August 20, 2026 09:00 AM PST
(PenniesToSave.com) – Goldman Sachs published research on Wednesday that analyzed employment growth across more than 800 occupations, and the finding lands in a very specific place. Artificial intelligence is now detectable in the employment data of major developed economies, and the pressure is falling hardest on the people trying to start a career [1].
That is a narrower statement than most of what has been said about AI and jobs over the past two years. It is also better supported. For much of that period, the loudest voices in the conversation have been forecasting outcomes rather than measuring them, and the gap between the two has grown wide enough that readers have a hard time knowing which numbers to plan around.
Coverage of the report described it as showing “visible but narrow AI-related labor market impacts in the US,” while noting that evidence in other economies is more limited [3]. Both halves of that sentence carry weight. Something real is happening, and it is not yet happening to most people.
What follows is what Goldman actually measured, who is absorbing it, how much confidence the larger predictions have earned, and what any of it reasonably means for a household budget.
Quick Links
- What Did Goldman Sachs Actually Measure?
- Why Are Entry Level Workers Absorbing the Most Pressure?
- How Much Weight Do the Larger Predictions Deserve?
- Does Learning a Trade Solve the Problem?
- What Does This Mean for the Average American Household?
What Did Goldman Sachs Actually Measure?
The bank compared employment growth against long run trend across more than 800 occupations in several developed economies. Its central observation is that industries with greater exposure to AI automation have generally seen slower job openings growth since the second half of 2022, with the relationship most pronounced in Germany, Australia and the United States [1].
Employment in information and communication services, one of the most AI exposed categories, has slowed across nearly all major developed economies since 2022. Outside the United States, however, employment in those industries remains near or above its long run trend [1].
Four sectors stand furthest from where the historical pattern pointed: call centers, software publishing, management consulting and advertising. Call centers are the clearest case. Employment there sits 39 percent below trend in the United States, 33 percent below in Canada and 27 percent below in Germany [1].
That figure deserves a plain explanation, because it is easy to misread. Below trend does not mean 39 percent of call center jobs were eliminated. It means current employment is that far short of where decades of hiring patterns suggested it would be by now. The jobs may never have been posted rather than cut.
Adoption gives that context. Goldman combined 11 separate surveys and estimated AI adoption of roughly 15 to 20 percent across major developed markets, with France, the United States, the Netherlands and the United Kingdom leading and Italy, Japan and New Zealand toward the lower end. Major emerging markets came in between 10 and 15 percent [1].
Two similar looking numbers are circulating and they measure different things. Goldman reported that a 10 percent occupational exposure to AI was associated with roughly a 0.1 percentage point drag on annual headcount growth in France, Canada and the United States [1]. Separately, coverage of the report attributed a 0.1 percent increase in the US unemployment rate to corporate AI usage [3]. Those are two distinct findings that happen to share a figure.
Below trend does not mean jobs were cut. It means the hiring that history predicted never arrived.
Why Are Entry Level Workers Absorbing the Most Pressure?
This is where the report gets specific in a way most coverage of AI and employment does not. Goldman found that AI related headwinds were strongest among entry level workers, with a smaller additional negative effect in occupations considered at high risk of displacement [1].
The size of the gap is the finding. Across the broader labor market, a 10 percent occupational exposure to AI was associated with roughly a 0.1 percentage point drag on annual headcount growth in France, Canada and the United States. For entry level workers, the same exposure ranged from more than 0.6 percentage point in Australia to more than 0.2 percentage point in the United States [1]. In the American case, that is roughly double the effect on everyone else.
What that looks like in practice is familiar to anyone who has hired recently. Entry level analysts now compete against systems that summarize reports, generate presentations and handle administrative work in seconds [4]. The tasks that used to justify a first hire are the tasks most easily automated.
The sectors line up across sources. Customer support appears among the pressured fields in contributor analysis [4], customer service was flagged among the highest risk categories in a Senate report last year [2], and call centers are the industry Goldman measured furthest below trend [1].
Senator Mark Warner has warned that the disruption could hit young people first and hardest, potentially driving unemployment among recent college graduates as high as 25 percent within two to three years [2]. That is a projection, not a measurement, and it should be read as one.
Still, the direction matters for households. The workers this data points to are the ones with the shortest employment history, the smallest savings and the freshest student loan balances. For a young adult in that position, paying down debt before income becomes unpredictable is a more useful response than tracking every new forecast.
How Much Weight Do the Larger Predictions Deserve?
The alarm case deserves a fair hearing before it gets scrutiny. Geoffrey Hinton, the computer scientist widely called the Godfather of AI, has said it seems very likely to a large number of people that AI will cause massive unemployment [2]. He left Google in 2023 to speak more freely about the technology’s risks and received the Nobel Prize in 2024 for his machine learning work [2].
His reasoning is financial rather than technical. Hinton argued that one of the main sources of the roughly one trillion dollars being invested in data centers and chips will be selling AI that does the work of workers much more cheaply [2]. On that logic, replacement is not a side effect of the buildout. It is the business model funding it.
Hinton also applied the brakes himself. He acknowledged AI will create new jobs, though he does not expect the number to approach the number eliminated, and he urged skepticism toward every prediction including his own. Comparing the problem to driving in fog, he said that “10 years out, we have no idea what’s going to happen” [2].
Not every number in circulation carries that humility. Senator Bernie Sanders warned in an October 2025 report that nearly 100 million US jobs could be displaced by automation, naming fast food, customer service and manual labor as highest risk alongside accounting, software development and nursing. That report was based partly on estimates generated by ChatGPT [2].
A figure of nearly 100 million and a measured drag of a fraction of a percentage point are not describing the same phenomenon. Readers are entitled to know which one rests on measurement, particularly if policy and spending eventually get built on the other. Worth noting on timing as well: the Hinton remarks come from a November 2025 discussion with Sanders at Georgetown University, republished this month rather than newly delivered [2].
None of that proves the forecasters wrong. Goldman measured a snapshot, not a ceiling.
Readers are entitled to know which number rests on measurement and which rests on a forecast.
Does Learning a Trade Solve the Problem?
The standard advice to displaced office workers is to learn a trade, and it has real merit. Demand is genuine in electrical work, HVAC and infrastructure maintenance, and younger people questioning the cost of a four year degree are already choosing those paths [4]. Skilled work is stable, valuable and undersupplied, and the skepticism finally being applied to expensive credentials is overdue.
Jennifer Schwab Wangers, president of Learning Source, which provides workforce development solutions for Career and Technical Education programs, argues in Kiplinger that the trades are being treated as a universal answer to white collar displacement and cannot carry that weight [4]. Her company operates in that market, which is worth knowing when weighing the argument.
Her financial point is the sturdiest part of the case. She notes that an entry level HVAC technician earns roughly 20 to 25 dollars an hour in many parts of the country, a substantial reset for a midcareer professional who spent 15 years building a six figure career in a different field [4].
The track record supports caution too. Large scale retraining efforts have struggled for decades to return displaced workers to their prior earnings, often because participants land in lower paying occupations with weaker upward mobility, and the earnings damage can persist for years after someone is working again [4]. Pace compounds it. Technical skills in data, software and operations evolve quickly enough that curriculum can struggle to keep up with the systems workers are being asked to learn [4].
Wangers proposes counseling, job placement assistance, mental health support and some form of income assistance alongside technical certification [4]. Those are proposals from an interested party rather than findings, and they carry costs she does not price. She also allows that AI could reshape more jobs than it eliminates, with many professions evolving through collaboration rather than replacement [4].
What Does This Mean for the Average American Household?
For most working Americans right now, the honest answer is that the measured effect is small. Goldman concluded that AI related hiring pressures are clearly visible in global employment data but remain limited to a relatively narrow set of industries and workers [1].
The households with immediate reason to plan around this are more specific. Anyone with a young adult entering the workforce belongs in that group, as does anyone working in call centers, software publishing, management consulting, advertising or customer support [1][4]. If that describes your household, the useful move is preparation, not prediction.
Preparation looks unglamorous. It means building an emergency fund on a schedule you do not have to think about, avoiding new debt taken on against an assumption of uninterrupted income growth, and keeping skills adjacent to these systems rather than in direct competition with them.
It is equally worth naming what the data does not support. It does not support abandoning a career on the strength of a headline, and it does not support treating a figure of nearly 100 million displaced jobs as a planning input when that figure was produced partly by a chatbot [2].
Sanders made a point in that report that lands across the political spectrum, writing that work, whether as a janitor or a brain surgeon, is an integral part of being human, and asking what happens when that is removed from people’s lives [2]. Disagreement about his numbers does not require disagreement about that. Work carries dignity and structure that no transfer payment replaces, which is exactly why the question of who absorbs this deserves better evidence than it has been getting.
Final Thoughts
Goldman’s report is useful precisely because it is modest. It does not claim a crisis, and it does not claim safety. It documents a narrow effect concentrated on a specific group of workers, and it puts a number on it [1].
Whether that narrow effect stays narrow is the open question, and nobody has measured it, including the people most confident about the answer. AI adoption sits at roughly 15 to 20 percent in developed markets [1], which means the current readings describe an early stage rather than a finished one.
For most households the practical response is the same one that works against any uncertain income risk. A household budget that assumes some income variability handles a slow hiring market, a layoff or a career change without requiring anyone to guess correctly about the next decade of technology.
That is a less satisfying conclusion than either side of this argument tends to offer. It also happens to be the one the evidence currently supports.
Works Cited
[1] Lee, Jenny. “Goldman Studied Where AI Is Squeezing Labor Markets. Here’s What It Found.” CNBC, 19 Aug. 2026, www.cnbc.com/2026/08/19/goldman-ai-impact-employment-jobs.html.
[2] Fore, Preston. “‘Godfather of AI’ Says Billionaires like Elon Musk Are Right about the Future of Work, but He Predicts Mass Unemployment Is on Its Way.” Fortune, 16 Aug. 2026, fortune.com/article/godfather-of-ai-geoffrey-hinton-massive-unemployment-warning-big-tech-replacing-workers/.
[3] “Goldman Sachs Is Reporting That AI Usage by Companies Increased the US Unemployment Rate by 0.1%.” Yahoo Finance Canada, 19 Aug. 2026, ca.finance.yahoo.com/photos/goldman-sachs-reporting-ai-usage-170135062/.
[4] Wangers, Jennifer Schwab. “We Can’t Weld Our Way Out of an AI Employment Crisis.” Kiplinger, 19 Aug. 2026, www.kiplinger.com/personal-finance/career-paths/ai-employment-crisis.