Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.
The Substitution Threshold: When You Stop Being the Cheapest Option
There is a single, unsentimental line that decides whether you have a job. The book calls it the Substitution Threshold: the moment when the cheapest reliable provider of a given service is no longer human.
Notice what the definition does not say. It does not say “when the robot is smarter than you.” It does not say “when AI achieves consciousness” or “when the machines wake up.” It says cheapest and reliable. Those are accounting words, not science-fiction words. The threshold is crossed by a spreadsheet, not by a singularity.
It Crosses One Job at a Time
The reason the Substitution Threshold is easy to miss is that it does not arrive economy-wide on a Tuesday. It arrives one kitchen, one help desk, one warehouse at a time, and each crossing looks local and survivable until you connect the dots.
In the book, Maria Delgado loses her cleaning job not to a policy or a recession but to a robot that does not get tired, does not need health insurance, and does not have a seven-year-old daughter. For her, the threshold crossed in one specific household on one specific morning. For the economy, that household is a leading indicator.
The receipts are already arriving on company letterhead. In February 2024, Klarna announced that its AI assistant had handled two-thirds of its customer-service chats in its first month, the work of about 700 full-time agents, worth an estimated $40 million a year. Those agents were not retrained or promoted; by the arithmetic the public numbers imply, they were mathematically obsolete. Then came the second act. In May 2025, after customers complained of generic answers and an assistant that lost the thread on complicated cases, Klarna started hiring people again, and its chief executive promised there would “always be a human if you want.” The machine did not retreat. By that November the company put its assistant at the work of 853 agents and $60 million saved. That is the threshold behaving exactly as a serious automation theory predicts: it crosses first for the repeatable tier-one work, keeps advancing there, and lingers, for now, on the cases that need someone who can read between the lines.
Why “Reliable” Is the Load-Bearing Word
A lot of comfortable arguments hide in the gap between “cheaper” and “reliable.” For years the honest defense of human labor was that the machine was cheap but flaky. The Substitution Threshold is the point where that defense expires, where the machine becomes not just cheaper but dependable enough that a rational manager stops keeping a human around as insurance.
Watch the cost curve and you can see it happening. Stanford’s AI Index found that the price of running a model at GPT-3.5’s level fell more than 280-fold between November 2022 and October 2024: what cost $100 cost about 36 cents two years later. The home humanoid sliding into Maria’s old job is listed at $499 a month, which works out to roughly $0.69 an hour if it runs around the clock, against a federal minimum wage of $7.25, though as of mid-October 2026 no customer delivery had been independently verified. (The full hardware accounting lives in the humanoid-robot trajectory.) When the cheaper option also clears the reliability bar, the threshold is no longer theoretical for that task. The math wins, and the math does not care about your politics.
The vendors know where the line sits, because they price against it. The prospectus for the merger taking Agility Robotics public, filed in September 2026, states the rule in one clause: its warehouse humanoid rents “at a discount to a customer’s fully burdened labor cost.” That comes to $8,500 a month for a machine whose build cost is falling from $150,000 toward $30,000. The final prospectus of October 13 keeps both numbers, and the company expects to trade as AGLT in November. The robot does not have to be cheap. It has to be a little cheaper than you. Crossing the threshold and passing the savings along are separate events, and the gap between them is where the fight over who benefits gets decided.
The spreadsheet also runs the other way. The Wall Street Journal reported in October 2026 that Walmart’s automated grocery warehouses draw up to $800,000 a month in electricity, against about $250,000 when people did the work, and that the engineers who keep the machines running cost more than the warehouse workers they replaced (Supermarket News summarized the report). A robot that stalls on an oversized cardboard box has not crossed the threshold, whatever the demo showed. Cheapest is a total-cost word, kilowatt-hours and maintenance included, and reliable is measured in a warehouse, not on a stage.
In licensed work, “reliable” carries a second meaning that has nothing to do with accuracy: someone specific absorbs the loss when the answer is wrong. That is what a malpractice premium buys and what a terms-of-service page does not, and it is why a chatbot can already out-read your doctor and still not replace them. The liability gap is where that distinction gets tested. The mirror image matters just as much: where the output can be checked mechanically, as a Lean proof or a passing test suite can be, reliability stops needing a credential at all and price decides on its own. That is the verifiability premium, and it explains why research mathematics felt the threshold before most clerical work did.
The Threshold and the Cliff
The Substitution Threshold is the micro-event. The Labor Cliff is what you get when you add up millions of those micro-events across an economy.
The connection is a flywheel running in reverse. Our economy is a loop: people work, earn wages, spend those wages, and companies use that revenue to pay more workers. Each crossing of the Substitution Threshold removes one worker from the “earn wages” step. One crossing is absorbed easily. Enough of them, and the flywheel loses torque outward, the displaced first, then the people who used to sell to the displaced, and on through the ring around that, until a corner grocery that used to close at nine closes at six. The cliff is not a wall everyone hits at once. It is the threshold, crossed enough times, unwinding the loop link by link. The count is already visible in the layoff notices: Challenger, Gray & Christmas records employers citing AI in 120,136 US job cuts through September 2026, about a fifth of the total and the year’s leading stated reason.
This is why the book insists the problem is arithmetic rather than moral. Companies can produce infinite goods with machine labor, but machines do not buy groceries or movie tickets or their kids’ birthday presents. The question nobody in power wants to touch is the one the Substitution Threshold forces: when a large share of the population is below the line where their labor is the cheapest reliable option, who buys the products?
What It Demands of Us
Naming the Substitution Threshold is not doom-mongering. It is the opposite of denial. Once you accept that the line is real and that it moves in only one direction, the policy debate stops being “how do we retrain our way around it” and becomes “what do we build for a world on the far side of it.”
You cannot out-train a falling cost curve. Every new skill you acquire becomes a new task on the wrong side of the threshold the moment a model gets cheaper at it than you are. So the book’s answer is not to race the machines back up the ladder. It is to decouple survival from labor entirely through the Foundation, and to decouple meaning from employment through Impact. The Substitution Threshold is the diagnosis. The rest of the framework is the prescription.
The threshold is a tool, not a tragedy. It tells you exactly which jobs go first and roughly when, which is precisely the information a civilization needs if it intends to build the floor before the people land on it.
Related Articles
- The 2025-2030 Labor Cliff: What the threshold adds up to when it is crossed across a whole economy.
- Anticipatory Displacement: Why employers cut jobs before the machine has actually crossed the line.
- Wage-Anchored Pricing: What happens after the crossing, when the machine is priced against the wage it replaced.
- The Liability Gap: Why licensed work crosses late, and what a license really sells.
- The Verifiability Premium: Why work that can be checked mechanically crosses early.
- Human Reserved: Whether a job can be fenced off from the threshold by law.
- The Vanishing Apprenticeship: The same threshold one level up, where the first rung of a career goes before the career does.
