Could AI take my job? How can I prepare?

The question starts with your working day

You open your inbox, prepare a report or try to organise meeting notes. Someone says AI can do it in seconds. The concern is understandable: if a tool can do part of what you are paid to do, what will happen to your job?

AI can automate tasks and affect employment opportunities. Its impact depends on the activity, how the employer introduces it and labour market conditions. This edition mainly covers generative AI: tools that produce text, images or code from instructions. No study can tell you with certainty what will happen to your individual job. [1] [3]

What exposure to AI means

In May 2025, the International Labour Organization and NASK estimated that one in four workers worldwide was in an occupation with some exposure to generative AI. The index examines tasks and their potential transformation; it does not count dismissals or calculate your personal probability of losing a job. The ILO considers transformation more likely than complete replacement for many roles. [1]

Start with an inventory: how much of your day involves writing, organising information, calculating, coordinating or resolving situations that need context? A job brings together different tasks. A tool that drafts the first email does not, on its own, handle the client relationship, negotiation or responsibility for what is agreed.

There are signals worth taking seriously

A Stanford working paper, revised on 12 August 2026, analyses US payroll records from ADP through June of that year. It finds weaker employment trends among workers aged 22 to 25 in AI-exposed occupations, mainly through reduced hiring. It does not find widespread displacement across the economy. [2]

The authors describe their results as descriptive indicators, not causal estimates. The magnitude also varies with the controls and data source. These are relevant signals for people seeking their first job; they do not prove that AI caused every reduction in hiring or automatically describe developments in other countries.

Where it could help during the day

As practice, you could request an email draft using fictional information, an outline for a report or questions to prepare for a meeting. If you are self-employed, try a fictional service description before using actual client information. These are suggested exercises to get started.

Microsoft 365 Copilot is one example of an assistant integrated into workplace applications. An evaluation published by HMRC, the UK tax authority, in July 2026 describes a trial held from September to December 2024. Participants reported average time savings of around an hour a week; the report cautions that this is an estimate, not a precise measurement of impact. It does not guarantee that saving for another person or tool. [4]

To assess whether it helps, count the whole process: preparing the request, reading the answer, verifying it and making corrections. Compare quality and errors with your usual method too. A quicker draft may offer little benefit if fixing it takes too much time.

What is worth learning

In its July 2026 paper, the OECD notes that AI affects sectors, locations and skill levels differently, and that adapting requires training and supportive policies. [3]

As practical guidance, work on skills connected to your profession: making a clear request, judging whether an answer is correct and explaining why you make a decision. In customer service, practise resolving a complaint; in administration, checking a calculation; in communications, verifying a claim. Knowing the work helps you notice what an apparently polished answer leaves out.

Learning to use AI can help you adapt, but cannot guarantee your job. The claim that “someone using AI will replace you” oversimplifies business and economic decisions beyond your control. Nor is it sensible to promise that a human skill will remain impossible to automate forever.

What to check before sharing information

The NIST generative AI risk profile, published in 2024, covers confidently presented false answers, privacy risks and excessive reliance on the tool. [5]

Check which service your employer permits and what information you may use with it. Practise with fictional data; do not paste contracts, payroll records or client data into a personal account without authorisation. Removing a name does not always prevent a document from identifying someone. Check figures and sources outside the chat before handing over the work. Asking AI to review its own answer cannot replace that check.

Practice with a work-related purpose

The aim is to identify which part of a task you can delegate, which responsibility you retain and which skill you need to develop. Choose something you do regularly. This connects learning AI to your work and lets you assess its usefulness.

Example: after a meeting, you need to send the team a list of agreements. AI could help organise the notes; you need to check that the list reflects what was agreed. These fictional notes illustrate the task: “Ana will send the proposal on Friday. Luis will review the prices; no date was set. Someone needs to call the client; no one was assigned.”

You could ask: “Organise these notes into a table with task, owner and deadline. Use only the written information. Where a detail is missing, write ‘to be confirmed’. Do not turn a suggestion into an agreement.”

The list should retain these details: send the proposal — Ana — Friday; review the prices — Luis — deadline to be confirmed; call the client — owner and deadline to be confirmed. Use this list as a reference when checking the exercise.

What are you practising? Giving precise instructions, spotting invented information and recognising what you still need to ask the team. If the system assigns the call to Luis or invents a deadline for reviewing prices, correct it. Asking it not to invent details cannot guarantee compliance. [5]

The potential benefit is spending less effort organising information. Check this by including review time and assessing whether the list is complete and accurate. The exercise cannot show that a job will disappear: it helps you decide whether AI could support this task and what you need to learn to supervise it.

A short plan for this week

An editorial suggestion: list the tasks you repeat most often and choose a low-risk one. Decide what would count as a useful result. Try it with fictional information, check the output and record the total time. Repeat with a different case to see whether it remains useful.

Keep a note of what worked, the errors and what you still need to learn. If you are job hunting, you can document your own exercise, clearly labelling it as practice without inventing professional experience or performance figures. Explaining how you checked and corrected an output shows more of your judgement than simply writing “I can use AI”.

Employers have work to do too

Ask which tools are approved, what training will be available, how responsibilities will change and how quality will be assessed. The ILO highlights social dialogue during this transition; the OECD also stresses support for skills development. [1] [3]

Preparing can start with one specific task tomorrow. The discussion about employment also needs decisions about training, recruitment and working conditions. It should not depend entirely on each person teaching themselves.