Free AI diagnosis: 30 minutes, and you leave with a plan

Method

How we read the job data, task by task

Every role page looks at one job through public data: its tasks, which ones AI could speed up, which ones people already use AI for, and what the job pays.

Last reviewed

What a role page shows

  • The tasks

    What the job is made of, from a government list

  • What people said

    Could AI do the task faster?

  • Already done with AI?

    Real use data from Anthropic

  • Pay and jobs

    Numbers from the U.S. government

Each role page takes one job and looks at its real tasks. For 8 to 15 of them, a table answers two simple questions.

Can AI speed this task up? And do people already use AI for it? Below the table you find the job's pay and number of jobs.

All role pages live in AI and your role. Older guides, like will AI replace jobs, stay where they are.

Where the tasks come from

The tasks come from O*NET, the U.S. government's big list of jobs and what each one involves. It is free, and it is updated four times a year.

It has 1,016 job titles, and 923 of them have full data. It links more than 19,000 task descriptions to the work people do.

Each role page uses the tasks that list gives for that job, in the list's own words.

Can AI speed the task up?

  1. 01

    Yes, AI alone

    The AI tool by itself does it twice as fast

  2. 02

    Yes, with extra software

    It needs other programs built around the AI

  3. 03

    No

    AI does not really speed it up

In 2023, researchers Eloundou, Manning, Mishkin and Rock checked every task in O*NET. They asked one question about each.

Could an AI tool like ChatGPT do this task at least twice as fast, with the same quality? People who studied each task gave the answers.

Each task got one of three answers. "Yes, AI alone" means the AI tool could do it by itself. "Yes, with extra software" means it would need other programs built around it. "No" means AI does not really speed it up.

The study found that an AI tool alone could do about 15% of all U.S. work tasks much faster. With extra software built on top, that rose to between 47% and 56%.

Is it already done with AI?

A task AI could speed up is not the same as a task people already hand to AI. For that, we use the Anthropic Economic Index.

It shows which tasks people actually do with Claude, Anthropic's AI assistant, at work. Anthropic found that 97% of those tasks were ones the 2023 study said AI could help with.

Tasks where AI alone could help make up 68% of the real use Anthropic saw.

When a task does not show up in the data, the page says "Not seen yet". That does not mean AI cannot do it.

Pay and jobs

Pay and job numbers come from the U.S. Bureau of Labor Statistics (BLS). Every year it counts jobs and pay for about 830 kinds of work.

Role pages show typical pay, which is the middle point: half of workers earn more, half earn less. They also show how many jobs there are and what the BLS expects by 2035.

For the bigger picture, see AI job displacement statistics.

How to read a role page

  1. 01

    Read the tasks

    What the job is made of

  2. 02

    See what stays human

    The tasks AI can't really speed up

  3. 03

    Check real use

    Where people already use AI

  4. 04

    Look at pay

    What the work pays today

If AI could speed up a task, that does not mean the job goes away. A job is many tasks, and people still decide, check the work and talk to customers.

The 2023 study did not try to guess when AI will be used or by whom.

The use data comes from people using one AI assistant. So it shows where things are heading, not the whole economy.

What this method does not measure

It does not predict job losses, pay or hiring at any company.

It does not say how good AI is at a task. It only says if AI could speed the task up and if people already use AI for it.

It does not show what a given business does. That depends on its own work, data and rules.

Role pages by sector

Each role page uses this method for one job, with its own task table and pay numbers.

Sources, licenses and updates

This page was last reviewed on November 2, 2026. Each role page shows the date of the data it uses.

  • O*NET database, U.S. Department of Labor, Employment and Training Administration: a free public database
  • Eloundou, Manning, Mishkin and Rock (2023), "GPTs are GPTs": the paper on arXiv; its task answers are published in the authors' repository under the MIT license
  • Anthropic Economic Index, labor market impacts data: CC-BY license for the data, MIT for the code
  • U.S. Bureau of Labor Statistics: OEWS estimates and the Occupational Outlook Handbook, May 2025 data

Next step

Which tasks can AI take off your team?

A free 30-minute call: you leave knowing which processes to automate first.