Robots Can Do 74% of Physical Work. They Win on Cost for 0.3%
Anthropic's new robot exposure index: robots can do 74% of US physical tasks, but are cheaper than people for just 0.3% of work. Here is who that hits first.
In-depth analysis and insights on AI automation, career trends, and the evolving workforce.
Explore AI impact analysis organized by industry sector
Latest research, data, and trends in AI and the labor market
Anthropic's new robot exposure index: robots can do 74% of US physical tasks, but are cheaper than people for just 0.3% of work. Here is who that hits first.
In a double-blind Swiss trial, AI-recommended placements raised refugees' three-year employment by about 10% at CHF 50 per adult. All of the gain came after COVID.
A medical-device firm made 5,000 staff send 200 AI queries a month. 31% were repeats or off-task. Where branches cut the quota to 100, filler made up 90% of the drop and sales rose 7%.
NBER WP 35793 reads AI productivity from stock prices: software engineering up the equivalent of 32.6% through 2025, GDP up 3.6%. One cost weight explains the gap and can halve it.
Unemployment for 22-to-25-year-old graduates hit 7.8% in June 2026. That's normal for June. An NBER study finds no AI spike yet, and shows who the data misses.
AI-exposed employment fell 0.6% in August, erasing July's rise. For 22-to-25-year-olds the gap to safer jobs narrowed — because the safe side fell 2.0% too.
Statistics Canada says 35.9% of workers used generative AI for their job in the year to March 2026. Run the arithmetic on how deeply they use it and that share collapses to about 4%. The same survey shows adoption is highest in jobs where AI assists — doctors, nurses, engineers — and slower in the jobs it might replace. Here is what that inversion means for your work.
Statistics Canada's new survey shows generative AI use jumped from 17% to 30% of workers in under a year — and degree-holders are five times more likely to use it. Here is where your job stands.
Nine of ten AI models say no to the same applicant 17.3% of the time after post-training, up from 5.6%. Older applicants are hit first. A COLM 2026 preprint.
Texas computer science enrollment fell 27.8% in a year while nursing grew 10.6%. The Dallas Fed finds AI-exposed majors now carry a 1.7-point hiring penalty and 5% lower first-year pay per 10 points of exposure. We summed its own chart data: the pivot away from exposed fields is real, uneven, and covers just 6.4% of students.
Under-30 employment in Korea's most AI-exposed jobs fell 13.4% from 2022 to 2025, but the slide began a year before ChatGPT. Five KEIS studies, read against their own press release.
A new ILO brief interviews 21 Chinese firms and surveys 1,591 workers. Big reported gains, 62% of firms with no metrics, and 39% expecting lower pay.
62% of youth use AI daily, 30% were taught how. UNESCO-UNEVOC's new TVET guide answers with five principles, a four-phase pathway, and a rule: fail one, do not buy.
Two MIT FutureTech economists tell Brookings that exposure scores are the wrong basis for policy. The numbers behind their five fixes are thinner than the headline.
Half of euro area workers now use AI at work, the figure behind Lagarde's speech. The ECB's own survey puts the economy-wide time saving at 3.8%. Mind the gap.
Only 15% of recruiters expected it. Randomly assigned AI voice interviews raised job offers 12% and starts 18%, no productivity loss. What it does not show.
Two-thirds of Texas firms now use AI. A Dallas Fed study of millions of job postings finds exposed occupations down 7-8% since ChatGPT, cut by incumbents.
ADP's July Canaries reading: AI-exposed employment up 0.1%, the first positive number in the series. Least-exposed jobs grew 1.1%. The gap widened, not closed.
Brookings shows why NAICS, SOC and O*NET miss frontier jobs: supply chains triple semiconductor postings, an uncoded job doubled, and a BLS forecast flipped.
$13,598 per person, +1.7 points of employment, +$800 a year. The largest meta-analysis of U.S. job training yet, and the one program type that beats it tenfold.
Yale Budget Lab's January/February 2026 CPS update finds AI exposure metrics flat, dissimilarity in historical range, and no employment-unemployment link. What is missing matters as much as what is there.
OpenAI's Signals Q1 2026 update shows ChatGPT going mainstream — Dominican Republic and Haiti rose 9 places, Japan 8, with feminine-name users now a majority. What this means for emerging-market labor markets.
The U.S. unemployment rate climbed to 4.3% in March 2026, and everyone wants to blame AI. Yale's Budget Lab dug into the data and found something almost nobody is saying out loud: the numbers point somewhere else entirely.
One in four workers globally is in an occupation with some GenAI exposure. The ILO's April 2026 brief explains why that number is widely misread — and what it actually means for your career.
"Shall" appears zero times, "should" 13 times. The ILO record of its April 2026 AI-in-manufacturing meeting shows which sentences governments, employers and unions deleted before signing.
NBER w35720 pre-registered RCT: AI raised patent draft quality 0.38 SD, juniors most. But durable judgment gains without AI went only to seniors (0.45 SD); junior scores split.
IMF economists price the time AI currently saves at $2.7 trillion a year — 3.4% of GDP across 86 countries. But 96% of it accrues to high-income countries, and in Tanzania nearly all gains flow to under 5% of workers. Who actually captures AI's value?
AI fell from the #1 stated cause of US job cuts to #4 in August 2026, its lowest count since December. One month is not a trend — here is what the data shows, including a hiring number up 725% on a startlingly weak base.
79% of Cambodia's surveyed platform drivers say a customer rating decides their earnings, and half have been penalized by an app. Inside the ILO's new diagnostic of a workforce whose boss is already an algorithm.
28% of EU workers already use AI at work — only 15% got any AI training. A six-agency report maps where the real skills gap sits.
The first task-level survey of genAI adoption finds use across 80% of US occupations, yet fewer than half of workers adopt in most tasks. Who adopts now matters more than what.
The US government just assigned every occupation an official AI exposure category - built partly on Claude and Copilot usage logs. BLS projects +5.9 million jobs by 2035, one-third the pace of the last decade. Here is what the first official AI exposure score actually measures - and what BLS insists it does not.
KPMG gave nearly 4,000 back-office workers GenAI for eight months. Their skill never improved — and training's effect vanished within a month. 713,564 prompts say why.
Occupations built of AI-exposed tasks are losing posting share at slope -1.33, but the task categories inside them decline at only -0.83. A 752.6-million-ad study of the Chinese labor market concludes: the tasks outlive the jobs that carried them.
Only 16% of 3,785 PhD students welcome AI everywhere. The largest group — 44% — draws a line through their own job. Here's exactly where that line falls.
Anthropic analyzed ~400,000 Claude Code sessions: users made 70% of planning decisions but only 20% of execution decisions — yet experts reached verified success at double the novice rate, and recovered troubled sessions at nearly four times the rate. The expertise premium did not disappear. It moved.
UK postings for programmers and software developers fell 73.4% from 2022 to 2024. But the occupation we measure at 0.0% observed AI exposure shrank too, and the care job we rate as more exposed grew faster than the one we rate as less exposed. The whole market fell 39.7% — and automatability does not sort the table.
On three of seven real work tasks, a worker model scored better with no AI help at all than with any of the ten AI assistants tested against it. A new UC Berkeley benchmark splits "can AI do your job" from "can AI help you do your job" — and finds the two barely track each other.
A stock of data-science credentials lost 82% of its predictive power during the AI transition. Then the audit found that half to three quarters of that loss had nothing to do with AI. Here is what 444,698 Kaggle entries actually say about whether your certificate still means anything.
Nearly a million Filipinos share one job code — and all 985,200 of them sit in the highest AI-exposure band the ILO measures. Three clerical codes carry 78% of the country's highest-risk employment. Here is what the primary data actually says, including one number in the headlines that turns out not to be Philippine at all.
Nearly 80 million ASEAN workers hold jobs generative AI could technically touch - and employment in exactly those jobs has grown every year since 2017. The ILO's new brief separates exposure from displacement, and the gap between them is the whole story.
About half of American workers don't hold a bachelor's degree, and almost none of the AI-and-jobs research is written about them. A Dallas Fed economist just put numbers on that gap: administrative roles at 60% automation risk, truck drivers at 12%, and a short-term credential worth about $5,000 a year. Here is where the Fed's own prescription breaks down.
ADP Research and Stanford linked 5 million job postings to 9 million payroll records to answer a question labor statistics almost never touch: what is each individual task inside a job actually worth? Eight tasks carry a wage premium, six carry a penalty — and five lost value after generative AI arrived. One of the five is the exact task everyone has been told to move toward.
Employment in America's most AI-exposed occupations has fallen for 33 consecutive months among workers aged 22-25 — down 4.3% in the past year. Across all ages, the same occupations fell just 0.2%. That gap is the entire story, and it says something very specific about where the damage is actually landing: at the door, not inside the building.
The share of UK businesses using AI nearly tripled — from 12% to 35%. Over the same period the number of AI tools the average adopter actually runs went from 1.4 to 1.6. New ONS official statistics name the exact job roles employers say are affected, and that gap between wide adoption and shallow use changes what you should expect for your own job.
Job postings for AI-vulnerable occupations fell 5.8% in high-income countries after ChatGPT — and barely moved in developing ones. A World Bank study of 555 million postings across 84 countries shows who is absorbing the first wave of GenAI displacement: entry-level workers, with US postings requiring no experience down 20% and administrative support down 40%.
The World Bank's flagship WDR 2026 finds 14.2% of jobs in high-income countries at risk from generative AI versus just 4.5% in developing economies — yet the productivity upside is nearly identical. Here's what the Bank's most comprehensive AI assessment means for your job.
The BIS says AI has not triggered mass layoffs — it has frozen hiring instead. Unemployment stays low while the job hunt gets brutal. Bulletin 130 explains why both things are true, which occupations show the first substitution signals, and what workers can do with the warning.
A new IMF paper finds 23.1% of Israeli workers hold high-AI-exposure, low-complementarity jobs. The most vulnerable are not who you would expect: sales, business administration, and ICT professionals top the risk list.
Between 75% and 85% of AI-related job postings sit in STEM occupations — in all 10 countries a new study examined, from the US to Indonesia. Health, education, services, and logistics each get under 5%, and the heaviest AI skill demands land on entry-level jobs. Here is what the data means for your career.
Anthropic will fund $5-30M grants testing retraining, income support, and worker equity ideas before AI displacement peaks. Here's what's in the agenda.
Here is a number that should change how you think about AI and your job: 27 percentage points. That is the gap between the European workers who say they need AI skills and the ones who have actually been trained. A new Cedefop foresight report maps four futures for 2040 — and argues that which one arrives is a choice, not a prophecy.
If AI automates 90% of your job, someone still does the last 10% — and that slice may be your safest, most human work. Economist Joshua Gans on why exposure scores are starting points, not forecasts.
Ask 203 Texas executives who actually use AI what it did to their staffing, and 76% say: nothing. But the same Dallas Fed survey shows expected job cuts running 2.5x hotter than actual ones - plus a 10-to-1 gap between the productivity AI delivers and the pay it returns.
About three-quarters of US adults fear AI will cost jobs. Fewer than 6% of private-sector workers have a union to bargain over how it gets deployed. A new Brookings framework (June 29, 2026) calls that gap "the great mismatch" — and argues the answer is not one silver-bullet policy but four levers used together: brakes, steers, buffers, and shifts. Here is what it means for your job, and the one line in the report that quietly contradicts the advice you have been given.
Your industry's AI exposure score predicts almost nothing. New Dallas Fed research across the US and 16 European economies shows the real variable is usage: high-exposure US sectors grew productivity 3.7% while identical EU sectors showed zero correlation. The US AI usage index is 3.69 vs the EU's 1.85. Exposure is assigned to you. Usage is chosen.
More than half of US and German workers think automation will raise unemployment. Fewer than three in ten think it will happen to them. A randomized experiment with 5,147 workers shows what happens when you show people the actual evidence - and where the reassurance stops working.
The UK government just put a number on the AI shift: 70% of workers are in jobs AI can already partly do. Graduates and juniors are feeling it first. Here is who is exposed, who is protected, and which jobs still grow to 2035.
A new RL Feasibility Index scoring 17,951 O*NET tasks finds AI exposure has been badly misjudged. Power plant operators and railroad conductors face far more risk than assumed, while musicians and physicians face far less.
The Bank for International Settlements says AI already saves 20-50% of task time — and the US sectors gaining the most productivity are hiring the least. Here is what the BIS Annual Economic Report 2026 means for your job.
June 2026 U.S. layoffs cooled to 45,849 — down 53% from May. But AI drove 31% of them, marking the fourth month running that artificial intelligence topped the list of reasons employers gave for cutting jobs.
Anthropic tracked 9,700 workers and millions of Claude conversations. Tax queries jumped 8x on April 14. And the people handing the most work to AI are the least worried about losing their jobs. Here is what the data actually says.
PwC analyzed over 1 billion job ads across 27 countries. The surprise: AI-exposed companies are growing headcount AND raising wages faster. Here is what it means for your career.
More than 70% of euro area firms now touch AI. But only 7% use it intensively — and the ECB just revealed what separates them from everyone else.
The European Central Bank just put a number on AI's labour-market footprint: a 15-point employment gap has already opened between high-risk and low-risk jobs since 2019. Your job may already be on one side of it.
A study of 2.26M Upwork contracts finds generative AI cut the value of skills, credentials and experience by 7.8% in exposed fields. Here is what it means for your career.
PIIE data shows AI adoption jumped from 4% to 12% at large firms in two years. But here is the twist: the industries adopting it fastest are the ones that already pay the highest wages. AI is following the money, not chasing the low-skill jobs everyone warned about.
Youth employment in AI-exposed jobs has fallen 13% since 2022 — not from layoffs, but because young workers cannot get hired in the first place. The Dallas Fed explains why.
In just two years, generative AI adoption in German workplaces jumped from 5% to 24% — a near five-fold leap. But the gap between who is using it and who is not tells the real story about your job.
ADP's Today at Work 2026 report finds 50% of workers now use AI weekly and 20% daily — yet daily users are 4x more likely to feel they are underperforming. Here is why this productivity paradox matters for your career.
Entry-level workers are taking the first hit from AI. Britain's answer is a £20m alliance training 400,000 students and almost 1 million young people. Here is what it means for your first job.
U.S. employers cut 83,387 jobs in April 2026, up 38% from March. For the second month running, AI was the #1 cited reason at 26% of all cuts — even as year-to-date layoffs fell 50%.
AI reshuffles roles before cutting them - reallocation vs redesign
Eurofound's EWCS 2024 survey of 36,644 workers across 35 countries shows just 12% use generative AI at work — but the bigger finding is what AI didn't do.
2022年,欧盟职业类岗位招聘比例跌至33%的低点。2025年,这一数字反弹至超过36%——这一逆转几乎与生成式AI热潮同步。以下是Cedefop数据对你职业规划的意义。
Australia's federal Jobs and Skills Australia agency analyzed 358 occupations and found 79% of workers face low or very low AI automation risk — but the 21% who don't are clustered in routine clerical work, while professionals see the highest augmentation gains.
Two years ago it was 8%. Now 28% of US workers use ChatGPT on the job — and Fortune 500 adoption hit 93%. Here is what OpenAIs April 2026 workplace report means for your specific occupation, and the gap quietly opening between knowledge workers and everyone else.
A new April 2026 paper tracks 40 years of finance productivity — and shows agentic AI is squeezing the middle layer hardest, with AUM-per-employee up 149%.
A new arXiv paper projects 35.6% of information-intensive Bay Area occupations will cross the moderate AI displacement threshold in 2026. Here is who, why, and what protects your role.
A new MIT FutureTech study flipped the automation forecast: instead of experts predicting AI impact, 17,000+ workers evaluated real LLM outputs on their own tasks. The results upend conventional wisdom about who is most exposed.
Job-changers earned 6.4% wage growth vs 4.5% for stayers in January 2026 — the narrowest gap since 2020. New-hire pay broke its 18-month $18/hr plateau, jumping to $19. And 45% of workers now work part-time, up 6 percentage points from 2019. ADP's structural pay-trends analysis.
ADP Research surveyed 39,000 workers globally and found just 25% feel their job is safe — 28% in the U.S. The disconnect between strong headline labor data and weak worker confidence is the most important labor-market signal of 2026. Plus: secure workers are 6× more engaged.
Even in AI-exposed occupations, entry-level workers are seeing relative employment declines. A May 2026 Brookings synthesis triangulates payroll data, OECD studies, and the Anthropic Usage Index to argue AI growth acceleration is plausible but its distributional effects are already showing up — and not in workers favor.
A new NBER paper compared 5 forecaster groups on AI's labor market impact. The median says GDP grows 2.5%/year. The rapid scenario says ~10M jobs gone by 2050. The disagreement reveals more than the numbers.
A US Federal Reserve governor used the phrase 'essentially unemployable' out loud last month — and he wasn't talking about a fringe scenario. Fed Vice Chair Michael S. Barr's February 17, 2026 speech laid out three AI futures the Fed is actively planning around, and signals the rate-cut narrative may not survive an AI productivity boom.
29% of US workers are in occupations with the lowest AI exposure. 18% are in the highest. And the share has not budged since ChatGPT launched. The Yale Budget Lab's February 2026 synthesis finds AI exposure is real and measurable — but it has not yet translated into measurable employment displacement.
A new MIT-led study shows full AI automation is almost never the cost-minimizing choice for firms. Here is what 11% actually means for your job.
A new arXiv paper tracks assets-under-management per employee across three tech waves and finds finance is not facing a cliff — it is on the next chapter of a 40-year transition. What this means for advisors, analysts, and back-office workers in 2026.
On April 22, 2026, Anthropic launched the Economic Index Survey, a monthly qualitative survey of Claude users covering AI adoption, productivity, and what workers want from the next decade. Here is what it asks and why it matters.
Anthropic's economists built a new way to measure which jobs are actually being done by AI right now. The first warning sign? Young workers entering high-exposure fields are seeing 0.5pp fewer hires. The full data tells a more hopeful story than you might expect.
OpenAI发布四维分析框架,覆盖921种职业——18%面临短期自动化风险。受压最大:法律支持、办公行政。受保护:律师、护士、教师。对你这周意味着什么?
**36%** 的女性从事的职业中,AI 可能重塑一半以上的日常工作——而男性这一数字为 **25%**。这不是四舍五入的误差,而是 Brookings 基于 ChatGPT-4 对 1,000 多个职业任务暴露评分得出的警示信号。
ILO与世界银行联合研究135个国家发现了一个鲜明对比:人工智能威胁发达国家的文案工作,而发展中经济体缺乏数字基础设施来获得好处。
年轻软件开发者正在失业——并且速度很快。根据斯坦福刚刚发布的2026年AI指数报告,22至25岁的开发者就业人数自2024年以来下降了近20%。
ChatGPT推出两年后,丹麦大多数知识工作者已经开始在工作中使用AI聊天机器人。他们的雇主推出了正式的人工智能计划。工人们报告了真实的生产力收益。然而——这是让研究人员感到惊讶的部分——他们的薪水和工作时间基本保持不变。[事实] 这是来自[NBER](https://www.nber.org/papers/w33777)新论文的主要发现,由Anders Humlum和Emilie Vester
在AI时代,57%的美国工作在技术上可实现自动化。但麦肯锡的新研究表明70%以上的技能将保持相关。
一项覆盖四国6,000名高管的大规模调查揭示了一个惊人矛盾:AI普及无处不在,但几乎没人能量化它对就业的影响。未来三年会有什么变化?
高盛研究发现AI每月替代25,000个岗位、增强9,000个岗位,净损奁16,000个。但摩根士丹利表示对失业率的影响仅为0.1个百分点。谁说得对?
2026年首次,AI在单月内超越所有其他裁员原因。Challenger Gray报告3月AI相关裁员15,341人,占总数25%。这对你的职业意味着什么。
MIT研究人员让17,000多名工人评估了3,000多项任务。结果?没有突然的AI替代,但每年稳步提升15个百分点的AI能力可能在2029年达到80-95%的成功率。
布鲁金斯发现:1560万非学位工人处于AI高暴露岗位,他们赖以晋升的职业路径近半也面临威胁。
韩国央行调查的是真实家庭,不是企业。结果:大多数韩国劳动者已在使用生成式AI,每周节省约1.5小时,最大赢家是经验最少的员工。
韩国央行自己的数据推翻了青年失业最常见的解释。真正的原因是AI、教育差距和一个结构性排斥年轻人的劳动力市场。
韩国有5.7万AI专家,增速是可比国家的两倍。但30%的企业无法定义AI岗位,国内工资溢价仅6%而美国为25%。问题不在数量。
2015-2022年美国研究使用工具变量发现,自动化AI减少低技能岗位的就业和工资,而增强AI为高技能岗位创造新角色并提高薪酬。
一项对近10,000个埃及职位发布的研究发现,在AI自动化高风险岗位中,只有24.4%的工人拥有可行的职业转型路径。其余面临的结构性障碍无法通过简单的技能提升来解决。
沃顿商学院新研究揭示了一个博弈论悖论:企业为了降本理性地自动化岗位,但集体行动却摧毁了它们赖以生存的消费需求。UBI和再培训都失败了,只有一种政策管用。
应届大学毕业生正在苦苦求职。斯坦福说是AI的锅。但EIG的新数据显示,没有学位的年轻人同样困难——而AI暴露度高的岗位本来就没几个年轻人在做。
使用AI超过6个月的员工,成功率比新手高10%。Anthropic 2026年3月经济指数揭示了学习曲线如何正在制造一种新型职场不平等——以及这对你的职业意味着什么。
49%的职业已经在至少25%的工作任务中使用Claude。但关键是:AI正以超出预期的速度向低薪、低学历岗位渗透,而新手和老手之间的差距越来越大。
Acemoglu、Autor和Johnson认为当前AI发展偏向自动化而非增强——并提出九项政策建议将其重新导向有利于工人的方向。
Anthropic调查了132名工程师,分析了20万份Claude Code记录。AI使用率翻倍至59%,生产力提升50%,27%的AI辅助工作是全新创造的。
首个企业级研究证明AI替代劳动力是真实的。企业每在外包劳动力上削减1美元,只在AI上花0.03美元——97%的成本节省正在重塑自由职业经济。
Anthropic的印度国别简报揭示了一个惊人悖论:印度贡献了全球5.8%的Claude使用量(仅次于美国),但人均采用率在116个国家中排第101位。四个IT中心占了一半以上的使用量,45%用于软件工作。
2024年全球企业AI投资达2523亿美元,AI岗位占比创历史新高4.2%。但与此同时,整体招聘减少了140万个。斯坦福和Indeed的数据指向同一个结论:劳动力市场正在一分为二。
美国劳工统计局首次在其十年就业预测中明确纳入AI因素。我们把他们的数据和我们癇10个关键职业的AI自动化风险数据做了对比。
一项分析了1050万份LinkedIn档案和失业记录的新研究显示,AI暴露职业在ChatGPT之前几个月就开始恶化——但接受过LLM技能培训的毕业生反而收入更高。
OpenAI联合创始人Andrej Karpathy对342个美国职业进行了AI暴露评分。42%的劳动者——5990万人——处于高暴露区。这对你的职业意味着什么?
布鲁金斯学会研究发现,610万美国劳动者困在AI高暴露、低适应能力的困境中。86%是女性,集中在办公室和行政岗位。
对11个国家AI普及率和失业数据的交叉分析揭示了一个反直觉的发现——AI用得最多的国家,失业率并不最高。
ILO对138个国家2,861项工作任务的分析发现:女性主导的职业面临29%的生成式AI暴露度,男性主导的仅16%。自动化风险差距更大:16%对3%。
斯坦福和哈佛的一项78人实验揭示了"AI墙"——当你缺乏足够的专业知识来驾驭AI时,AI就帮不了你了。构思能力提升了,但真正的写作功底依然顽固地属于人类。
大多数公司打着AI的旗号缩减初级岗位。IBM反其道而行:校招人数翻三倍,每人每年必修40小时技能培训。他们的首席人力官说出了背后的逻辑。
五项独立研究描绘了一个悖论:AI在裁减岗位的同时推高工资。真正的故事是关于谁受益、谁受损,以及为什么企业在为潜力而非表现裁员。
四个独立研究来源——达拉斯联储、ADP/斯坦福、EIG和HBR——都指向AI相关职业的入门级就业下降。ADP数据显示22-25岁群体下降6%。但EIG认为下降始于生成式AI之前。
哈佛商业评论揭示了一个令人不安的模式:大公司正在基于AI预期而非实际成果裁减白领岗位。Gartner数据显示,50个AI投资中仅1个具有变革性价值,5个中仅1个实现正ROI。
Anthropic经济指数分析了超过10万次真实Claude对话。理论上1.8%的生产力提升在计入任务成功率后降至1.0-1.2%。程序员AI任务覆盖率75%,但复杂任务成功率仅66%。
PwC全球AI就业晴雨表显示:AI高度渗透行业的生产力增长是其他行业的4倍,拥有AI技能的工人薪资溢价56%。然而,AI渗透最少的职业就业增长却快了20倍。
Challenger Gray报告2026年2月裁呑48,307人(比1月下降55%),但AI相关裁员年初至今达12,304人,招聘计划同比暴跌56%。运输行业裁员飙升872%。
布鲁金斯数据显示,ChatGPT发布33个月后,AI高暴露职业的就业保持稳定。但企业77%的自动化率、早期职业脆弱性和编程的过度代表性,说明故事远未结束。
从1960年代的MDTA到今天的WIOA,美国政府的再培训项目战绩并不光彩。当AI威胁新一波失业潮时,布鲁金斯问:什么才真正有效?
用AI的企业连20%都不到。高AI暴露岗位的青年就业在下降——但失业率并没有上升。布鲁金斯说,AI劳动市场研究还在“第一局”。
ILO预测2026年全球失业率稳在4.9%,但就业缺口高达4.08亿人——与此同时AI正在重塑四分之一的工作岗位。这种脆弱的稳定对你的职业意味着什么?