Will AI Take My Job? How To Assess Your Risk
Short version: probably not the whole thing. Likely a chunk of it, and which chunk matters more than your job title. Whether AI takes your job depends less on what you're called and more on what you actually do all day. A job is a bundle of tasks. AI doesn't come for bundles. It comes for tasks.
That distinction is the whole game, and most of the "will AI take my job" coverage skips it. A headline that says "customer service reps are doomed" is useless to a specific customer service rep whose day is 60% de-escalating angry callers and 40% logging tickets. AI is already good at one of those and bad at the other. Your risk isn't the average for your occupation. It's the sum of your own tasks.
This page gives you a way to work that sum out. If you'd rather just answer a set of questions and get a score, our free How AI-Proof Is Your Job? assessment does the same math for you in about 2 minutes. Either way, the framework below is what's underneath it.
What the research actually says (and doesn't)
Start with what's real, because the numbers get mangled constantly.
The World Economic Forum's Future of Jobs Report 2025 projects that by 2030, AI and related tech will displace about 92 million jobs globally while creating roughly 170 million new ones (a net gain of 78 million), with total churn touching about 22% of jobs (World Economic Forum). So the aggregate picture isn't a jobs apocalypse. It's a reshuffle, and a big one. The catch buried in that net-positive number: the jobs created and the jobs destroyed don't go to the same people, in the same places, at the same time.
Then there's the study that got the most attention, from Microsoft Research. Researchers analyzed 200,000 real conversations between people and Copilot to measure which occupations' tasks overlap most with what AI can already do, scoring 785 occupations (Microsoft Research). Interpreters and translators topped the list, followed by historians, writers, customer service reps, and sales reps (Fortune). Physical jobs, roofers, equipment operators, sat at the bottom.
Here's the part people drop: the Microsoft team explicitly said this measures exposure, not elimination. In their own follow-up, they warned against reading the list as a death row for those jobs. Exposure means AI touches the tasks. It doesn't mean the job vanishes (Microsoft Research blog). A translator's job includes negotiation, cultural judgment, and accountability that a model doesn't carry.
And then the sharpest data point, from Stanford economist Erik Brynjolfsson and colleagues, using ADP payroll records covering 25 million workers. Early-career workers aged 22 to 25 in the most AI-exposed occupations saw a 13% relative decline in employment after generative AI took off, while older workers in the same roles held steady or grew (Fortune). Entry-level software and customer service roles were hit hardest.
Read those three together and a shape emerges. AI isn't erasing occupations wholesale. It's thinning out the routine, learnable, entry-level slices of exposed jobs first, the exact tasks a junior person used to do to build up skill. That thinning is the real risk, and your title alone won't tell you if you're in it.
The task audit: how to assess your own exposure
Forget your job title for a minute. Grab the last two weeks of your actual work. Here's the process.
Step 1: List your real tasks
Write down what fills your time, in plain language, not the résumé version. Not "stakeholder alignment": "I sit in meetings and take notes, then email a summary." Not "content strategy": "I write first drafts of blog posts and edit other people's." Aim for 8 to 15 items, including the boring middle of your week, not just the parts you'd brag about.
Step 2: Score each task on two axes
For every task, ask two questions.
Can current AI do a first pass? Rate it high, medium, or low. Drafting a routine email, summarizing a document, generating boilerplate code, cleaning a spreadsheet, translating standard text: those are high. AI does them competently today. Reading a room, owning a decision that gets someone fired if it's wrong, coordinating people who don't want to be coordinated: those are low.
Does a human need to be accountable for it? Accountability is the axis most people forget. Some tasks AI can do, but nobody sane lets it do unsupervised: signing off on financial numbers, telling a patient bad news, deciding which employee to let go. High accountability is a moat even when technical capability is high.
Step 3: Sort into four buckets
Plot every task:
| Bucket | AI capability | Human accountability | What it means |
|---|---|---|---|
| Exposed | High | Low | AI can and likely will absorb this. This is your risk. |
| Protected-by-trust | High | High | AI assists, but you stay the one on the hook. Safer than it looks. |
| Human-anchored | Low | High | Judgment, relationships, physical presence. Your durable core. |
| Quiet | Low | Low | Niche or manual work AI ignores for now. Stable but not a growth area. |
Now count. What share of your week sits in the Exposed bucket? That's your rough exposure score. If it's under a quarter, you're in decent shape. Around half, you've got real reason to act (not to panic, to act). Most of your week in Exposed, and the read is that your role as currently defined is on a clock, and the smart move is to shift your time toward the other three buckets before someone else decides for you.
If you want this scored properly, with peer comparison so you can see how people in similar roles land, that's exactly what the job assessment does. It won't flatter you, which is the point.
Why the same job title can carry very different risk
Two paralegals. One spends most of the week on document review and drafting standard contracts, both squarely in the Exposed bucket now that legal AI tools handle first-pass review fast. The other spends the week interviewing witnesses, managing anxious clients, and prepping attorneys for hearings: human-anchored work. Same title on LinkedIn. Wildly different exposure.
That gap is why occupation-level lists mislead. When you read that "writers" rank high on the Microsoft list, that lumps together the person churning out templated product descriptions (very exposed) and the investigative reporter building sources over months (not). The label is the same. The task mix isn't.
Use your own task audit as the real answer. The table below shows how the pattern plays out for eight common roles, so you can see the reasoning applied to a specific title before you apply it to yours.
What this looks like across common roles
Same pattern, different jobs. The automatable slice on the left is real. The slice on the right is what's still standing once it goes.
| Role | What AI already handles | What still needs a human | Exposure |
|---|---|---|---|
| Copywriters and content writers | First drafts, product descriptions, routine social copy | Brand strategy, voice, judgment about what works for a specific audience | High for drafting, lower for strategy and editing |
| Paralegals | Document review, contract summarization, legal research | Client-facing judgment, courtroom logistics, the liability of filing | High for review, low for accountability work |
| Customer service reps | Password resets, order status, routine billing questions | De-escalation, judgment calls on policy, customers who want a person | High for scripted interactions |
| Bookkeepers and accountants | Data entry, reconciliation, routine categorization | Advisory work, tax strategy, audit judgment on incomplete information | High for entry-level tasks, low for advisory |
| Graphic designers | Variations, format resizing, rough concepting | Client relationships, brand strategy, choosing which option fits the brief | High for production, low for direction |
| Admin and executive assistants | Scheduling, routine correspondence, travel logistics | Competing-priority judgment, reading unstated preferences, trust built over months | High for logistics, low for the trust relationship |
| Software developers | Boilerplate code, routine debugging, first-pass implementations | Architecture decisions, translating business needs, novel debugging | High for routine coding, low for architecture |
| Teachers | Grading routine assignments, generating practice materials | Classroom management, reading a room, the relationship of actually teaching | Low overall, with current tools |
| Consultants | Research synthesis, first-draft decks, data analysis, benchmarking | Client trust, accountability for the recommendation, in-person facilitation | High for production work, low for judgment and delivery |
None of this is fixed. Roles built almost entirely from the left column shrink over time. Roles that shift weight toward the right column tend to hold, or turn into something new. Don't see your job above? The task audit two sections up works on any role, since it scores what you do rather than what you're called. Consultants get a full breakdown of their own: see whether AI will replace consultants.
The signals that actually predict risk
Beyond the task audit, a few patterns reliably separate higher-risk situations from lower-risk ones.
Routine and repetition raise risk. If your task follows the same steps most of the time and the inputs are digital, it's a strong automation candidate. Variety and judgment lower it.
Being early-career raises risk right now, uncomfortably. The Stanford data is blunt about this: the jobs that used to be how you learned a profession are the ones thinning fastest, because they're the most routine (TIME). If you're junior, the play is to get to judgment-level work faster than the ladder's bottom rungs disappear.
Physical presence and regulated accountability lower risk. Trades, hands-on healthcare, anything where a licensed human has to be legally responsible: AI can assist but can't be the one who signs. Our guide to jobs AI can't replace goes deeper on where that protection comes from.
Working near the AI, not against it, lowers risk. People who use AI to do more of the high-value work, and become the person who knows how to deploy it well, are consistently better positioned than people whose entire role was the routine slice AI now covers.
What to actually do about your result
The point of assessing risk is to act on it, so here's what to do at each exposure level.
Low exposure. Don't get smug. Low today isn't low forever, and the frontier moves. Learn to use AI tools in your workflow anyway, so you're the one raising your own output rather than the one who gets compared unfavorably to a colleague who did.
Moderate exposure. Moderate exposure is most people, and it's the productive place to be. Deliberately shift hours from your Exposed tasks toward the human-anchored ones. Volunteer for the work that needs judgment and relationships. Get fluent with the AI tools in your field so you're operating them, not competing with them. If reskilling makes sense, target the adjacent skill that moves you up the value chain, not a random bootcamp, and reskilling without quitting your job lays out how.
What this looks like in practice, from a reader: "I used to write first drafts of client reports. That was about 40 percent of my week: gathering the data, structuring it, writing it up in plain language. Now a tool drafts it in about ninety seconds. It isn't perfect. It's close enough that rewriting it from scratch would waste everyone's time.
The first few months, I didn't have a plan. I just did the same job with a new tool bolted on, and it felt precarious, like one product update from being unnecessary. That turned out to be the useful part. It forced me to look at what my job was actually for, not just what my tasks had been.
Clients weren't paying for the drafting. They were paying for someone to know which numbers mattered, catch the ones that looked wrong, and turn all of it into a recommendation they could act on. The drafting was just the visible part of that. Once it got automated, the judgment underneath it didn't go anywhere. It became a bigger share of what I actually did.
I spend less time typing now and more time reviewing, the way an editor reads a draft: assuming it's competent, not assuming it's right. It took about six months to stop feeling precarious and start feeling normal. The instability was about not having reframed the role yet, not the technology itself."
High exposure. Take it seriously and give yourself a real runway: six to eighteen months, not a weekend. Map where the human-anchored work in your field lives and build toward it. If the whole role is exposed, that's a signal to move roles while you still have the standing to choose, rather than waiting to be moved. It's not the message anyone wants. It's the one the data supports.
One thing worth saying plainly: the people who do worst in a transition like this are usually the ones who decided early that it wouldn't touch them and stopped paying attention. You're already past that by reading this.
Using AI on your resume
If you're updating your resume because of what you just read, one question comes up constantly: does using AI on it hurt you? Short answer: not the tool itself. The trouble comes from what the tool produces when nobody checks it.
Most hiring managers aren't offended by an AI-assisted resume or cover letter. Plenty use AI themselves for job descriptions and outreach messages. What draws complaints is what's always drawn complaints: generic, one-size-fits-all applications. AI just makes a mediocre generic application faster to produce at scale, which is where the reputation comes from.
A few things reliably flag an application as low-effort: language that could apply to any job at any company with the title swapped in, claims about the company that are slightly off (a sign nobody checked the output against reality), and formatting quirks tied to specific AI tools that recruiters have simply seen enough times to recognize.
Using AI to tighten grammar, restructure bullet points, or get past writer's block on a first draft rarely bothers anyone. The output still has to be true, specific to you, and specific to the role. Where this catches up with people is the interview: a polished application that doesn't match how someone actually talks creates a credibility gap the moment the conversation starts, and that gap costs more than an imperfect but authentic application would have. Read the final version out loud before you send it. If it doesn't sound like something you'd say, that's the gap showing up early, while you can still fix it.
Keep the tasks that keep you employed
Knowing your exposure level is step one. The free 5-Day AI Reset is a five-email course built around what to do with that number: Day 2 has you take one task back from the tool and do it unassisted, which is the single best test of which skills you still own. Get the free course.
Frequently asked questions
Will AI take my job in the next five years?
For most people, AI will take some of your tasks, not your whole job, within five years. The WEF projects large churn (about 22% of jobs disrupted by 2030) but a net increase in total jobs (World Economic Forum). Your personal risk depends on how much of your week sits in tasks AI can do without human accountability.
Which jobs are safest from AI?
Roles built on physical presence, licensed accountability, complex human judgment, and relationship-building are safest. Microsoft's research put physical and manual jobs lowest on the exposure list (GeekWire). But "safest" is about task mix, not title: a hands-on role can still contain exposed tasks.
Is it true AI is only hitting entry-level jobs?
Not only, but disproportionately. Stanford's payroll analysis found a 13% relative employment decline for 22-to-25-year-olds in the most exposed occupations, while older workers in the same jobs were stable or growing (Fortune). The routine tasks juniors do are the most automatable.
Does using AI at work make my job safer or riskier?
Usually safer, if you become the person who deploys it well rather than the person it replaces. The dividing line is automation versus augmentation: roles where AI does the whole task are shrinking, roles where AI makes a skilled human faster are often growing.
How accurate are online "will AI take my job" tools?
It varies a lot. Older tools rely on a 2013 automation model and score by title, which misses your actual task mix. A good assessment scores your specific tasks and accountability, not just your occupation. Ours does the latter, and we compare the options in our guide to the will AI take my job calculator.
Do companies use AI detectors to screen resumes?
Some do, but detector accuracy is unreliable enough that interview alignment matters more than any tool's verdict. A resume that doesn't match how you actually talk in person is the bigger risk either way.
Should I disclose that I used AI to write my resume?
Not for routine drafting help, similar to not disclosing that you used spellcheck. It matters more if you're asked directly in an interview, where a straight answer tends to land better than dodging the question.
Ready to stop guessing? Take the free How AI-Proof Is Your Job? assessment: it runs the task-based analysis on your real work and shows you where you stand against people in similar roles.