Of the twenty-two major occupational groups in this release, the one with the highest median exposed share is computer and mathematical occupations, at 57.2%. It is ahead of office and administrative support, which is at 55.4% and is the group most people would have guessed.
| Major group | Occupations | Median exposed share |
|---|---|---|
| Computer and mathematical | 36 | 57.2% |
| Office and administrative support | 51 | 55.4% |
| Sales | 22 | 49.7% |
| Business and financial | 48 | 47.3% |
| Education | 62 | 42.2% |
| Management | 56 | 41.3% |
There is an obvious irony here and it is worth getting past quickly, because the substance underneath is more interesting than the joke.
Why the number is this high
Software work decomposes, in O*NET's task statements, into an unusually pure form of symbolic production. Writing code. Reading specifications. Documenting systems. Testing against requirements. Analysing performance data. Almost none of it requires a body in a place, almost none of it requires a licence, and the work product is text that can be checked by running it.
That last point matters more than it looks. Verification cost is one of the five frictions we score, and software has an unusual property: the output largely verifies itself. Code either compiles and passes its tests or it does not. Compare that to a social worker's assessment or a physician's diagnosis, where an error can sit undetected for a long time and cost a great deal. Cheap verification removes one of the strongest brakes on adoption.
So software scores high on capability and low on four of the five frictions. The arithmetic does the rest.
What the score does not capture
Two caveats, and both are real limits of this index rather than rhetorical hedges.
First, O*NET's task statements describe the job as surveyed, and software work has been changing faster than occupational surveys can track. The tasks in the database are a reasonable description of the work, but the parts of the job that have grown most in the last two years, such as directing and reviewing systems that write code, are under-represented in any occupational taxonomy.
Second, exposure is measured per task, and a job is not only its tasks. Deciding which system to build, arguing for it, and being answerable when it fails are activities that sit awkwardly in any task list, and they are where a great deal of senior software work actually happens.
The useful reading
The finding that survives both caveats is this: being technical is not protective. The protection in this index comes from embodiment, presence, accountability, private context and the cost of being wrong. Software work has unusually little of any of them.
That has a practical implication for anyone in the group. The parts of the role that hold up are the parts that carry accountability for a system rather than authorship of it, and the parts that depend on knowing what this particular organisation needs, which is context, our largest single friction across the whole index.
Figures from release v2026.Q3, built on O*NET 31.0. Group medians are computed across the occupations in each SOC major group. Exposure is not displacement: see the methodology page.