How AI Will Erase Entire Industries Without Automating Them
Notes on the shape AI displacement will actually take
Hey there, I’m Alberto! 👋 Each week, I publish long-form AI analysis covering culture, philosophy, and business for The Algorithmic Bridge. Paid subscribers also get Monday news commentary and Friday how-to guides. I publish occasional extra articles. If you’d like to become a paid subscriber, here’s a button for that:
I’ve been reading a few things this week that kind of clicked together—featuring David Oks, Nat Eliason, Andrej Karpathy, and Dwarkesh Patel—and I wanted to share my notes before they get stale. What follows is more thinking-out-loud than a polished argument. Will write more once the ideas here take a definite shape.
I. ATM THINKING / DAVID OKS
David Oks, a researcher at Andreessen Horowitz that I’ve found recently on Substack, published an essay a few days ago about bank tellers and ATMs that actually laid out the future of AI’s impact on jobs.
You know the parable: ATMs were supposed to kill teller jobs, but they didn’t. Oks’s first contribution is pointing out that another technology did kill them: the iPhone. ATMs automated tasks inside branch banking, but branch banking stayed, so tellers stuck around (they just became salespeople). The iPhone made branch banking irrelevant. When you “carry a bank branch in your pocket,” as Bank of America’s CEO said, there’s no branch. When there’s no branch, there’s no teller. Oks recounts that Bank of America closed 40% of its branches between 2008 and 2025. The trend is generalized.
Oks frames this as a general principle: automating tasks inside an existing paradigm rarely displaces workers. What displaces workers is when someone builds a new paradigm where those tasks don’t need to exist. The ATM tried to do the teller’s job, but the iPhone made the teller’s job pointless. (Oks emphasizes the distinction between automation and irrelevance, which is the point where most people get stuck: it’s easy to visualize a world where your job has been automated, but really hard to visualize one where your job is irrelevant because the entire ground has transformed.)
I think this is the right frame for thinking about AI and work. The default way people think about AI and jobs is what we might call “ATM thinking,” to continue with the analogy. You have a company, it has people doing things, and you ask: Can AI do some of those things? You slot an AI into the existing workflow and, assuming you know how to make it work—which is itself not trivial—it will start enhancing existing tasks. Maybe it writes more emails or the same number but faster, or generates code that’s almost production-ready, or handles first-pass customer support, whatever.
This is the “drop-in remote worker” vision, and it’s where most of the industry’s attention goes—because that’s what the industry can visualize as the next step!
And it mostly doesn’t work that well. One reason is that AI is unreliable and not that good and not that easy to make work, but another reason is that companies see some productivity gains, but can’t really reduce headcount as much as expected. Oks argues this is predictable because you can’t fit capital into “labor-shaped holes.” You hit the frictions and inertias of the world that I’ve written about many times. The edge cases still need humans. The institutional structure resists change. As long as the paradigm stays the same, the humans stay too (more or less).
A world made by and for humans will find a way to need humans (we have plot armor!); at most, you move the bottlenecks around while keeping progress advancing at the speed of meat.
The much more interesting way to leverage AI is by skipping all of that and building a new structure around AI from scratch. AI as the new paradigm; AI as manufactured irrelevance. That’s what writer and entrepreneur Nat Eliason is doing as we speak.
