Will AI take your job? How to find out for your role, in an afternoon

The short answer is that AI is not taking jobs. It is taking tasks, and a job is a bundle of tasks with a title on top. Which is why the question "will AI take my job" has no useful general answer and a very precise personal one. You can have that answer by tonight.

The evidence for the general picture is not in dispute. The World Economic Forum's Future of Jobs Report 2025, built on more than 1,000 employers representing 14 million workers in 55 economies, expects 170 million roles to be created and 92 million displaced by 2030, a net gain of 78 million, with 22% of today's jobs churned along the way and 39% of the skills they need changing. Challenger, Gray & Christmas counted 116,175 announced US job cuts attributed to AI in the first eight months of 2026, about 22% of all cuts, the leading stated reason for the year. Anthropic's Economic Index, measured on conversations up to May 2026, finds the work split almost evenly between people working alongside the model (51%) and people directing it to complete a task outright (49%).

None of those numbers tells you anything about your Tuesday. They describe a market. Your position in it is decided by which of your tasks a competent system can do with your instructions, what the market pays for the rest, and what your employer has decided to measure. All three can be checked in an afternoon.

This guide is the method. Five steps, each with a time budget, and a reading of the result by stage. The cluster of War Room articles under it carries the detail on each step; this page carries the whole.

Why the question is asked the wrong way round

Ask whether a job will be automated and you get a forecast. Ask which tasks in a job a system now performs and you get an inventory. Forecasts are wrong at the edges and useless in the middle; an inventory is checkable today.

PwC's 2026 Global AI Jobs Barometer, built on more than a billion job advertisements across 27 countries, shows what the inventory does to a labour market. AI is sorting roles into two tracks. In roles it professionalises, the routine layer is absorbed and human judgement becomes the product; PwC names radiologists and recruiters. In roles it democratises, the tool makes the work easier for a non-expert to do; PwC names IT service managers and medical secretaries. Professionalised roles show twice the job growth and 42% faster salary growth. Same technology, opposite outcomes, and the outcome is set by the task mix. Which track your own role is on is the first thing this method establishes.

The Anthropic Economic Index makes the same point from the usage side. It matches what people ask a model to do against the task catalogue behind every occupation, and reports usage by task rather than by job. Tasks common to computer and mathematical work take 23.8% of observed usage; arts, design and media 13.6%; teaching and library work 12.8%; sales 9.1%; office and administrative support 7.9%; management 5.9%. Anthropic is careful to say the data describes what the model is used for, not who is using it and not whether any job is safe, and that caveat holds here too. The point is narrower: the unit being absorbed is the task, and tasks are what you can list.

The War Room's standing claim, written before any of these reports, is the same. Jobs are not disappearing all at once. Tasks are absorbed, roles are hollowed out, and one restructuring cycle later the title goes too. You do not own your job; you own the record of what you do in it. So start with that record.

Step 1: the task ledger (one hour)

Open a blank page. Write down everything you did at work last week, one line per task, in the verbs you would use to explain it to a colleague. Not your job description; your week. "Reconciled the supplier ledger." "Drafted the board note on the Q3 numbers." "Talked the client out of the cheaper option." "Checked the model's summary of the contract before it went to legal."

Aim for twenty to forty lines. Under twenty and you are summarising; over forty and you are logging keystrokes. Add a rough share of your time to each line. It does not need to add up; it needs to be honest about where the hours went.

Two rules. First, include the things you did that are not in your title. The unpaid coordination, the review nobody asked for, the phone call that unblocked someone else's week. Those are usually the lines that matter most in step 2. Second, include the things you did with a model. "Asked the assistant for a first draft, then rewrote it" is one task; write it as it happened.

That page is the asset the rest of the method works on. It is also, as the entry-level market analysis points out, the document nobody else in the building has. Roles are rewritten by people who do not know what you do all week, so they use the description on file. Yours is now more accurate than theirs.

Step 2: sort every line into three piles (one hour)

Go down the ledger and put each line in one of three piles.

Instructable. Someone with two years less experience than you could produce an acceptable version of this output using a current AI tool and your instructions. This is the democratisation test from PwC's data, and it is deliberately unflattering. It is not a comment on your talent. It is a statement about where the market will set the price of that task, because a widening pool of people can now produce it.

Judged. The line needs a decision under uncertainty, an outcome someone will be held accountable for, or a relationship that the task depends on. "Chose which of the three forecasts to put in front of the board." "Signed off the release." "Kept the account after the outage." A model can draft the options; it cannot own the call, and it has no relationship to spend.

Hybrid. A system produced most of it and you checked, corrected or shaped the result before it left the building. This pile is the one growing fastest, and the one most people misfile. Checking a model's output is not instructable work. It is judgement applied at speed, and it is paid as judgement only when it is written down as judgement.

Now add up the time shares. The number that matters is the share of your week in the instructable pile. Below a quarter and your week is already mostly the layer above task execution. Between a quarter and a half and you are in the hollowing zone: the routine layer is large enough to be automated away, and the title has not moved yet. Above half and the sorting PwC describes has already started on your role, whether or not anyone has told you.

Three things to check before you trust the number. Move any line where you wrote "with the assistant" from instructable to hybrid if you changed the output materially; the check is part of the task. Look at the judged pile for lines that are really instructable with a nice name; "reviewed" often means "read". And look at the instructable pile for the lines that carry a relationship, because the relationship is not instructable even when the task is.

The three skills AI cannot replace are the sorting key for the judged pile: synthesis under uncertainty, accountability for the outcome, and relationship capital. If a line has one of those in it, it is judged, however routine it looks.

Step 3: check the market (thirty minutes)

The ledger tells you what you do. The market tells you what it pays for, and the answer has moved.

Lightcast, reading more than 1.3 billion job postings, found postings that list AI skills advertise 28% higher salaries, close to $18,000 a year, and that 51% of those postings sat outside IT and computer-science occupations as of 2024. PwC puts the average wage premium for AI skills at 62%, up from 57% a year earlier, and finds jobs requiring specific AI skills growing 69% against 9% for the market as a whole. The same PwC data shows the premium running as high as 118% in consumer markets and as low as 16% in government and the public sector. Identical capability, priced seven times apart, because the price is set by what the capability is attached to.

Two things follow for your ledger. The premium is paid for AI skills applied inside a function, at that function's rate, and half the demand is already outside technology roles. And the premium attaches to the written record: postings, titles, performance summaries. A skill nobody can read about is not priced.

So do this. Pull ten live postings one level above your role, in your sector, and extract the AI language. Not the tools; the verbs. Design a workflow. Verify a model's output. Own an outcome a system produced. Count how many of those verbs you can evidence with a dated artefact you could show a stranger. The gaps are the list for the next ninety days. The re-labelled market piece walks through the same exercise with the Indeed and Handshake numbers behind it.

If you are early in your career the market reads differently, and there is a section for that below.

Step 4: check the employer (thirty minutes)

Your role is set twice: once by the task mix, once by the company. Three things to establish about yours.

Is AI use being scored? Meta told staff in 2025 that AI adoption would count in performance reviews, and by February 2026, as eWeek reported, the policy was fully active and feeding promotions and bonuses. Janelle Gale, Meta's head of People, put it plainly: "As we move toward an AI-native future, we want to recognize people who are helping us get there faster." Meta is the loudest case, not the only one. Find out whether your review template has a line for it yet. If it does not, the unscored cycle is where the standard gets set: the first people to show attributable results define what "good" looks like for everyone after them.

Has the hiring filter moved? Computerworld reported in August 2025 that AI-assisted interview fraud pushed Google, Cisco and McKinsey back to in-person interviews, citing a Gartner survey in which 72.4% of recruiting leaders said they now interview in person to counter it. Greenhouse's 2025 AI in Hiring report found 70% of hiring managers trust AI to make faster and better hiring decisions while only 8% of job seekers call it fair. Written polish is free now, so it carries no signal. The filter is in the room, and what gets checked there is whether you can walk through your own work unassisted. Your ledger is that walkthrough.

What does the law require your employer to tell you? In the EU, the AI Act entered into force on 1 August 2024. Its obligations for high-risk systems in recruitment and employment were scheduled for August 2026; under the Digital Omnibus the deadline moves to no later than December 2027 or August 2028, depending on how the system is classified, as Crowell & Moring's 2026 overview sets out. When they apply, Article 26(7) requires an employer to inform worker representatives and directly affected employees, clearly and comprehensively, before a high-risk system is used on them, and the system must allow effective human oversight. The deadline that moved and the one that did not covers what this means for you in practice. The short version: the right to know how a system scored you is coming, and the person who already knows how the system works is the one that right benefits.

One more number belongs in this step. MIT's Project NANDA reviewed more than 300 public AI initiatives, interviewed 52 organisations and surveyed 153 senior leaders for its July 2025 report, The GenAI Divide, and found 95% of enterprise generative-AI pilots producing no measurable return on $30 to 40 billion of spending. The failures trace to brittle workflows and systems that do not learn from their context. Read that from inside your ledger: your employer's AI programme is, on the odds, not working. The person who makes one workflow in it produce a number is not competing with the system. They are the reason it exists. Why most AI initiatives fail is the operator's reading of that report.

Step 5: score it (ten minutes)

The AI readiness assessment asks seven questions and places you on a 21-point scale in one of four stages: Early, Aware, Operational, Command Level. Take it now, with the ledger in front of you, and answer from the ledger rather than from how you would like to answer. The seven questions are the ones this method has been circling: how often you use the tools in your actual work, whether your organisation has a strategy, how much of your role you judge automatable in two years, whether you can explain how a language model works to a colleague, how you treat a model's recommendation, whether you are building what a model cannot do, and what you do first when your industry moves.

The stage is not a verdict. It is a position, and each position has a different next move.

Early. The instructable pile is large and you have not yet used a model on any of it. The ledger is the win here; most people at this stage have never seen their week written down. The move is to pick the three biggest instructable lines and run each through a current tool this week, with your instructions, and keep the output next to your own. You are establishing the gap, not closing it yet.

Aware. You know the picture and the ledger confirms it, and nothing in your written record shows it. The gap between knowing and doing is where careers stall. The move is documentary: rewrite your own role description, one page, the outcomes you own and the systems you already operate, and send it to your manager as an input to the review cycle. The version on file becomes the version read during restructuring.

Operational. The hybrid pile is your largest and you are the person in the team who checks what the system produced. The move is to convert one instructable output into an owned one this month: stop producing the thing and start owning the standard for it, the quality bar, the review, the sign-off, the consequence. Producing a forecast is instructable. Being accountable for whether the forecast is trusted is not.

Command Level. You are already shaping how your domain uses the technology. The question the assessment puts to you is the right one: are you building systems, or only being excellent individually? The operator stack is the answer for one person: four layers, rules of engagement, a ninety-day audit that removes anything your work no longer depends on.

The first ninety days

The free AI Survival Kit is the plan for the stage you landed in: seven moves and a 90-day schedule, 46 pages, written for exactly this afternoon. It is the download on the home page. Three moves from it stand on their own.

Days 1 to 30: make the record match the week. The ledger becomes the role description. Every hybrid line is written as judgement ("verified the model's contract summary against the source before it went to legal"), not as assistance. Two named AI applications inside your function go on the record with a shipped result and a number attached. Lightcast's finding is that half of AI-skill demand sits outside technology roles; that demand is in your function, waiting for someone to claim it in writing.

Days 31 to 60: move one task up a pile. Take the largest instructable line and own its standard instead of its production. Take the largest hybrid line and make the check visible: a log, a sign-off, a named decision. Building an AI-proof skill portfolio is the longer version of this move.

Days 61 to 90: keep a decision journal and re-run the ledger. One entry per decision, what you expected and what you were unsure of, reviewed at day thirty. Judgement is the only one of the three durable skills you can measure in yourself. Then write the week down again and sort it again. The instructable share should have fallen, and the fall is the number you take into your next review.

If you are early in your career

The entry-level picture is the sharpest version of everything above, and it deserves its own reading.

The Federal Reserve Bank of New York's labour-market tracker for recent college graduates put their unemployment rate at about 5.6% and their underemployment rate at 42% in the second quarter of 2026. Handshake's Class of 2026 report, covered by CNBC in April, found postings on its platform between July 2025 and March 2026 down 2% on the year before and 12% below the same months of 2019 to 2020. Indeed Hiring Lab recorded entry-level postings down 7.5% year on year as of May 2026. The same Handshake data found 4.2% of full-time early-career jobs calling for AI skills as of March 2026, nearly double the year before, and 10.3% of internships mentioning AI. Sixty-two per cent of graduating seniors told Handshake they feel pessimistic about their careers, up from 46% in 2024; 85% said they use AI tools.

Two honest caveats. Economists at the Federal Reserve and elsewhere argue the entry-level weakness is equally consistent with interest rates, a post-pandemic correction in hiring, and general payroll cooling; the dispute is live. And PwC's barometer found AI-exposed entry-level roles seven times more likely to demand judgement and leadership than roles AI does not touch, which means the first rung of the traditional apprenticeship, three years of routine work while judgement is absorbed by proximity, has been removed. Judgement now has to be built deliberately.

The method is the same; the emphasis moves. Your ledger is shorter, so weight step 3 heavily: the ten postings one level up are your syllabus. Your hybrid pile is your evidence, because you were hired into a market where the routine work is already someone's model. Build the artefacts that prove you can check a system's output and own a small decision, date them, and put them where a hiring filter that now sits in a room can ask you to walk through them.

Questions readers ask

Is my job safe if I work in healthcare, the trades or a classroom? Safer than most from task absorption, and not exempt from the sorting. The WEF report expects growth in care roles, educators, delivery and farm work alongside the technology jobs. In Anthropic's usage data the shares for construction (0.1%), healthcare support (0.6%) and installation and repair (0.6%) are small, though Anthropic itself warns that observed usage is not a measure of safety. The administrative and documentation tasks around those jobs are exactly the instructable pile, and they are being absorbed first. Run the ledger anyway; the result is usually a smaller instructable share and a very clear picture of which paperwork to hand over.

Should I learn to code? Only if your function pays for it. Lightcast's data says half the premium is outside technology roles, and PwC's says the premium attaches to the function the skill is applied in. The specific applications your function values, evidenced by a shipped result, are worth more on your record than a general programming credential. If your function is software, the answer is different and you already know it.

Do certificates help? They help a filter find you. They do not change the ledger. The written record that moves your price is a role description that names the systems you operate and the outcomes you own, and a dated artefact behind each claim. A certificate is one line in that record, not a substitute for it.

Are companies really replacing people with AI, or just saying so? Both, and the ratio is visible. Challenger's August 2026 report has restructuring as the leading stated reason for the month, 16,173 cuts or 31%, with AI fourth at 3,462 after leading for five months, while AI's year-to-date count stands at 116,175 of 529,914 announced cuts. The label moved; the ledger did not. A narrow cut lands by task exposure, not by headcount, which is why averages describe a market and exposure describes you.

How often should I redo this? Every quarter, and always in the month before your review cycle is drafted. Most annual reviews are written in the fourth quarter from material submitted earlier, so the ledger and the role description need to be on file by then. The whole method is an afternoon the first time and an hour after that.

What if the answer is that most of my week is instructable? Then you have the information most people will get from a restructuring announcement, a year early, while the choice is still yours. The Kit's first month is built for that answer: record, then move one task up a pile, then measure. AI does the routine work. You do the thinking. The ledger tells you where the line between them currently runs in your job, and the point of the next ninety days is to move it.