 million jobs lost” figure anytime in the last two years, update your talking points. (Source)
Which Jobs Actually Shrank, and Which Grew?
This is where the 2023 panic meets the Bureau of Labor Statistics, and the panic loses.
Office and administrative support, the occupation group Goldman Sachs and McKinsey both ranked as *most* exposed to AI automation, did not shrink. BLS data shows employment in the group grew 5.5% between February 2024 and February 2026. Paralegals and legal assistants, another flagged group, held essentially flat. Lawyers, supposedly near the top of the automation risk list, are projected by the BLS to grow 5% between 2025 and 2035, faster than the average occupation. (Sources, BLS)
None of this means AI had no effect. It means the effect showed up as *augmentation*: the same workers, doing the same jobs, with AI handling chunks of the workload. Exactly what the careful forecasters said would happen, and exactly what the headlines ignored.
What about outright AI-driven layoffs? They are real and rising. Challenger, Gray & Christmas, which tracks US job cuts, counted 54,836 layoffs explicitly attributed to AI in 2025, twelve times the number when it started tracking in 2023. Big names appear on the list: Amazon cut 14,000 corporate roles, Microsoft around 15,000 across 2025, Salesforce shrank its support team from 9,000 to about 5,000 with its CEO saying AI now does half the work.
But keep the denominator in mind: those 54,836 AI-cited cuts were just 4.5% of all US layoffs in 2025. And the New York Federal Reserve found that only about 1% of firms reported cutting staff because of AI in a six-month window, even as 40% of service firms were using AI tools. Several analyses conclude that “AI made us do it” is at least partly cover for ordinary restructuring. The layoffs are real, but AI is often the justification rather than the cause. (Source)
The Entry Ladder Is Being Pulled Up
If mass unemployment did not arrive, something subtler and arguably more worrying did. The sharpest finding in the 2026 data is not about layoffs at all. It is about hiring that never happens.
Stanford’s Digital Economy Lab analyzed ADP payroll records covering millions of workers and found that employment among 22 to 25 year olds in the most AI-exposed occupations sits roughly 19% below the expected trend. Workers over 30 in the same occupations show no such gap. The mechanism is not firings. It is reduced hiring: companies simply stopped bringing in juniors for work AI can now draft, summarize, or screen. (Source)
The supporting numbers line up. The New York Fed reported recent-graduate unemployment at 5.6% in mid-2026 versus 4.1% for all workers, with 42% of young graduates underemployed. Entry-level job postings fell 16% while applications per opening rose 26%. New graduates now make up just 7% of Big Tech hires, down by more than half since 2019.
In other words: the 2023 forecasts got the *where* right and the *how* wrong. The pain is concentrated exactly in the occupations the exposure studies flagged, exactly as OpenAI and UPenn’s paper implied it would be. But it is landing on people trying to enter those occupations, not on the people already in them. The experienced paralegal keeps her job and gets an AI assistant. The 23-year-old who would have become a paralegal never gets hired.
The Scorecard: Grading Each 2023 Prediction
2023 AI Job Forecasts: The 2026 Scorecard
Every major prediction, graded against the actual data
What the Forecasters Got Right (and What They Missed)
They got right: that AI would touch an enormous share of work quickly. With 88% of organizations now using AI in at least one function, the exposure story played out roughly on schedule. They got right that augmentation would dominate replacement: 87% of IBM’s surveyed executives expected augmentation, and the employment data backs them. They got right that reskilling would become the central labor-market story of the decade.
They missed: the distribution of the pain. The models assumed that highly exposed occupations would shed workers. Instead, exposed occupations kept their workers and the hiring pipeline narrowed. Nobody’s 2023 model had a clean prediction for “the jobs survive but the on-ramps close.” They also missed how slowly enterprises would move from experimentation to transformation: three years in, two-thirds of companies are still piloting, which is why the productivity boom keeps getting postponed.
They could not have known: that the WEF would reverse its own net-jobs call within two years, or that “AI took my job” would become such a convenient corporate euphemism that economists now have to disentangle genuine automation from ordinary layoffs wearing an AI costume.
The Honest Bottom Line
So, did AI take our jobs? Not the way the 2023 headlines said it would. Administrative employment grew. Lawyers are still being hired. The robot apocalypse did not arrive.
But something real did happen, and it is worth saying plainly: the 2023 forecasts described *exposure*, the press reported *elimination*, and the truth landed in a third place nobody headlined. AI is quietly reshaping who gets hired, not who gets fired. If you are established in an exposed occupation, you likely have an AI assistant now. If you are 23 and trying to break into one, the door is 19% narrower than it should be.
That is the prediction worth watching for the rest of the decade. Not the unemployment rate. The hiring rate.



