Fungibility, Not Abundance

A plumber's trip costs $1000 to fix a leaking faucet. The labor is physical. The skill is moderate. By the standard hierarchy — knowledge work above trades, cognitive above manual — this shouldn't happen. But the plumber is non-fungible on three axes simultaneously: few people can substitute, fewer are willing, and the one who does has to be here, in your house, today. A remote copywriter fails on all three. Hence the price difference runs opposite to the hierarchy's prediction.

The knowledge > physical hierarchy was always wrong. Value tracks fungibility rather than labor type.

Three axes of fungibility

What makes labor substitutable isn't the nature of the work. It's the intersection of:

Substitutability — can someone else do this? Physical labor is theoretically abundant. Cognitive labor used to be scarce. LLMs just changed that for generic cognition. "Summarize this document" is now substitutable by anyone with API access.

Willingness — will someone else do this? Trades are facing shortages not because the work is hard but because the supply of willing entrants is dropping. Meanwhile, millions are willing to do remote knowledge work. Willingness constrains supply independent of skill.

Locality — can someone elsewhere, or at another time, do this? The plumber must be local in space. The ER nurse must be local in time. A pandemic reclassifies who's essential overnight — not by changing what the labor is, but by changing where and when it's needed. During COVID, maintenance workers were essential personnel. Faculty members were not. Same skills, different positional value.

Any ranking of labor types that strips out these three axes is a fiction. A convenient fiction for organizing societies — but one that gets corrected when the demand context shifts. Sometimes the correction is a market adjustment, sometimes heads roll.

What LLMs actually reclassify

LLMs don't make cognitive labor worthless. They change its fungibility profile — unevenly, positionally.

Generic cognition on generic knowledge is now fully fungible. "Write a React component," "draft a marketing email," "analyze this spreadsheet" — substitutable by anyone with a model, willing participants everywhere, no locality constraint. This is the copywriter problem.

Cognition on domain-specific knowledge remains non-fungible — but only where the domain specificity is real, not performative. "Which herbicide rotation avoids resistance buildup in Palmer amaranth given this field's three-year history" requires knowledge the model doesn't have and can't get from its training data. The complexity budget explains why: C(M) + C(X) ≥ C(Y). The model supplies C(M). But domain-specific C(X) — the prompt that encodes knowledge absent from the weights — is where the human is non-substitutable. No matter how large M gets, you still need the right-shaped X to elicit the right Y.

The social hierarchy mistook cognitive labor to always be above physical labor. The fungibility lens says: it depends on substitutability in this market, on willingness in this labor pool and on locality in this moment. The plumber was always going to outlast the copywriter. (Let's not get started on robotics just yet)

The positional move

Find your domain specificity. Locate it close to where the demand is — in space, in time, in adjacency to the problems that need solving now. Generic capability is a commodity. Positional, non-fungible knowledge is not. The model gives everyone cognition. What it can't give them is yours.