Employers are under tremendous pressure to embrace AI and ditch the workforce. Investors and CEOs are obsessed with reducing costs and maximizing opportunities; Every CIO is pushed to come up with an AI plan, to keep pace with competitors. Dreams of AI-driven transformation are everywhere.
But leaders should not feel that they have to rush to accept a future that already exists. There are many reasons to be cautious. Here are the nine:
The “experts” have often been very wrong in their predictions. Nobel laureate and AI pioneer Geoffrey Hinton said in 2016, “People should stop training radiologists now… But few if any radiologists have been replaced ten years later. Google co-founder Sergey Brin promised in 2012 that self-driving cars would be everywhere by 2017. Today, 14 years after the promise that (and many following Elon Musk), fully autonomous vehicles are still a limited experiment, available only in a few cities with a good climate.
Big Tech wants you to believe that it has created artificial general intelligence. That doesn’t make it true. When tech CEOs warn of a career Armageddon, they may be covering their bases if that happens, but then again, maybe they just want you to raise their company’s values. Take every guess they make with a grain of salt.
When it comes to job impact, the AI giants’ numbers don’t back up their claims. Anthropic’s CEO has been warning about the job, but recent Anthropic research has shown a gap between perception and reality. The company offers great opportunities for that AI power do in fields like finance and architecture. But what it called “supervised AI” (a fancy term for what’s happening in the real world) has made up a surprisingly small part of that theoretical achievement. What they think AI can do and what it actually does are light years apart.
Current AI is “jagged” (good at some things but not others), meaning it can rarely replace a human. AI can certainly help the productivity of some workers, but even in jobs where AIs work well, models and agents often make silly mistakes, some of which are hard to detect. And jobs aren’t jobs: Even if AI can do a part of a human’s job, it doesn’t mean it can do all of that person’s job.
Current AI models still struggle to get beyond language. Some white-collar jobs involve only words, but many involve visual understanding: interpreting pictures, charts, diagrams, plans, maps, etc. It may seem easy to think of AI taking over every job, especially if that job is some kind of magic. But once you realize that modern AI is a tool, with strengths and weaknesses, you begin to realize that technology can displace workers from some jobs and not others (and will often increase human jobs). Even in areas like customer service that may seem obvious, the results are often disappointing. The Remote Worker Index focused on tasks that could be fully accomplished online, and found that less than 4.5% could be adequately completed by AI agents.
Most of the physical work is beyond what current AI can do. Don’t expect AI to replace plumbers, carpenters, auto mechanics, nurses, house cleaners, rangers, chefs, repairmen, gardeners, or many other jobs anytime soon.
Many of the gaps attributed to AI are not actually related to AI. That may have been the case for the recent crowdfunding on the fintech Block; others saw this as CEO Jack Dorsey’s attempt to regain investor confidence after its stock slumped. In many cases AI may act as a fig leaf to cover jobs that are actually driven by financial inefficiencies or earlier hiring.
Some AI-induced declines are not sustainable. I call this the Klarna Effect, after buying now, the latest company of Klarna, which has proudly done great AI jobs to bring it back. Many of the people who were laid off worked in customer service, but after 11 months Klarna decided that (at least in some cases) “real people” were needed at all.
The overall impact on productivity and return on investment in AI so far has been modest. Every company is investing in AI, but so far many are not getting much profit.
All this is subject to change; maybe one day it will – but probably not until we see major advances in AI, which could be a decade or more. In the meantime, simple advice: Don’t focus on replacing people. Focus on how you can use AI to help those around you.
Gary Marcus is a distinguished professor of psychology and cognitive science at NYU, and the author of six books, including Taming Silicon Valley.
This article appears in the April/May 2026 issue of Good luck titled “9 Reasons We Shouldn’t Be Scared (yet) About AI.”
#Reasons #Wont #Job #Good #luck