Is AI coming for my job? A 30-minute task-by-task check and a skills plan

Short answer

Nobody can tell you for sure whether AI will replace your job, and lists of "jobs AI will kill" are guesses. What you can do is look at your job as a set of tasks and mark which ones AI already does, which it helps with and which it can't do yet. The best current evidence, a Stanford study of payroll data, finds fewer young people being hired into the most AI-exposed jobs, but the authors call it an early signal, not proof of cause. Use the check below to see where your work sits, then pick two skills to practise for 30 days.

Who this is for. Workers, students and career changers who are worried, or just curious, about AI and their work. You need 30 minutes and a pen or a spreadsheet.

What the studies measured, and what they didn't

The study most often quoted is "Canaries in the Coal Mine?" by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at Stanford. Its August 2026 version uses ADP payroll records (pay data from a large payroll company) up to June 2026. It finds that employment of 22- to 25-year-olds in the jobs most exposed to AI "now stands 19% below where it would be had it kept pace with that of their less-exposed peers", and that experienced workers show no comparable gap (Stanford Digital Economy Lab, August 2026, working paper, not peer-reviewed).

What the authors say about their own result:

It is descriptive, not causal. They call the findings "early, descriptive indicators" rather than causal estimates. Something else could explain part of it.
It is mostly fewer hires, not layoffs. The gap comes "primarily through reduced hiring of young workers rather than increased separations".
It depends on how AI is used. Declines are in jobs where AI mostly does the task instead of a person. Where AI mostly helps people do the task, employment is "flat or rising".
There is no economy-wide wave. They find "no evidence of widespread, economy-wide job displacement from AI."
The size is uncertain. Adding a control for how many workers have a degree roughly halves the estimate in the most exposed group. The gap is bigger in ADP's data than in national surveys, and the groups were already diverging around the COVID-19 pandemic.

"Exposure" in studies like this means a model or a usage index rated the job's tasks as ones AI could do or is used for. It is a rating of tasks, not a count of jobs lost. That is why the check below works at the task level too.

The 30-minute check

1. List your real tasks (10 minutes)

Write down 10 to 15 things you actually do in a normal month, each as a verb and an object, such as "answer customer emails" or "plan the weekly rota" (illustrative). If you get stuck, look up your occupation on O*NET OnLine (a US Department of Labor database of occupations and their tasks) and edit its list to match your real job.

If you have already filled in Where could AI help this week?, start from that task list. It covers your repeated tasks. Add the ones that aren't weekly, such as training a new colleague or handling a complaint that went wrong.

2. Mark each task (10 minutes)

Mark	Meaning	How to decide
Does	AI can already do this task end to end, with a quick check	You or a colleague have seen an assistant do it to a standard you'd accept
Helps	AI speeds up part of it, but a person decides, checks or finishes	It drafts, sorts or summarises, and you still own the result
Can't yet	It needs your presence, judgement, trust, hands or knowledge of people	You can't picture handing it over this year

The marks are our suggestion. For a second view, browse your occupation in the Anthropic Economic Index (usage data from Anthropic, which makes the Claude assistant, so vendor data). It shows which tasks people use Claude for, and whether they mostly hand the task over or work alongside it. It covers one company's users only.

The best test is your own. Try one "Helps" task with an assistant and time it with and without, using Am I actually faster?. Don't paste anything confidential: What not to paste.

3. Read your result (5 minutes)

Count your marks. Our rough guide, not tested by anyone:

Mostly "Can't yet". AI will change some of your admin more than your job. Learn to use it for that admin.
Mostly "Helps". In jobs where AI mostly helps, the Stanford study saw flat or rising employment. Learn to brief the assistant and check its work.
Several "Does", especially your core tasks. Your job is likely to change, and entry-level hiring in similar roles may slow. Shift your time towards the "Can't yet" tasks and the checking.

Weigh tasks by how much of your time they take. Five small "Does" tasks matter less than one big one.

4. Pick two skills and a 30-day plan (5 minutes)

An AI-adjacent skill, such as writing a clear brief for an assistant, checking its output against sources, or working with data in a spreadsheet. Learning with AI without losing the learning explains how to practise without letting the tool do the thinking.
A human-heavy skill in your field, such as handling a difficult client, negotiating, teaching a colleague, or hands-on work AI can't reach.

My 30-day skills plan · v1.0

Date: <TODAY>
Tasks marked "Does": <DOES_TASKS>
Tasks marked "Helps": <HELPS_TASKS>
Skill 1 (works with AI): <SKILL_1>
  Practice: <WHAT_I_WILL_DO>, <HOW_OFTEN>
  Done when: <A_RESULT_I_CAN_SHOW>
Skill 2 (human-heavy): <SKILL_2>
  Practice: <WHAT_I_WILL_DO>, <HOW_OFTEN>
  Done when: <A_RESULT_I_CAN_SHOW>
Next check: <DATE_IN_SIX_MONTHS>

Replace the following:

<DOES_TASKS> and <HELPS_TASKS>: from your table.
<SKILL_1> and <SKILL_2>: such as checking an AI summary against the source and running a team meeting (illustrative).
<WHAT_I_WILL_DO> and <HOW_OFTEN>: something small, such as check one AI summary against its source and twice a week.
<A_RESULT_I_CAN_SHOW>: a thing you could mention in an interview, such as a weekly report I now produce in half the time, with no errors found by my manager.

If you are job hunting, the result line is the kind of concrete example Job search and interview prep asks for.

5. Redo it every six months

Tools change faster than jobs. Put the next check in your calendar and keep your old table, so you can see what moved.

Limits

Sources

Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, August 2026. Working paper using ADP payroll data. Link
US Department of Labor, Employment and Training Administration. O*NET OnLine, read October 2026. Government occupation database. Link
Anthropic. Anthropic Economic Index, last updated June 2026. Vendor usage data. Link

Changes

v1.019 Oct 2026first version