Where the Work Resists
Occupational categories are the wrong unit.
"Knowledge work" and "manual work" both split internally, along the same lines. What determines exposure is not what a worker is called — it's the conditions under which the work happens.
Four resistances determine when AI enters a task. Low resistance is displaced first. High resistance persists — and as adjacent tasks are absorbed, often appreciates.
This predicts order, not safety.
Input Completeness
Is the material already on the table?
Low resistance when inputs exist in retrievable form. High resistance when the input must be drawn from what hasn't been said — unstated intent, tacit convention, knowledge no one wrote down because no one thought to.
Retrieval precedes processing. Only processing has been automated.
Environmental Structure
Is the setting the same each time?
Low resistance in defined environments — fixed workflows, standard forms. High resistance where conditions vary per instance and can't be enumerated in advance.
This is why physical labor didn't uniformly survive. The boundary is environmental, not corporeal.
Answer Determinacy
Does a correct answer exist before the work begins?
Low resistance where correctness is knowable in advance. High resistance where validity is established only by consequence.
Training on prior outcomes doesn't resolve this. A model reproduces the distribution of past judgments. It doesn't become right about cases with no precedent.
Accountability Load
Is bearing the consequence part of what's sold?
Low resistance where output is evaluated on content alone. High resistance where someone absorbs the cost of being wrong.
This resistance isn't technical. It moves at the speed of institutions — the slowest of the four to erode.
Sequence, Not Safety
| Resistances | Trajectory |
|---|---|
| 0 | Displaced first. |
| 1–2 | Survives in reduced headcount. Rates compress. |
| 3–4 | Persists. Scarcity rises as surrounding tasks are absorbed. |
One occupation frequently contains tasks at both extremes. Aggregating them under one job title produces a prediction wrong for both.
What This Doesn't Say
Persistence is not prosperity.
Whether high-resistance work converts into income depends on supply, bargaining position, and the ability to pass cost through. A task can be irreplaceable and precarious at the same time.
The two questions are separate. Conflating them is the second error, after the occupational one.
The question was never which work survives. It was always which friction the work was made of.