|

AI and Jobs: Practical Skills for Working Alongside Generative AI

8
Support

Learning to use generative AI can help with some workplace tasks, but it cannot guarantee that a job is safe. A useful starting point is to examine the work you actually do: which tasks can be assisted, which need checking and which should remain under human control?

This guide separates research about occupational exposure from predictions about redundancies, then offers a practical exercise for testing a tool without putting confidential information or important decisions at risk.

What does research say about AI and jobs?

The International Labour Organization’s 2025 occupational exposure index estimates that one in four workers globally are in an occupation with some exposure to generative AI. About 3.3% of global employment falls into its highest exposure category. Clerical occupations remain particularly exposed.

Those figures describe the potential for technology to affect tasks. They are not a forecast that one in four workers will lose their jobs. The study concludes that transformation is the most likely overall effect because most occupations still contain tasks requiring human input. Actual outcomes depend on adoption, costs, organisational choices, regulation and demand for the work.

Exposure also differs between countries and occupations. A global average cannot tell an individual whether their employer will recruit, retrain or reduce staff. Nor does learning a particular tool guarantee a promotion or a higher salary.

Start with a task map, not a list of fashionable tools

Write down five tasks you complete regularly. For each, record its purpose, the information involved, what a satisfactory result looks like and who is responsible if it goes wrong. Then sort the tasks into three groups:

  • Low-risk assistance: generating questions for a meeting, organising public information or creating alternative outlines for a document.
  • Assistance with verification: drafting a factual summary, suggesting spreadsheet calculations or preparing code that a qualified person must check before use.
  • Restricted or unsuitable use: uploading confidential records without permission, making consequential decisions without appropriate oversight or using a tool your employer has prohibited.

These are suggested categories for an exercise, not a replacement for your employer’s policies. A seemingly simple task can become sensitive when it contains personal information, commercial secrets or information about vulnerable people.

A worked example: preparing a weekly sales summary

Imagine an administrator who prepares a weekly summary from an approved spreadsheet. The job includes checking totals, identifying unusual changes, drafting an explanation and sending the final document to a manager. This is an illustrative example, not a measured workplace trial.

The administrator might use a permitted AI tool to suggest headings or explain an unfamiliar spreadsheet function. They should not assume that a fluent explanation proves a calculation is correct. Totals must still be reconciled with the spreadsheet, and any claim about why sales changed must have supporting evidence.

A safer first experiment uses invented sales figures rather than real customer records. Ask the tool to organise the figures, then check every number independently. If the tool invents a reason for a decline, remove that explanation rather than polishing it into something that sounds credible.

The useful skill is therefore not simply writing a prompt. It is specifying the task, recognising uncertainty, checking the result and deciding whether the output is suitable for its audience.

Four skills worth practising

  1. Source checking: open the original material behind factual claims. Check that quotations, figures and dates really appear there. A plausible citation is not enough.
  2. Problem definition: describe the audience, constraints and required output before requesting assistance. Explain what the tool must not infer.
  3. Domain judgement: strengthen the subject knowledge that lets you notice mistakes. A general-purpose model does not replace the qualifications required for regulated work.
  4. Communication: explain what was assisted, what was independently checked and what remains uncertain. The person receiving the work needs to understand its limitations.

NIST’s AI Risk Management Framework resources include a generative AI profile addressing risks such as confidently presented false information. That risk is one reason a polished answer should be treated as a draft to evaluate, not as evidence.

A one-week experiment you can evaluate

Choose one permitted, low-risk task and define success before using a tool. For example: a meeting outline must cover five agreed topics, contain no invented commitments and be ready within a specified preparation period.

  • First attempt: complete the task without AI and record the time and any corrections.
  • Assisted attempt: use public or fictional inputs and record time spent prompting, checking and rewriting.
  • Comparison: compare accuracy, completeness and total effort. Include checking time; faster drafting alone is not a productivity gain.
  • Decision: keep, revise or abandon the workflow. A tool that adds mistakes or requires extensive repair may not be useful for that task.

This is a suggested learning exercise, not evidence of a guaranteed efficiency improvement. Use several comparable attempts before drawing conclusions, and do not generalise the results to high-stakes work.

Before paying for training or changing career

Check actual job descriptions in your intended field and look for repeated skill requirements. Distinguish a recognised qualification from a certificate that merely confirms course attendance. Avoid courses promising guaranteed employment or claiming that one prompting technique makes professional knowledge unnecessary.

Start with a manageable project that demonstrates a skill you can explain and defend. A portfolio example should identify the task, your contribution, the checks you performed and the limitations of the result. Use fictional or authorised material rather than exposing a previous employer’s records.

Explore our Jobs & Opportunities hub for career resources and the AI & Technology section for related coverage. No article can predict an individual’s employment outcome; the practical aim is to make better-informed learning and workplace decisions.

Sources and scope

Updated 8 November 2026: revised to distinguish exposure from job-loss predictions and add a sourced, practical task-evaluation exercise. The worked example is hypothetical.

8
Support

You May Also Like

Leave a Reply

Your email address will not be published. Required fields are marked *