The Scarce Resource
In this essay, I unpack how LeBron James managed to become a billionaire without knowing anything about Wiener Processes.
Why has LeBron James made so much more money than me? He doesn’t know much about stochastic processes, causal inference, or industrial organization—at least, as far as I can tell.
Because LeBron James is a scarce resource. There isn’t anyone quite like him. You can’t replicate him by hiring the next best alternative. There’s a huge drop-off. There are a lot of people like me. A dime a dozen in college towns. You can buy yourself a Zach on any street corner fairly cheap. He’s a commodity, really, just without enough demand to trade him on the CME.
So, you want to be the scarce resource—rare, hard-to-find, collectible.
I doubt the trick to make yourself employable in the Robot Future is to learn how to use the Robots well.
For one thing, they’re easy to use, and the Robot companies have every incentive to make them ever easier to use.
It’s easy to learn AI. The trick is to think to use it. “Oh wait, I can just type that into the computer now.”
Most of the increased productivity I’ve got from the LLMs looks like this:
“Please search the folder for bugs. Report any bugs you find. Just report them. Don’t edit the code.
<Bugs>
<I fix the bugs>
“Continue searching for bugs.”
Anyone can do this, and any problem that can be solved cheaply by a technology that everyone knows how to use will not demand much of a wage.
You want to do the opposite, of course. You want to know something that most people don’t. Call it “the return on secrets.”
So, you learn hard things, the kind of things that take years to learn, because no one wants to spend years learning. The difficulty is what makes the skill scarce.
It’s not a theoretical law—not even close—but an empirical regularity: technology tends to complement the productivity of the human input.
[Armchair theory: technology complements human productivity because humans innovate, and we create what we imagine we will use. There’s probably a very smart economic history book on this that I’ll have to read (I’m currently working my way through this one, which I highly recommend).]
To convince yourself this is still the case today in the world of LLMs, take the problem space they’ve improved the most: computer programming. If you’re not a programmer, try to write an app by burning Claude credits. Then, have a programmer do the same thing, but they can only use Claude prompts. The programmer’s result will be so much better that you will wonder whether you even used the same tool. It’s not because they know how to use AI better. They know how to program, so they’ll create better programs, regardless of medium.
It turns out that it still pays to know something—which sucks, of course.
Thanks for reading!
Zach
Connect at: https://linkedin.com/in/zlflynn