Dossier

AI skills and work — train for changed tasks

Connect learning, role design and practical evaluation instead of counting course attendance.

Begin with the work

A generic AI course can improve vocabulary without improving a workflow. Identify the tasks people will perform differently: finding information, drafting, checking, deciding or supervising automation. Design learning around those changes.

Separate three capabilities

Users need to recognise useful and unsafe outputs. Builders need data, integration and evaluation skills. Managers need to select problems, allocate ownership and judge evidence. These groups overlap, but they should not receive an identical training programme.

Teach verification explicitly

Use examples with omissions, outdated sources and plausible errors. Ask participants to find and explain the failure, not simply produce a better prompt. Include information-handling rules and the point at which a person must escalate rather than continue.

Measure changes after training

Look for sustained task completion, correction effort and appropriate use several weeks after a course. Ask whether employees can explain the limits of the tool. Attendance, licences and satisfaction scores are useful programme signals but do not establish operational benefit.

Support the transition

Give teams a route to ask questions, report failures and suggest workflow improvements. Avoid making staff responsible for quality while denying them time to review outputs. Recognise that some tasks should remain human-led even when a model can generate a response.

Discussion for EventChanger

What should an AI-capable role be able to demonstrate in practice? A useful working-group output is a set of observable tasks and evaluation examples, with versions for users, managers and builders.