Wondy.
KO|EN
·AI & IT News·9 min read·WONDY

Will AI Take My Job? A 15-Minute Self-Audit Using AI Itself

New AI models shipped in April, May, and June 2026, and the job-loss headlines followed each one. Instead of doomscrolling, ask the AI about your own job. Here is a copy-paste prompt that scores your daily tasks for AI exposure, what the actual US labor data says, and how WONDY read the results without panicking.

Last week a coworker spun his laptop around at lunch and showed me a client report he'd finished in four minutes with AI. It used to take him a day. He was grinning. The rest of the table went quiet, because everyone was quietly doing the same math on their own job. That scene is playing out everywhere right now: three frontier models shipped in three months (OpenAI's GPT-5.5 on April 23, Anthropic's Claude Opus 4.8 on May 28, GPT-5.6 announced June 26), and every launch makes the lunch-table silence a little heavier.

Here is the thing nobody says out loud: the fastest way to find out what AI can do to your job is to ask the AI, about your job, specifically. Not "will AI replace workers." Your tasks, your week, scored one by one. I'm WONDY, I write about AI and money for a living (a job with plenty of exposure, as you'll see), and I ran this audit on myself before writing this. Let's set the facts first, then do yours.

What the US data actually shows

Start with the scary numbers, then the context that defuses them. The World Economic Forum projects AI will displace 92 million jobs by 2030. The same report projects 170 million new ones, for a net gain of about 78 million. Goldman Sachs' famous figure, 300 million jobs affected, counts tasks that could be automated, not people who will be fired. A job is a bundle of tasks, and AI eats some of the bundle, rarely the whole thing.

What makes this wave different is who is exposed. It is white-collar work this time. Management analysts, lawyers, and software engineers show the highest theoretical exposure, and the WEF's high-displacement list names paralegals, junior consultants, and research analysts. Meanwhile only about 3% of workers without a high school diploma face high exposure. The old automation story, machines coming for factory floors, has flipped.

But exposure is not unemployment. Oxford Economics concluded in January 2026 that firms "do not appear to be replacing workers with AI on a significant scale." The Yale Budget Lab reported in February 2026 that unemployment duration for AI-exposed jobs is unchanged and there is no macro-level evidence of AI disruption in the aggregate employment data. Anthropic's own labor research found no systematic rise in unemployment for highly exposed workers since late 2022, though it did find suggestive evidence that hiring of younger workers has slowed in exposed occupations. So the honest picture: no mass firings yet, a real squeeze at the entry level, and nobody sure how long "yet" lasts.

15-minute AI job self-audit infographic — list your week as 5-10 tasks, score each 0-5 for AI exposure, hand off the 4s and 5s and own the 1s and 2s; WEF projects 92M jobs displaced and 170M created by 2030 for a net gain of 78 million
The 15-Minute AI Job Self-Audit

The 15-minute self-audit

Averages are useless for a personal decision. What you want is a task-level map of your own job. Three steps.

Audit your own job with AI
  1. 01
    List your tasks

    Break your week into 5 to 10 concrete tasks.

  2. 02
    Score each task 0 to 5

    Ask AI how much of each task it can do today.

  3. 03
    Read it as a map

    Hand off the 4s and 5s, own the 1s and 2s.

Here is the prompt. Paste it into ChatGPT or Claude and swap the bracketed parts for your own job. I used a paralegal, one of the named high-exposure roles, as the example.

You are a labor-market analyst. Audit my job honestly.

- My job: [paralegal at a mid-size law firm]
- What I do in a typical week: [legal research, drafting standard motions,
  summarizing depositions, managing filing deadlines, client intake calls,
  coordinating with attorneys]

Do the following in a table:
1. Break my week into 5-10 discrete tasks.
2. Score each task 0-5 for how much of it AI can do TODAY
   (0 = humans only, 5 = AI does nearly all of it).
3. Mark high-scoring tasks as "hand off to AI, reclaim the time."
4. Mark low-scoring tasks as "double down, this is your moat."
5. End with one verdict: "this role faces restructuring" or
   "this role faces replacement risk," with your reasoning.

No fear-mongering. Be specific and cite what today's tools can and cannot do.

The two rules that make this work: force task-level scoring, and demand a verdict with reasoning. Ask "will AI replace paralegals" instead and you'll get a hedged essay that scares you without telling you anything.

AI 답변 (요약)ChatGPT

Here is the task-level audit for your paralegal role.

TaskAI exposure (0-5)What to do
Legal research4Hand off first drafts, verify citations yourself
Drafting standard motions4AI drafts, you review
Summarizing depositions5Hand off, spot-check
Deadline management3Automate reminders, keep ownership
Client intake calls1Double down, this is your moat
Attorney coordination1Double down

Verdict: this role faces restructuring, not replacement. The research and drafting hours shrink, but the judgment, client-facing, and coordination work remains human, and firms still need someone accountable for accuracy.

That answer matches what the macro data suggests: the role gets rebuilt around the low-scoring tasks rather than deleted. When I ran my own job (research, drafting, data cleanup all scored 4 to 5; deciding what to write about and verifying facts scored 1 to 2), the verdict came back the same way. Seeing the 5s stung for a minute. Then it read less like a threat and more like a to-do list: hand off the drafting hours, spend them on the judgment work.

Where this audit goes wrong

Three failure modes, so you can use the tool without being used by it.

First, AI overrates itself. Chatbots claim capabilities with total confidence, so scores skew high. When a task gets a 5, push back: "can you really do this end to end today, or just a first draft?" The score often drops to a 3 under cross-examination.

Second, exposure is not a pink slip. The Yale and Oxford findings above are the reason to read a high score as "here is where I reclaim hours," not "here is my termination date." The people getting hurt in the current data are mostly new entrants in exposed fields, because companies hire fewer juniors when AI does junior work. If that's you, the audit tells you which skills to stack so you're not competing with the machine at its strongest.

Third, the map expires. Three model releases in three months means your scores from January are stale by July. Re-run the same prompt every six months and watch the direction, not any single snapshot.

FAQ

Is AI actually replacing jobs in the US right now?

Not at scale, per the data so far. Oxford Economics (January 2026) found firms are not replacing workers with AI on a significant scale, and the Yale Budget Lab (February 2026) found no macro-level evidence of AI disruption in aggregate employment statistics. The real, narrower problem is slower hiring for entry-level roles in exposed occupations.

Which jobs have the highest AI exposure?

White-collar knowledge work, reversing the old automation pattern. Management analysts, lawyers, and software engineers show the highest theoretical exposure; the WEF names paralegals, junior consultants, and research analysts on its high-displacement list. Only about 3% of workers without a high school diploma face high exposure.

How do I check my own job's AI exposure?

Break your job into 5 to 10 concrete tasks and have a chatbot score each from 0 to 5 for how much AI can do today. Task-level scoring shows which parts of your week to hand off and which to double down on. Vague questions get vague, scary answers.

If my exposure score is high, should I switch careers?

Not on the score alone. Exposure measures what AI can assist with, not whether your employer will cut the role, and augmentation is the more common pattern in the data. Shift your time toward low-scoring tasks (judgment, relationships, direction) and re-audit every six months.

Closing thought

Nobody, including the labs shipping these models, knows how this plays out. This post isn't here to calm you down or scare you; it's here to replace a vague dread with a specific map. Fifteen minutes, one prompt, your own numbers. Anxiety mostly lives in the not-knowing.

Sources