The short answer
No report can tell you exactly what AI will do to your organisation’s jobs. A useful response is to examine tasks, test a narrow workflow and develop the skills needed to check and improve the result. Use a 90-day planning window to gather evidence rather than make a headcount promise.
What the evidence does — and does not — say
The ILO studies task exposure to generative AI; the World Economic Forum records employers’ expectations about jobs and skills. These are different kinds of evidence. Neither establishes a predetermined staffing outcome for your business. Use them to frame questions, then test your own work.
ILO — Generative AI and Jobs (20 May 2025) · World Economic Forum — Future of Jobs Report 2025
A suggested 90-day preparation plan
This is SundAI’s planning template, not a delivery promise. Adjust the pace to your organisation, consultation needs and risk. Keep the three phases sequential so learning from one phase informs the next.
| Planning window | Work and decision |
|---|---|
| Days 1–30: understand | List recurring tasks with the people doing them. Choose one task with a clear output, manageable data and a named owner. Record the current time, error rate and review effort. |
| Days 31–60: test | Use approved tools and representative, permitted data. Compare human-only work with AI-assisted work against the same quality criteria. Include correction time, tool cost and failures. |
| Days 61–90: decide | Review the evidence with the team. Expand, revise or stop the pilot. Document responsibilities, training needs, change triggers and the next review date. |
An example: preparing a weekly operations brief
An illustrative team starts with a recurring brief assembled from approved internal sources. It defines “done” before introducing AI: correct figures, current sources and clear exceptions. The trial compares the complete workflow, including fact-checking and corrections, with its earlier process. If drafting becomes quicker but review takes longer, the team records the trade-off. No saving is claimed until the net result has been measured. The next experiment might improve the source data rather than give the model more freedom.
Prepare for three possibilities
Consider three scenarios: AI remains a drafting assistant; it supports several connected tasks; or it becomes capable enough to justify tightly bounded actions. These are planning scenarios, not forecasts or assigned probabilities. For each, ask which skills, data and controls would still be useful. Source verification, clear ownership and the ability to stop a failing workflow are sensible candidates to test across all three.
Make learning part of the operating model
Ask colleagues to explain when they would reject an AI answer, what information must not be uploaded and who can resolve an uncertain case. Use real work examples with safe data. Record misunderstandings and adjust the training. A readiness review can organise these decisions; a workshop can let the team practise them. General risk-management and secure-development guidance provide the wider context for keeping this work iterative.
NIST — AI Risk Management Framework · NCSC — Guidelines for secure AI system development
Turn uncertainty into a practical readiness discussion
Share the workflow or team you want to prepare. Explore a Readiness Review or an AI literacy workshop, with scope and availability agreed first.
Request a proposal →Keep reading
- Before your AI agent gets access: seven decisions to make
- A document can redirect your AI. What should stop it?
Sources and method
ILO — Generative AI and Jobs (20 May 2025) · World Economic Forum — Future of Jobs Report 2025 · NIST — AI Risk Management Framework · NCSC — Guidelines for secure AI system development
AI-assisted editorial guide by SundAI. Sources checked on 18 September 2026. Evidence confidence: high for the cited descriptions of risks and research; this worksheet and its examples are proposed practice, not measured client outcomes or a security guarantee. Future scenarios are not predictions.