Is My Job Safe From AI? How to Audit Your Role Task by Task

Is any job really safe from AI?

The honest answer is that "safe" is the wrong word. Lists of safe and doomed job titles are popular because they are simple, but they do not match how the research is done or how work changes in practice. A job title is a label on a bundle of tasks. A paralegal reviews contracts, yes, but also reassures clients, keeps the precedent library in order and remembers what was agreed with the other side three years ago. Those tasks sit at very different distances from what an AI system can do.

The International Labour Organization made this point clearly in its May 2025 working paper, produced with Poland's research institute NASK. It scored individual tasks before drawing any conclusion about occupations, and found that about one in four workers worldwide are in an occupation with some degree of generative AI exposure, with 3.3 per cent of global employment in the highest exposure category. The authors' own conclusion is that, because most occupations include tasks that need human input, transformation of jobs is the most likely effect.

So the useful question is not whether your job is safe. It is which of your tasks are changing, and what you intend to do about each one.

What does "AI exposure" actually mean?

Exposure is a measure of technical potential: how far an AI system could perform, or help perform, the tasks in an occupation. It says nothing certain about what employers will do, how quickly, or what happens to the people involved. A highly exposed task might be automated, or it might be done faster by the same person, who then has time for something more valuable.

The numbers that circulate need the same care. The World Economic Forum's Future of Jobs Report 2025 reports that employers expect about 170 million roles to be created and 92 million displaced worldwide by 2030, with nearly 40 per cent of on-the-job skills changing. Those figures come from a survey of more than 1,000 large employers about their expectations. They are projections, and they describe a world in which far more jobs change shape than vanish.

PwC's 2025 Global AI Jobs Barometer, based on close to a billion job advertisements, found that the skills employers ask for are changing 66 per cent faster in the most AI-exposed occupations than in the least exposed. It also reported that job numbers were still growing in exposed occupations. Exposure, in other words, is a signal of change, and it is not the same as job loss.

How do I audit my own job task by task?

You need about two hours, your calendar and a spreadsheet. The method below is the one built into the Career Exposure Audit workbook in our AI-Proof Your Career kit, but a blank sheet will do to begin.

• Step 1. Pull up your last two weeks: calendar, sent emails, chat and any ticket or project system.
• Step 2. Write 15 to 30 tasks as verb plus object, such as "prepare the monthly variance report". Avoid vague labels like "admin".
• Step 3. Estimate hours per week for each task. The total should be close to your real working week.
• Step 4. Add the invisible tasks: the questions colleagues bring to you, the clients who call you directly.
• Step 5. Rate each task from 1 to 5 for exposure, value and human edge (explained in the next section).
• Step 6. Choose an action for each task: automate, augment, protect or drop.
• Step 7. Calculate a weighted exposure score: multiply each task's hours by (exposure minus 1) divided by 4, add them up and divide by total hours.

How should I rate each task?

Use three ratings. Exposure asks how much of the task current AI tools could do with the access your organisation would realistically give them. Value asks how much your employer or client cares about the outcome. Human edge asks how far the task depends on judgement, relationships, accountability or local context.

Rate what tools can do today, not what a conference speaker says they will do in five years. And be honest in both directions: people tend to under-rate exposure on tasks they enjoy and over-rate it on tasks they fear. An AI assistant can give a useful second opinion on your ratings if you remove confidential detail first.

What should I do with each task once it is rated?

The ratings point to one of four actions. The weighted exposure score that results is a thermometer, not a verdict. A score of 60 per cent does not mean a 60 per cent chance of redundancy. It means about 60 per cent of your current hours go on work where AI can help or take over, which tells you how much of your week is likely to change and how much time you could free.

The most common mistake is to treat highly exposed, high-value tasks as a threat to hide. They are where learning to work with AI pays back fastest, because the organisation already cares about the result.

Which skills are hardest for AI to replace?

Four qualities keep a task with a person. Judgement: deciding what matters when rules run out or information is incomplete. Relationships: trust built over time between specific people. Accountability: a named human who must answer for the outcome to a customer, a regulator or a board. Context: the unwritten knowledge of this organisation, this market and this week.

These are not soft extras. The World Economic Forum's employer survey lists analytical thinking, resilience, leadership and collaboration among the skills growing in importance, next to AI and data skills. The practical move is to look at your exposed tasks and ask how you could add one of the four qualities. Drafting a report is exposed. Presenting it, answering questions and making a recommendation you will stand behind is much less so.

If you want an independent baseline before you start, the Everyday AI Judgment at Work assessment on AssessAll measures how well you decide when and how to rely on AI output.

How do I build AI skills without taking another course?

Courses help, but fluency comes from repetition on real work where you know enough to spot a bad answer. Take the task you marked Augment with the most hours, and work on that one task for two weeks: brief the tool as you would a bright new colleague, criticise the output, check every number and name against the source, and save what worked.

There is room to stand out here. Stanford University's AI Index Report 2025 notes that 78 per cent of organisations surveyed reported using AI in 2024, up from 55 per cent in 2023. Yet in a 2024 Upwork Research Institute survey of 2,500 people, 77 per cent of employees using AI said it had added to their workload, and 47 per cent did not know how to achieve the productivity gains their employers expected. Someone who can show one specific task done better, with the check they applied, is still unusual in most teams.

Keep a simple log: the task, minutes before, minutes after, how often it recurs, whether quality rose or fell, and the human check you applied. Record the failures too. Before you put any work material into a tool, read your employer's AI and data policies and use only approved tools.

When should I talk to my manager, move internally or pivot?

Talk to your manager once you have two or three logged improvements. Bring a proposal, not a question about safety: what you tested, what it freed, where you suggest the time should go, and a request for a short pilot. Managers are usually under pressure to do something with AI and rarely have task-level evidence. If you want to formalise the skills you plan to build, an individual development plan on Jobulary gives the conversation a structure and review dates.

If your exposure score is high, your list of protected tasks is short and proposals to reshape the role go nowhere, look at moves in order of cost: reshape the current role, move internally, pivot to an adjacent role, and only then consider a full change of field. An internal move keeps your context and relationships, which are hard for anyone else to match.

Before any move that risks your income, work out your runway: accessible savings, less one-off transition costs, divided by the gap between essential monthly costs and any income that would continue. This is a planning figure, not financial advice, so speak to a qualified adviser and check your contract and local employment law before you resign.

How often should I repeat the audit?

Every six months, and whenever your organisation introduces a significant new tool. AI capability moves quickly, and a task you rated 2 for exposure this year may be a 4 next year. LinkedIn's Work Change Report estimates that by 2030 around 70 per cent of the skills used in most jobs will have changed, which is an estimate, but a reasonable prompt to keep your own picture current.

A re-audit takes less than an hour once the task list exists. Compare your weighted exposure score, your hours in protected work and your log of improvements with last time. The trend tells you more than any single number.