· AI

Office Admin Work Is 90 Percent AI-Capable. Only 40 Percent of It Is Actually Automated.

The jobs with the highest theoretical AI exposure are not where the real disruption shows up yet, and the redeployment problem turns out to be smaller and more specific than a department-wide cut assumes.

Computer programmers now carry the highest measured AI exposure of any occupation: about 75 percent of their tasks are things AI is actually doing, not just things a model could theoretically handle. Customer service representatives sit at 70 percent. Data entry and medical records specialists are at 67 percent. Those figures come from a measure Anthropic's economists built to separate theory from practice, and the gap between the two changes the question a business owner should ask before cutting a team.

Most restructurings still get sized by department: trim the back office, hold engineering, cut support hardest. The exposure data says that unit is wrong. Exposure concentrates in specific tasks scattered across almost every role, and the number of people with nowhere else to go once those tasks are automated is far smaller than a department-wide cut assumes.

The honest case for cutting first

There is a real, large-scale case for acting now rather than waiting. The World Economic Forum's Future of Jobs Report 2025, based on surveys of more than 1,000 employers across 55 economies, projects 170 million new jobs created and 92 million displaced by 2030, a net gain of 78 million, but with real churn in between. Forty one percent of employers surveyed say they plan to shrink their workforce specifically because AI is automating tasks those workers did. That is not a hypothetical; it is what employers running real companies say they intend to do, at a scale too large to dismiss as caution.

But the same survey complicates its own number. Almost half of those employers also expect to redeploy staff out of AI-exposed roles into other parts of the business rather than let them go. Employers are roughly split between cutting and moving people, and the exposure data below says that split should not be a coin flip.

What is actually exposed, and what is not yet

Anthropic researchers Maxim Massenkoff and Peter McCrory built their "observed exposure" measure to track actual usage against the theoretical ceiling, and the gap between the two is the whole story. Office administration, as a broad category, sits at roughly 40 percent observed exposure against a 90 percent theoretical ceiling, according to reporting on the study by Fortune. The computers and math category runs 33 percent observed against 94 percent theoretical. Construction and agriculture show almost no observed exposure at all, and fields like healthcare and the life sciences carry high theoretical exposure but little real automation yet, likely because legal constraints, software integration, and the need for human verification slow adoption even where a model is technically capable.

The researchers also found, as reported by The Register, no systematic rise in unemployment among highly exposed workers since late 2022; the gap versus less-exposed workers is statistically indistinguishable from zero. The one real signal is narrower: a roughly 14 percent decline in the job-finding rate for younger workers in exposed fields, a result the researchers call barely significant. So far, the effect shows up as slower hiring at the entry level, not mass layoffs of people already in the role.

The redeployment math that actually matters

A January 2026 NBER paper by Sam Manning and Tomás Aguirre puts a number on who can actually move. Of the 37.1 million workers in the top quartile of measured AI exposure, 26.5 million are in occupations that also carry above-median capacity to manage a job transition, meaning the education or transferable skills to shift into adjacent work. The harder group is smaller and specific: 6.1 million workers, about 4.2 percent of the workforce the study covers, combine high exposure with low adaptive capacity, and they cluster in clerical and administrative roles, the rules-based scheduling, paperwork, and data entry work that shows up in the office-admin exposure numbers above.

What this means for your next restructuring

Before cutting a department, break every role in it into its actual tasks and check which ones land in the 60 to 75 percent exposure range and which sit near zero. Most people doing the exposed tasks, according to the NBER data, have real capacity to move into the judgment work a model cannot do: handling exceptions, reviewing the agent's output, serving the customer who does not fit the template. Save the hard conversation, the real retraining time and an actual new role, for the narrower clerical and administrative slice where that capacity is thin. A department-wide layoff cuts both groups at once and throws away the first group's ability to run the system you just built.

It is also the cheaper call, not just the fairer one. Letting go of someone with transferable capacity means hiring and training a replacement later, right when the automated system needs the oversight that person already had. That is the task-by-task design work behind an AI agent fleet: deciding what moves to automation and who specifically needs a real plan to move with it, before the headcount decision gets made instead of after. If you are looking at a restructuring and want that map before you cut anything, that conversation is worth having.

Sources

References used in this article. Links also appear alongside the relevant claims.

Let's talk

Let's make it happen.

You don't need a polished brief. A couple of lines about where you want your company to go is plenty, and we'll come back with what we'd tackle first.

We scope the work around your goals and timeline before quoting anything, so you know exactly what you're getting.

LocationBoca Raton, Florida
CoverageSouth Florida + remote nationwide
Status Now accepting clients