Will AI take my job? Evidence on tasks, hiring and skills

What employer and vacancy evidence can tell us

The OECD's June 2026 synthesis reports that fewer than 1% of workers are likely to need advanced AI skills such as model development. For many more, the relevant work is using digital tools, checking data and interpreting results alongside problem-solving and communication. In the employer surveys it reviews, around 40% of non-adopters in manufacturing and finance identified skills as a barrier. More than half of workers using AI reported employer-funded training, and trained users were more likely to report better performance and working conditions. These are survey associations, not proof that a short course will protect any particular position. The OECD warns that its most recent underlying data date from late 2024.

An IMF staff discussion note published in January 2026 examines demand for new skills across labour markets. It reports that around one in ten vacancies in advanced economies asks for at least one new skill, roughly twice the incidence in emerging economies. Vacancies asking for AI skills offer higher wages, while the authors also find lower employment in exposed, low-complementarity roles where those skills diffuse. The pattern raises concern for young entrants and middle-skilled roles. It is a regional and occupational finding, not a prediction that every new AI skill creates a job or that every exposed worker is displaced.[2][3][4]

Plans are changing faster than measured outcomes

McKinsey's August 2026 online survey asked 1,719 respondents in 97 nations about AI use and business effects. Thirty-nine percent expected a reduction in total head count in the next year, while 14% at AI-using organisations said AI had contributed to an overall reduction in the past year. Those questions refer to different time periods and rely on respondent attribution. They should never be presented as a measured 39% probability of losing a job. Firms may revise hiring plans for many reasons, and a consultancy selling AI advice has a commercial interest in the topic. The ILO task index and the McKinsey survey measure different things and cannot validate one another's percentages.

A practical assessment starts with your actual role. List the recurring tasks, the information and tools they require, which mistakes would matter, and which parts need a person to approve, explain or repair the result. Ask whether your employer is changing a process, merely adding a tool or reducing entry-level work. Check whether time saved is measured after review and correction, whether quality holds, and whether staff have a route to challenge a bad decision. This is a way to ask better questions, not a personalised employment forecast or a promise that training alone removes risk.[4][1]

Three moves for workers and managers

First, learn the tool used in your own workflow and practise verifying its output against a reliable source or a known baseline. Second, build the skills around the tool: domain judgement, data interpretation, communication with customers or colleagues, and knowing when to stop an automated action. Third, ask for training and measurement before accepting a claim that AI has made a role redundant or safe. Managers should publish what tasks change, provide time to learn, test failure cases and track job quality as well as speed. These steps follow the evidence more closely than a generic instruction for everyone to become a prompt engineer.

The next evidence needed is longitudinal: observed hiring, wages, task quality and worker transitions by age, role, gender and country, with comparison groups and full operating costs. Women have a higher share of jobs in the ILO's top exposure category than men globally, and exposure differs sharply between low- and high-income countries. A single worldwide prediction would hide those differences. We will update this guide when stronger observed outcomes change the assessment, keeping the original publication dates visible rather than silently presenting old research as new news.[1][2][3][4]

A task checklist: what could change in your own work?

Choose one recurring task rather than scoring your whole job. Write down the input, the required output, the person accountable for it and the cost of a mistake. Separate drafting or summarising from approving a decision, caring for someone or handling an unfamiliar situation. This is a practical exercise informed by task-based research, not a validated personal risk assessment.

Try an authorised tool on a non-confidential example. Compare the result with your normal method, including the time spent checking and correcting it. Record an error the tool made and the skill needed to catch it. Then ask your manager which tasks will change, what training is available and whether saved time will improve service, change workloads or affect hiring. A faster first draft alone cannot answer those employment questions.[1][2]

Common questions about AI and jobs

Which jobs are most exposed? The ILO places clerical occupations among the most exposed to generative AI and identifies growing exposure in some highly digitised professional roles. Exposure describes possible task change; it does not establish that a whole occupation will disappear.

Do I need to become an AI developer? The OECD synthesis distinguishes specialised model-development skills from the broader digital, analytical and human skills needed across many roles. Start with the tools and judgement used in your own work; the research does not support one universal training prescription.

Can AI create jobs as well as change them? The IMF staff research examines demand for new skills alongside uneven employment effects. New opportunities and displacement can coexist. The studies do not provide a reliable forecast for a particular person, employer or local labour market.[1][2][3]