The comfortable assumption about automation is that it works from the bottom up. Machines take the routine jobs first, the ones needing least training, and education is the ladder out. That story described the last three waves of automation reasonably well. It does not describe this one.
O*NET assigns every occupation a Job Zone, a five-point scale for how much preparation the work requires. Zone 2 is a job you can learn in a few months. Zone 5 needs a doctorate or a professional licence. Sorting our 923 scored occupations by that scale gives this:
| Job Zone | Preparation required | Occupations | Median exposed share |
|---|---|---|---|
| 2 | Some, a few months | 337 | 16.4% |
| 3 | Moderate, vocational or associate | 208 | 23.9% |
| 4 | Considerable, bachelor's degree | 226 | 41.1% |
| 5 | Extensive, doctorate or licence | 152 | 35.6% |
Two things in that table are worth sitting with.
Exposure more than doubles between Zone 2 and Zone 4
A job requiring a bachelor's degree has, at the median, two and a half times the exposed task share of a job you can learn in a few months. That is the opposite shape to the one automation has had for two centuries.
The reason is not mysterious once you look at what the work consists of. Zone 2 jobs are disproportionately made of handling objects, moving through physical space, and doing things to material. Our activity family for handling and moving objects has a mean exposed score of 2.4%, and general physical activities sits at 0.6%. Current systems cannot do any of it, not because the work is intellectually hard but because it happens in the world rather than on a screen.
Zone 4 jobs are disproportionately made of reading, analysing, drafting, recording and deciding. Documenting and recording information scores 60.4%. Analysing data scores 57.7%. Working with computers scores 58.6%. That is what a bachelor's degree often buys: the right to do symbolic work. Symbolic work is exactly what these systems produce.
The fall from Zone 4 to Zone 5 is the more interesting number
If exposure simply tracked how symbolic the work is, Zone 5 should be the most exposed tier of all. Doctors, lawyers, professors and senior engineers work with symbols almost exclusively. Instead exposure falls by five and a half points.
The friction scores explain it. Zone 5 work carries the accountability that Zone 4 work usually does not. Somebody has to hold a licence and sign the output. A radiologist's read is not a document, it is a legally attributable clinical act. A structural engineer's drawing carries a stamp. That requirement does not care whether a machine could have produced the same artefact.
This is the clearest evidence in the index for something we have argued from the start: capability and adoption are different questions, and the gap between them is where most people's working lives actually sit.
What this does not mean
It does not mean a degree was a bad investment, and it does not mean Zone 2 work is safe in any general sense. Zone 2 occupations are exposed to different pressures entirely, including the cheaper and older kind of automation this index does not measure. A warehouse does not need a language model to shed jobs.
It also does not mean Zone 4 workers are about to be replaced. Exposure is a measure of what current systems can produce, not a forecast of employment. Whether exposed work becomes a lost job depends on firm economics, regulation and organisational inertia, none of which this index models, and all of which move slowly.
What the table does mean is narrower and more useful. If you are trying to work out which parts of your own job are in play, the honest signal is not how long you studied. It is whether your output is a document, a calculation or a message, and whether anyone is legally required to sign it.
Figures from release v2026.Q3, built on ONET 31.0. Job Zone is sourced from ONET without modification. Exposed shares are computed from the published rubric. The full method, including the inter-rater reliability of every judgment involved, is on the methodology page.