I saw this: “AI Is Looking More and More Like a Normal Technology,” by Richard Hanania.
This is a great article, and it agrees with what I’ve been saying since 2023 or so: AI will not lead to meaningful, if any, net job loss. It also agrees with Freddie deBoer’s position, and he has been right since he began writing about this around 2024. Our views represent a tiny minority among “experts” who have staked their intellectual reputations on the certainty of AI leading to a job apocalypse rivaling even the Great Depression, yet so far this has not been borne out.
There are sooooo many jobs AI cannot yet automate–things like service sector work, blue-collar jobs, or individualized services. Many white-collar jobs involve coordination and dealing with edge cases. Things like insurance or advertising require meeting clients’ unique needs or dealing with counterparties, like cross-examining a witness in a courtroom or assessing an insurance claim. If an insurance company does not pay out, you are not going to resolve this with AI; you’re going to be making some phone calls. So this represents a huge chunk of jobs…roughly 30% of workers have zero measured AI exposure, according to Anthropic’s own research, and many more are only lightly exposed.
Then you have white-collar jobs. This is similar to the lore of the “nonviolent offender who is serving a huge sentence” used to justify prison reform: the argument depends on there being far more people who fit the description than there actually are. There just aren’t that many white-collar workers who can just be replaced. The crux of the AI “doomer” argument (I hate this word, but I cannot find a better one) hinges on the existence of huge swaths of the workforce that can be replaced with AI. In actuality, there just aren’t that many, or as many as needed to lead to the scenarios forecasted. A lot of white-collar work involves coordination and edge cases that AI cannot yet handle, or dealing with the unique needs of customers or clients, as mentioned above.
To echo Mr. Hanania, AI is amazing at math or coding, but destroying a job takes much more than merely doing well at individual tasks when instructed to. A job requires–to borrow a buzzword–“agency.”
On a related note, Zvi and Patio11 ask in a podcast why AI hasn’t changed the world more, or why what they call “diffusion” has been so slow:
Zvi: Absolutely. I think so far, I would summarize it as: I have been surprised the models have gained capability as quickly as they have so far. And, like everybody else, I have been surprised, conditional on how much capability they have gained, at how little the world around us looks different as a result of that.
Diffusion has been slower than I would have expected, but ultimately that doesn’t matter. And as the models themselves are capable of driving the diffusion, the diffusion will speed up. So yes, expect high weirdness.
Or about jobs not being replaced despite the power of AI:
Patrick: I am not at liberty to say who it is yet, but I can say I’m aware of at least one person who spends more than a million dollars a month in inference for the acceleration of their professional job, and whose employers look at their inference spend every month and are like, “Great.
That continues to grow. Wonderful.” And there don’t have to be all that many people in the world hitting a million dollars a month in spend that they can personally authorize for these to be wild, wild businesses. And again, the job of this person in the status quo resembles the job that they were doing two years ago, except they’re now doing just a whole lot more of it, because they have teams of minions. Virtual minions.
From my own experience, a major reason is integration: the pieces each work well on their own, but connecting them is hard. For example, uploading files to a public GitHub repository is baffling to a new user because the interface is cluttered. A month ago, I set up a GitHub repository for a trading system I was testing, in order to generate backtests. Having never used GitHub before, I asked AI to help. Even with the latest Ultra Astra model, I got incorrect instructions because the interface had changed and the advice was dated.
It eventually took me 20 minutes to set it up, and I still need help making changes. An expert could have told me how to do it in a minute or done it himself; ergo, an employee. This is why coding jobs won’t just “go away” despite AI progress. There is always some bottleneck where two technologies, such as AI and GitHub, do not work seamlessly together. Clumsy interfaces improve much more slowly than AI.
Another example is the so-called “walled garden,” where platforms restrict access. This renders AI crawlers useless on those sites and dampens productivity, as these tools cannot be used to their full potential. It is perfectly rational for sites to block AI crawlers: they get no compensation while paying server costs and opening themselves up to spam and exploits.
Indeed, AI was unable to create the GitHub account for me. Here was Claude’s response (on the latest model, no less): “I can’t create the GitHub account for you. Signing up means accepting GitHub’s terms, verifying an email address and passing a CAPTCHA, and you need to do those yourself because the account is yours.”