The Flexibility Trap
Why so much of the economy isn't already automated
The first “white collar” job I ever had sucked.
I did data entry for a local retailer. I would take a registration card someone had filled out in a store and input it into a spreadsheet.
For a 4 hour shift, 3 times a week, I would squint at someone’s sloppily written email address and type it in.
I hated it. Like a lot.
So much so that I found a very primitive app that would let me take a picture of the card, and it would transcribe it for me. Then I figured out enough code to build a macro that would input that form into the spreadsheet. It was bad and clunky and only I could do everything just right to get it to work, but it turned a 4 hour job into more like 20 minutes.
At one point I asked my boss why a software couldn’t just do this, and her incredibly candid answer was that it was likely cheaper to pay me $10.50/hr (!) than it was to buy a software or hire someone to build one.
Today an AI would do that job in seconds, for pennies.
Automation has always demanded a bargain: pay a lot upfront, save a lot later.
You buy the robot first. You install the software first. You build the factory line first.
The capital expenditure is enormous, but once it’s running, each additional unit costs almost nothing. High leverage — but only if volume is predictable and demand is stable.
Human labor offers the opposite deal. You don’t buy a person. You rent them.
Wages are a recurring cost, and each hour of output costs roughly the same as the last. It’s expensive per unit, but it’s flexible.
You can hire a college kid for a semester and let him go when the registration drive ends. You can redirect a worker from one task to another with a conversation. You don’t need to know the future to justify the investment, because there isn’t much investment to justify.
At $10.50 an hour, I was cheaper than any software. Not because I was better (I was dramatically worse) but because I could be turned on and off like a spigot.
No contract. No implementation. No risk.
This tradeoff has quietly governed which parts of the economy automate and which don’t since the dawn of automation.
High-volume, predictable demand? Automate.
Volatile demand, variable tasks, tight cash flow? Keep humans.
Over time, this has created a sharp divide between large and small businesses.
Big companies are good at forecasting and they have the resources to absorb the risk if they’re wrong. So they automate.
Small businesses are more cash-constrained, less able to predict next quarter let alone next year. They mostly don’t automate. They hire people instead.
And this matters more than it might seem: the vast majority of American businesses are small businesses and close to half of all American work for small businesses. Huge portions of the economy have remained essentially un-automated not because the technology didn’t exist, but because the economics never made sense at that scale.
The result is that human labor hasn’t just survived alongside automation — it’s been structurally preferred across much of the economy. Not because people are better at every task, but because flexibility has real economic value. The ability to scale up and down without sinking capital is worth paying a premium for.
This is the part most people skip when they talk about automation replacing jobs. It’s never been a simple question of capability. It’s a question of cost structure. And for decades, the cost structure of automation had a built-in limit: it only made sense where the math was predictable.
AI breaks the math.
Not because it’s smarter than previous automation (though it often is), but because it has a fundamentally different cost structure.
There is no robot to buy. There is no factory line to install.
You pay per token, per task, exactly when you need it, scaled precisely to what you need.
The startup cost approaches zero. The marginal cost is low. And critically, you can stop paying the moment you stop needing it.
This is new. Not “incrementally better” new, but “changes which parts of the economy are vulnerable” new.
Consider a small legal practice. They could never justify a custom document automation system — the capital outlay doesn’t make sense for their volume. But an LLM that drafts contracts at pennies per page, with no commitment beyond maybe $20/mo? The math suddenly works.
AI doesn’t primarily compete with human labor where automation already won.
It competes with human labor where human labor was specifically preferred because of its flexibility.
The traditional refuge of human labor — “we’re more adaptable, we’re more flexible, you don’t need to make a big bet on us” — starts to erode when the alternative also doesn’t require a big bet. AI-as-a-service scales like a temp agency. It carries no fixed overhead.
This doesn’t mean universal displacement.
AI still can’t do physical work, can’t navigate novel social situations, can’t exercise the kind of judgment that comes from actually caring about an outcome.
The flexibility of a human being extends far beyond their cost structure — it includes creativity, empathy, improvisation, and the ability to do things no one anticipated needing done.
But the economic moat is narrowing.
And if history is any guide, when the math changes, the economy follows. Not overnight. Not evenly. But eventually, and in ways that the people inside the transition rarely see coming.
The question isn’t whether AI is as good as a person. The question is whether the reason we were paying for a person was ever really about the person at all.
Originally published on Griff’s Substack on 12 March 2026.