Compensation, capital and counting.
When measurement is expensive you pay for the person. When measurement gets cheap you pay for the work.
Walk into any grocery store in America and you’ll find self-checkout lanes. And cashiers. That’s a rational economic calculation made by organizations that looked at full replacement and decided the math didn’t work. Shrinkage goes up. Cart abandonment goes up. Accessibility becomes a liability. It turns out that the unmanned checkout experience has costs that don’t appear in the labor line.
The tech didn’t fail. The full cost of operating without humans ended up meaning that the humans in stores were worth it at their labor cost.
The price was right. The price is about capital. Capital doesn’t flow to importance. It flows to returns. Those have historically overlapped enough that we conflate them but they are not the same thing.
For the better part of a decade the most richly valued work in the American economy was software that helped midsized companies manage sales pipelines or sell ad attention at scale. Those weren’t necessarily civilization defining solutions. But recurring revenue and high gross margins produced returns that capital liked. SaaS founders got richer than energy founders not because software was more essential to human life but because software had better margin structure. The market was pricing the packaging, not the problem.
Right now capital is flowing to civilization scale problems and the money is real. Fusion, longevity, frontier energy, even AGI. But these problems didn’t suddenly become more meaningful. Their returns got more defensible.
The pricing of work has always been about returns too. In most knowledge work the returns were clear at the company level and less precise at the individual one.
AI is dismantling the pricing logic that protected certain kinds of work from direct measurement. The output of knowledge work was often the product of a team, relationships and complex chains of decisions. You couldn’t follow the work precisely from effort to outcome so you priced the person instead. The price was set by what it would cost to replace them and paid out against time like hourly rate or yearly salary. Rational but not strictly outcome driven.
AI makes knowledge work production costs visible. The path from input to output goes from obscure to traceable. When the input cost becomes transparent the question of what the output was worth gets much harder to avoid. You can see exactly what was produced and what it cost you to do so. And what 10x more would cost you. For the first time there’s a token price on a unit of knowledge work.
So the fully loaded cost of knowledge work becomes two line items. The person and the compute. An engineer’s cost now includes their compensation and their tokens. A marketer includes theirs and the models producing their creative. Unlike a software subscription seat, tokens scale with what gets produced and each call returns a discrete result you can evaluate on its own. Right now compute is cheap because investors are subsidizing adoption. The point isn’t the ultimate ratio. It’s that now the new math exists.
When the result is traceable and the cost is known the market finds a way to price the outcome instead of the person.
This logic is thousands of years older than the technology. Ancient Mesopotamia had brokers working on commission. Before 1980 truckers were on weekly union pay because what happened between terminals couldn’t be cheaply observed. Deregulation and electronic logs made miles, hours and safety events measurable so pay moved to outcomes in cents per mile. Real estate agents aren’t getting salaries any day now.
Outcomes based pricing already existed wherever the work could be easily measured. AI is widening what can be measured.
Supply and demand still runs within that logic. Top talent will still command top compensation and an attractive package will still include a big salary or other guaranteed pay. Where companies are competing for talent, leverage holds the old structure firmer in place.
The jobs most exposed to this shift aren’t vulnerable because they lack value. They’re vulnerable because the work can now be priced on output instead of replacement cost. Every compensation structure and headcount justification in knowledge work was built on the assumption that individual contribution couldn’t be cleanly isolated. Organizations didn’t choose ambiguity. They built on it because there was no alternative. Now they have an (imperfect) alternative.
The transition won’t be clean because no one is rebuilding compensation architecture from scratch while also running a company. The repricing will happen the way most structural shifts do. Unevenly and reactively. Spreading from one “how are you guys handling this?” to another. New measurable bonus targets will come before slashing salaries. Collective pushback will succeed in pockets.
Ultimately capital keeps chasing returns. Returns reward measurement. Knowledge work is getting newly measured.