How to read an AI exposure score, one task at a time

If your company has an AI exposure score for every role, try a small test. Pick any role and name one responsibility behind its number. Most people cannot, and that silence is where a lot of workforce AI planning quietly stalls. A score of 67 tells you something is happening to a role, but it does not say which parts of the job are moving or in what order.

People ask me all the time whether AI will take their job. My answer is that the useful unit of measurement is the task, so tasks are what we measure at JobRoute.ai. A role score only becomes useful for planning when you can open it up and read the tasks inside it.

How the score is built

We start with the role’s actual responsibilities, the task statements in the Department of Labor’s ONET database, version 30.3. We weight each task against real AI usage observed in the Anthropic Economic Index, and the occupation’s exposure reflects the share of its daily tasks that current AI tools can already perform. Employment counts come from BLS state employment data for May 2025. We have scored all 1,016 ONET occupations this way, and the methodology is published.

The national AI readiness score is 63.7 out of 100, across 144,555,948 workers in the BLS employment data. It is a useful headline, but on its own it tells a planner about as much as the 67 above. The value is in the breakdown underneath.

A paralegal, task by task

Take a paralegal. O*NET breaks the role into responsibilities such as legal research, drafting affidavits, reviewing contracts, preparing for trial, meeting clients and filing pleadings. Legal research scores 94 out of 100. Every other task in the role, including drafting affidavits, trial preparation, client meetings and filing with the court, scores 29, which puts them in the durable band.

That gives the paralegal role one task at 94 and seven at 29. The job will rearrange itself around those seven, and the paralegal who owns them and hands research to AI becomes more valuable to the firm. The person whose whole day was legal research has a real problem. I think that person deserves to know now, while there is still time to move toward the other seven.

Exposure follows the tasks

The same logic explains the top of the national list. Ranked by the share of daily tasks current AI tools can already perform, the most exposed occupations include these:

• Customer Service Representatives, exposure 84, with 2,595,760 workers
• Computer Programmers, 84, with 91,270 workers
• Market Research Analysts and Marketing Specialists, 83, with 449,790 workers
• Business Intelligence Analysts, 81, with 87,468 workers
• Medical Records Specialists, 80, with 194,040 workers
• Wholesale and Manufacturing Sales Representatives, 79, with 1,238,180 workers

Customer Service Representatives are the single most exposed large occupation, at 2.6 million workers. Programmers and business intelligence analysts sit right next to them, alongside medical records specialists and sales representatives. What these roles have in common is the shape of their daily tasks, since a large share of those tasks are things current AI tools already do, whatever the seniority of the title.

Automation and augmentation call for different decisions

A task score still leaves one question open, which is what kind of AI use is happening on that task. Each responsibility in JobRoute Companies carries a plain band (AI is here, Shifting or Durable), and where the Anthropic Economic Index has observed real usage on a task, the drilldown also shows whether that usage looks like automation or augmentation. When I wrote about this in July, observed usage was available for 2,450 tasks.

Automation means AI doing the work. Augmentation means AI helping a person do the work. A task that is being automated points toward redesigning the role around what remains, while a task that is being augmented points toward training people to work alongside the tool. Those are different futures for the same score, and they need different decisions from HR and from line managers.

My caveat is the most important part of reading any of these numbers. A high exposure score means a large share of a role’s tasks are changing, and I read it as a signal to reposition people early. Treating it as a layoff list would misread what it measures, partly because the score describes tasks and knows nothing about how a particular team does its work, and partly because observed usage covers only part of the task catalogue so far.

If you want to start with your own role, the individual AI Ready Score at ready.jobroute.ai is free. Look at the tasks behind your number before you look at anyone else’s.