I think the editors of the Washington Post are failing to make an important distinction in their new editorial on the implications of LLM AI on “white collar” jobs. That distinction is between what generative AI is actually likely to accomplish in the near term and how it is being sold by AI service providers and how American managers are interpreting that. The editors remark:
Artificial-intelligence catastrophism is everywhere. This month saw stock-market ructions after Anthropic, a leading AI company, announced that it had developed and deployed AI “agents” to autonomously execute legal, marketing and sales tasks. The economic potential is real, but the hand-wringing is overblown.
Yes, AI appears to be a transformational technology, and that transformation will sometimes be disruptive. But it won’t render humans obsolete, and the disruption of white-collar work might do a lot of good.
The editors treat AI as a technological phenomenon whose labor effects depend on capability. But corporations do not wait for capability; they act on forecasts. Whether those forecasts are accurate is economically secondary. The layoffs occur at the moment executives believe the technology will soon substitute for labor not when it actually does.
Even if the editors are technically correct, their conclusion about disruption being overblown is wrong because economic behavior precedes technological reality. The economic impact of AI will be driven less by actual technical capability and more by managerial belief in vendor claims. Consequently, job loss can occur even if the technology is immature.
Just a few days ago Microsoft AI CEO Mustafa Suleyman predicted that AI would automate most tasks in white collar jobs with human-level performance within 18 months. Anthropic CEO Dario has repeatedly warned that AI would eliminate half of all entry-level white collar jobs within the next 5 years. IBM CEO Arvind Krishna paused hiring at IBM, explaining that a large number of jobs will be “replaced by AI”. Those are the sales pitches.
How these sales pitches are being interpreted and utilized by Fortune 500 CEOs is equally clear. In 2024 Carol Tomé, CEO of UPS, laid off 14% of the company’s managers, explaining:
We will constrain head count growth through the end of the year, in addition to a limited reduction in roles across the firm… These targeted steps are consistent with our priorities of gaining more agility and creating the right team structures in order to implement effective AI solutions
This year the company has already laid off a significant number of non-managerial workers due to loss of contracts. The contracts were lost in large part due to AI. The CEO of the financial technology company Klarna has frozen hiring—AI can do all of the jobs. Last year David Solomon, CEO of Goldman Sachs, announced a hiring freeze due to AI. I cannot distinguish between the actual results of AI and confidence in the future promise of AI but these CEOs are responding to the sales pitches by improving their companies’ bottom lines by reducing headcount. In the final analysis it doesn’t make any difference whether they’re right or wrong. The jobs are gone in either event.
That’s not alarmism. Those layoffs and freezes are cold, hard facts with serious real world implications for young people who now have thousands of dollars in student loans in anticipation of jobs that are not materializing.
Update
Mohammed El-Erian echoes a number of the things I’ve said above at the Financial Times:
There are also reasons to believe that this period of decoupling of employment from growth may prove more persistent and more consequential.
This time around, it may well last longer because we are just at the start of the AI adoption process, with robotics just around the corner and quantum computing further behind. Moreover, the current mindset of many firms in their initial consideration of AI does not help. Too many executives seem to think of AI more in terms of its labour cost minimisation potential (doing the same with fewer workers) rather than the bigger productivity potential (doing more with the same or additional workers). The latter comes with increasing the capabilities of existing and new workers.
This decoupling of GDP growth and employment also comes at a time when affordability is already a big political and social worry. An intensifying gap between robust growth and a weak labour market would likely increase income and wealth inequality in an economy already featuring a large divide in the fortunes of the wealthier and less well-off.
This would undermine low-income household consumption as an important driver of growth. It is also taking place at a time when the US Federal Reserve, already under political scrutiny, now faces the prospects of a bigger conflict between the two components of its dual mandate — maximum employment and price stability.
The most dangerous four words in economic and investment analysis — “this time is different” — could well end up applying here. Rather than gradually slowing to a crawl as has happened in the past, we should expect this decoupling phase to accelerate absent any holistic mitigation by companies and the public sector. Indeed, part of the challenge for 2026 is managing a lot better the economic, political and social risks of an economy that, without corporate and policy adjustments, may no longer need as many workers to grow.
Something unmentioned by Mr. El-Erian is that consumption spending by higher income individuals comprises an unusually large proportion of the whole. If job freezes continue to transition into actual job cuts, that could have serious implications for the greater economy.
The theory of creative destruction suggests that new jobs will come along for those who are displaced, like happened with agriculture and other changes. What it doesnt account for is how long it takes to adapt and develop those new jobs. Looking back at the history of the Industrial Revolution it looks like that can take quite a while and it can be pretty rough for those who have been displaced.
Steve
There are more master weavers now than there were in 1800. The original Industrial Revolution temporarily eliminated a relative handful of jobs, created far more, and improved well-being enormously. As I’ve illustrated in the past the Digital Revolution eliminated some jobs, created others. So far the numbers are about equal. The AI Revolution will produce a relative handful of jobs and eliminate millions of jobs. It may improve well-being but it won’t be a net job creator.
Many of today’s jobs were not possible in 1980, and many of the jobs in 2070 are not possible today.
The AI world being envisioned cannot exist. The economy and stock market is based upon expanding debt. If enough workers are unemployed or underemployed, debt cannot expand, and the economy will slow or shrink.
At that point, many of the products and services produced by AI become superfluous. Company profits will dwindle, and the company, along with its products and services, will cease to exist.
Eventually, AI will be out of a job.
Basically, the world being envisioned is the same as today but with fewer jobs. Somehow people will have no income, but GDP growth will continue. Fewer products will be purchased but profit margins will increase. Debt will be increased, but people will not be able to repay it.
Facebook produces no tangible product, and it relies on luxury products for jobless people. When people cannot buy electronic devices, they cannot use Facebook. When Facebook loses most of its users, its stock price will plummet. When its stock price plummets, Mark Zuckerberg’s wealth will plummet. He may not become a pauper, but his world will change, dramatically.
I just don’t get it.
TastyBits:
They’re taking a strictly IBGYBG strategy.
Dave Schuler: The original Industrial Revolution temporarily eliminated a relative handful of jobs
The Industrial Revolution eliminated most jobs. In 1800, three-quarters of Americans worked in agriculture. Now, it’s only a small percentage. That’s most jobs right there. That doesn’t include craft jobs that were also eliminated.
What happened is that a human-operated steam machine replaced a number of manual workers. The guy who knew how to operate the machine still had a job. The disruption absolutely caused hardship. In the long run, though, increased productivity meant the development of new products and new services (telephones, automobiles), or products that were once out of reach became available to the masses (textiles, clocks).
Similarly, it is quite possible that AI will develop along similar lines, with AI-assisted workers replacing a number of other workers. So somebody will still have a job. The problem is the rapidity of change. The faster the change, the more disruptive, and the longer it will take to correct itself.