Key Points:
- Goldman Sachs economists found AI is erasing approximately 16,000 net jobs per month in the U.S. — 25,000 lost to substitution, 9,000 added via augmentation
- Gen Z workers are structurally vulnerable: concentrated in routine white-collar roles — data entry, customer service, legal support — that AI automates most effectively
- Wage gap between entry-level and experienced workers has widened 3.3 percentage points per standard-deviation increase in AI exposure
- The displacement is happening before the new opportunities have time to materialize — “the destruction is hitting first, faster, and harder”
On April 6, 2026, Goldman Sachs published research that should make anyone entering the workforce today profoundly uncomfortable. The bank’s economists found that AI is erasing approximately 16,000 net jobs per month in the United States — 25,000 positions eliminated by AI substitution every four weeks, partially offset by 9,000 jobs added through AI augmentation.
Fortune reported that the pain is falling hardest on a generation that has grown up native to AI tools: Gen Z. They use the technology more fluently than anyone else. They are also the ones getting automated first.
The mechanism is structural, not incidental. Gen Z workers are disproportionate concentrated in the roles AI is best at automating: data entry, customer service, billing, insurance claims processing, legal document review. These are not jobs that require decades of judgment to master. They are, by design, the routine cognitive tasks that fit neatly into a language model’s training distribution.
Without accumulated experience or specialized expertise to serve as a buffer, young workers have little to insulate them when their employer decides to run the workflow through an API call instead of a human. Fortune reported that the unemployment gap between workers under 30 and those aged 31 to 50 has widened sharply relative to pre-pandemic averages — not because older workers are gaining jobs, but because younger ones are losing them.
The wage data is equally uncomfortable. For every standard-deviation increase in an occupation’s AI substitution exposure, the wage gap between entry-level and experienced workers has widened by 3.3 percentage points. That is not a projection. That is a measurement of what has already happened.
The productivity gains from AI are accruing to the people who own the AI and the people whose judgment the AI augments — not to the people whose jobs the AI has taken. OpenAI, which on April 7 published a policy paper proposing to tax robots to fund the transition, appears to understand this dynamic better than most.
As Frontierbeat reported, the company’s robot tax proposal is an explicit acknowledgment that AI destroys more tax revenue than it creates — by hollowing out the labor income base that funds social insurance programs.
The Automation Paradox: AI-Fluent Workers, First Out the Door
The irony cuts deep. Gen Z is the generation most likely to be building AI agents, running side projects on large language models, and natively fluent in the tools that are eliminating their job prospects. AI literacy is not protecting them. If anything, their fluency makes them more efficient at the tasks that are being automated — which makes the economic case for replacing them cleaner. A worker who can produce 200 AI-assisted outputs per day is more dispensable than one who produces 50 unassisted ones, not less. The tools that are supposed to make workers more valuable are making some of them more dispensable faster.
Goldman’s research is careful to note that it does not fully capture the offsetting hiring surge tied to AI infrastructure investment — data centers, power systems, fiber networks, construction — or the incremental labor demand driven by AI-enabled productivity gains in other sectors. Those effects exist. They are also harder to measure and slower to materialize. The destruction is immediate and measurable. The creation is indirect, delayed, and distributed across industries that have not yet been identified.
As the Goldman economists noted: “The problem for Gen Z is that the destruction is hitting first, faster, and harder in the roles they are most likely to hold. The creation of new opportunities, if history is any guide, will take longer to materialize and may require very different skills to access.”
Goldman’s data aligns with Anthropic’s own research on observed AI exposure, which Frontierbeat reported in March. Anthropic found that 94% of theoretically automatable tasks have not yet been automated in practice — but that the gap between theory and reality is narrowing fastest in the roles that young workers hold.
The theoretical threat is real. The observed displacement is already here. The question for policymakers — and for companies like OpenAI proposing robot taxes — is whether the redistribution mechanisms can move faster than the labor market deterioration driving the demand for them.
The 3.3-Point Warning Sign Nobody Is Talking About
The 3.3 percentage point wage gap widening per standard-deviation increase in AI exposure is the most important number in Goldman’s research, and it is getting the least attention. That is not a projection of future harm. It is a measurement of harm that has already occurred. It means that for every incremental step AI takes into an occupation, the experienced workers in that occupation gain relative to the entry-level ones.
The AI does not replace the senior analyst who knows which question to ask. It replaces the junior analyst whose job was to find the answer. That is not a technology problem. That is an incentive structure problem that technology is making worse.
Meta’s internal AI transformation — which the company disclosed requires engineers to produce 50% to 80% of their code with AI assistance — is the most aggressive corporate embodiment of this dynamic. As Frontierbeat reported, Mark Zuckerberg’s stated goal is “100x engineers” commanding fleets of AI agents.
The implication for junior engineers is straightforward: the path from junior to senior that once required years of apprenticeship now requires demonstrating judgment that AI cannot yet replicate. The bottleneck is no longer technical skill. It is experience. And Gen Z is at the beginning of accumulating it, just as the economic value of that accumulation is being disrupted by the technology they are most fluent in.
The caveats in Goldman’s research are worth acknowledging. The model does not fully capture AI infrastructure jobs — data center construction, power engineering, networking — that are growing in response to AI demand. It also does not capture the long-term productivity gains that may eventually create new categories of work that do not yet exist. History suggests that technological revolutions eventually create more work than they destroy, just not for the same people in the same industries at the same time.
The 16,000 jobs disappearing each month is a present-tense problem. The new opportunities that absorb them are a future-tense promise. Gen Z is living through the gap between those two tenses, one automated workflow at a time.
