Will you still have a job in 2030?

Why AI might not replace us after all

We could automate many aspects of food service, from ordering to paying. But people like the personal touch.

When I quit my job at The Atlantic to start The Argument, I used AI every day to help me figure out how to build a new company. I had no idea how people raised money or what kinds of lawyers I should be looking for or which forms of liability insurance were really necessary.

In bygone years, I would have probably just done some reading and made an educated guess here or there. But in this age of large language models (LLMs), I constantly turned to my colleague ChatGPT for help setting up the new business.

It turns out that I am not the only American to have had my entrepreneurial ambitions nurtured by an LLM. Ernie Tedeschi, Stripe’s chief economist who formerly served on the Biden White House’s Council of Economic Advisers, recently pointed out that AI is likely fueling solo entrepreneurs like yours truly:

Source: Stripe

All of this might make you think that I’m bullish on AI eliminating large swaths of white-collar jobs. But, in reality, it has made me incredibly skeptical.

According to The Argument’s recent polling, one of the most widespread fears about AI is unemployment — 70% of respondents said AI may cause large-scale job losses within the next five to 10 years.

It’s very hard to know what people consider to be “large-scale job losses.” In a healthy economy, there are nearly 2 million layoffs every month.

The reason this doesn’t turn into a full-blown recession is that many millions more are being hired at the same time.

So when people express worry about the impact of AI on the labor market, are they making a narrow point about the disruption they fear in their industry, a broader point about expecting to see a lot more layoffs, or an apocalyptic point about how AI is going to usher in a new age of mass unemployment?

One of the biggest AI doomers — or optimists, depending on how you look at it — is Anthropic CEO Dario Amodei. Amodei has repeatedly said that he believes U.S. unemployment could reach 10% to 20%. For reference, in April of 2020, when the whole world shut down due to the spread of COVID-19, the unemployment rate didn’t exceed 15%. Great Depression joblessness peaked at 25%.

So what Amodei is talking about is more than the usual labor market churn. And others who expect massive disruption have forecast something even scarier than high unemployment: that human labor will become a minor input into production, making most human beings unnecessary for the production of most goods and services.

I feel very skeptical about all this for a few reasons:

1. A job is more than the sum of its tasks. AI can automate a lot of tasks, but that doesn’t mean it can automate an entire job. AI could help me come up with a development pitch, help fact-check my articles, sort through my email, and produce slide decks for meetings.
   
   It can even do many of these tasks as well as — or better and faster — than I. And yet, an AI could not run The Argument.
   

2. AI diffusion will be slowed by the same things that slow high-speed rail. It is technically possible to have high-speed rail from San Francisco to Los Angeles. It is technically possible for nurses to prescribe Schedule II drugs safely. It is technically possible to build bus rapid transit infrastructure.
   
   And yet, in many places where this would easily and quickly improve the lives of millions of people, we don’t make these improvements. Not because we don’t have the money, but because lots of stakeholders don’t want the current economic arrangement to change; they are benefiting from it.
   

3. People like other people. There are cheap, automated versions of lots of things. There are vending machines for books and food, and there are QR codes for menus. But the preference for a human being to do stuff for you continues to provide demand for jobs. One fact that I think speaks to this at another level is that in May 1997, IBM’s Deep Blue supercomputer defeated world champ Garry Kasparov in chess — and yet chess has never been more popular.

This is not to say that people should calm down.

As I have written before, the economic growth spurred by the printing press was costly: “For hundreds of years following the invention of the printing press, millions of people died in religious conflicts spurred by the Protestant Reformation. The rise of the nation state enabled the mass slaughter of millions in world wars, not to mention brutal colonial empires. Modernity and the industrial revolution came after decades of conflict over who got to control the canonizing institutions.”

But the policy responses, while politically difficult, are extremely straightforward: Expand unemployment insurance, invest in job training and matching, conduct expansionary monetary policy, and make other necessary investments in the social safety net.

My podcast cohost and The Argument columnist Matt Yglesias thinks I’m being way too complacent, so I brought on our mutual friend and first-ever guest of the pod — Understanding AI’s Tim Lee — to browbeat him with me.

Watch or listen wherever you get your podcasts.

The Argument. Libbing out.

The transcript will be after the paywall in this post for paying subscribers.

WATCH THE EPISODE ON YOUTUBE HERE

New episodes post every Thursday.

For an ad-free version and full transcript, subscribe at TheArgumentMag.com.

Correction:

• Around 00:28, Tim describes memos sent to Starbucks staff “once about a decade ago and then once again more recently.” The recent memo was sent in 2025, but the previous one was sent in 2007, which is closer to 20 years ago.

Show notes:

• R.U.R. (Rossum’s Universal Robots), 1920 play by Karel Capek, referenced by Matt, which is widely credited with introducing the word “robot” to the English language: Project Gutenberg page

• “Why AI hasn’t replaced software engineers, and won’t,” article referenced by Tim Lee discussing the concept of jobs as a sandwich, of which AI can only replace a part: AI as Normal Technology article

• “The Case Against the AI Job Apocalypse,” podcast episode in which economist Alex Imas draws an analogy between AI and a world-class chef, whose perfect work would be ruined if it were oversalted at the end: Plain English episode

• METR chart referenced by Tim and shown in the episode, showing the growing length of task (based on how long it would take a human) that AIs are able to accomplish with reasonable accuracy: METR page

• “Canaries Dashboard,” project by Stanford Digital Economy Lab using ADP Research data to show changes in employment of AI-exposed industries. It shows early career software engineering jobs dropping and early career home health aide jobs rising: Stanford page

• “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” study by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen showing relative employment decline for young workers that didn’t affect older workers: Stanford research

• “The Broken Ladder: AI, Remote Work, and Early-Career Hiring,” study by Peter John Lambert and Yannick Schindler suggesting the decline in employment among young workers could be attributed to the growth in remote work: SSRN page

• Average Is Over, book by Tyler Cowen discussing the decrease in appreciation of competent, but not extraordinary, people: Goodreads page, Amazon page

• Peer Review: “Zoning: Externalities or Misallocation?” paper by Yu-Hsin Ho, Chang-Tai Hsieh, Wen-Tai Hsu, and Yu-Jhih Luo using data from Taipei to show that mixed-use development brings economic advantage: NBER working paper

• “Why Do Rich People Love Quiet?” article by Xochitl Gonzalez arguing that an expectation of quiet is at odds with minority cultures: The Atlantic article

Transcript: