Audit activities and measure the complete workflow before drawing conclusions from a job title or an AI demonstration.
A job title is a poor unit for understanding what artificial intelligence might change. Two people called “project manager” may spend their days doing very different things: one prepares routine updates, while another negotiates competing priorities and resolves disputes. A system that drafts a polished update does not automatically take responsibility for the project. For a useful assessment, break a working week into activities, identify what success requires, and ask where AI can reliably help. This approach will not tell you whether a particular employer will reduce headcount. It can tell you which changes are worth preparing for.
Exposure is not a prediction of replacement
The ILO–NASK occupational index, published in May 2025, estimates that one in four workers globally is in an occupation with some generative AI exposure. Its highest exposure category covers 3.3% of global employment. The study assesses potential effects on occupational tasks; it does not count jobs already lost or forecast an individual worker’s dismissal. The researchers identify transformation as more likely than complete replacement because occupations contain activities that still need human input.
Read any dramatic employment headline through that distinction. Technical capability, dependable deployment, commercial incentives, and an employer’s decision are separate questions. A model demonstration answers only part of the first one.
Map a working week before choosing a course
AI Affect’s suggested task audit starts with five ordinary working days. Record activities in plain language, rather than copying a job description. Include time spent correcting errors, chasing missing information, explaining decisions, and helping colleagues. Those activities are easy to overlook when evaluating a polished AI demo.
| Activity | Useful question |
|---|---|
| Draft a recurring report | Can its inputs and factual claims be checked? |
| Handle an unusual complaint | Who judges context and authorizes a remedy? |
| Update a spreadsheet | Can the result be reconciled against the original? |
| Agree a project trade-off | Who owns the consequences of the decision? |
Then classify each activity as suitable for a small AI trial, suitable only with close review, or unsuitable for your current setup. These are working categories, not permanent verdicts about what technology can achieve.
Measure the whole task, including review
Evidence from real workplaces is mixed across settings. The published 2025 study of 5,172 customer-support agents found a 15% average increase in issues resolved per hour with an AI assistant. Gains varied across workers. This was a specific tool and work environment, not a standard improvement to expect in every occupation.
For a personal trial, choose a repeatable, permitted task and define an acceptable result first. Compare similar tasks with and without assistance. Count preparation, checking, corrections, and the subscription or usage cost. Keep a short error log. If drafting becomes faster but verification takes longer, the relevant outcome is the combined effort.
Build evidence of your contribution
A useful skills portfolio contains completed work and a clear account of the judgment behind it. Show how you identified a problem, selected evidence, evaluated alternatives, and checked a result. Tool familiarity can be part of that story, but a list of product names says little about whether the output was useful.
Ask your manager which responsibilities are likely to change, what training is available, and how performance will be assessed. If you work independently, discuss the outcome your customer values before automating the delivery. The practical aim is to improve your ability to deliver and explain reliable work while staying alert to changes in demand.